Fix send-file-descriptor, receive-file-descriptor

[samuel-hunter]

The msghdr struct initialized no iovecs when preparing to send or receive file descriptors. On my machine, when the client calls sendmsg() with no iobuf, it succeeds (return val is not -1), though the receiving side receives no message, therefore failing to send a file descriptor.

I presume this is because it "optimizes" an "empty" message away as no message, therefore still calling it a success.

This change initializes an iovec with a size of 1 byte, the bare minimum required to create a successfully sending message.

Tested on Linux 5.10.10 x86_64

Essentially, the change would be synonymous with this C snippet:

+char nothing_name;
+struct iovec nothing_iov;
+iov.iov_base = &nothing_name;
+iov.iov_len = 1;

struct msghdr msg;
msg.msg_name = NULL;
msg.msg_namelen = 0;
+msg.msg_iov = &nothing_iov;
+msg.msg_iovlen = 1;
msg.msg_flags = 0;
msg.msg_control = cmsgbuf;
msg.msg_controllen = sizeof(cmsgbuf);

struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
cmsg->cmsg_level = SOL_SOCKET;
cmsg->cmsg_type = SCM_RIGHTS;
cmsg->cmsg_len = CMSG_LEN(sizeof(shm_fd));
memcpy(CMSG_DATA(cmsg), &shm_fd, sizeof(shm_fd));

// call fn...

Testing the C code before the changes verified that initializing a 1-byte iovec was needed. Applying the changes to C fixed the issue and I've verified that translating it back to iolib completed the fix.

[sionescu]

Do you know since when that's happening ? I wrote the code on Linux years ago and it used to work.

[samuel-hunter]

I'm not sure when it started happening. I found this bug independently a few months ago, ported the code to C and replicated the same issue.

[samuel-hunter]

Reviving this PR @sionescu, what can I do to get this merged? iolib is the only library I know of to support sending file descriptors, so I would love to see this fixed upstream!

[sionescu]

@samuel-hunter I'll try to look into it sometime in the next week.

[samuel-hunter]

Following up @sionescu, are you able to look into this?