GUACAMOLE-600: Correct handling of non-blocking socket for timeout.

[necouchman]

Okay, I think I've got the issue figured out after some Google Research on how to handle non-blocking sockets in Linux. Let me know if this looks sane or needs any work. I re-tested and am able to establish SSH connections...

[necouchman]

Tested with a couple of different timeout values (5s and 30s) against an IP address that isn't up/doesn't respond, and it appears to work as expected.

[mike-jumper]

Should we restore the default (blocking) behavior prior to returning the socket, so that callers are not surprised to find that the socket is non-blocking?

[neandrake]

I was looking into something similar, flipping a socket to non-blocking and back, for when read-only connections are created and sync display so it doesn’t cause the main connection to lag then. I found it challenging to think through the behavior of the socket. Would the socket need polled until all expected data comes in before switching back to blocking?

[necouchman]

Should we restore the default (blocking) behavior prior to returning the socket, so that callers are not surprised to find that the socket is non-blocking?

Yeah, probably a good idea. I've added a block in to do that, too.

[mike-jumper]

Would the socket need polled until all expected data comes in before switching back to blocking?

I think the blocking vs. non-blocking behavior comes into play only when actually servicing a system call, like a read(). The socket should buffer any received data regardless of whether O_NONBLOCK is set.