Libssh blocking instead of timing out

[Chris Backas <chb@xxxxxxxxxxxxxxxxxx>]

Hello all,

I wrote a few months ago about a problem I was having with libssh 0.2  
hanging under certain conditions, and at the time we decided I should  
upgrade our code onto 0.3 to see what happened.

So I did that back then, pretty uneventfully, and it's been fine until  
last night. Last night we had two automated processes hang trying to  
connect to an SSH host. In both cases, the situation seems to have  
played out like this:
1) We dialed up and connected to a remote system, and established a  
PPP session
2) We tried to connect to the remote host over the PPP link
3) The PPP link failed almost right away, with PPPD ultimately  
terminating with "Peer not responding to echo requests"
4) libssh never returned (we found it over 12 hours later)

Dial up links with some of our customers can be flakey, so the fact  
that the PPP link was never entirely stable isn't shocking.  However,  
we can't have libssh hanging in a failure case.  I took this stack  
trace from the process; it was identical on both of the machines that  
had a hang:

ssh_connect
	ssh_get_kex
		packet_wait
			packet_read2
				ssh_socket_wait_for_data
					ssh_socket_completeread
						ssh_socket_unbuffered_read
							recvfrom
								recvfrom

Reading the man page for recvfrom (ssh_socked_unbuffered_read actually  
calls recv, but the man page says that call is redundant and may go  
away, so I'm guessing it's just implemented as recvfrom) it sounds  
like the ability to have a timeout is dependent on the socket being  
created/opened with a timeout flag set.

So... is there something I should be doing with the library to enable  
timeouts, or is it something that libssh needs to be doing?

Thanks for any help.

Chris Backas
Software Engineer
Bristol Capital Inc.