Question

I have a python application that creates a subprocess, opens a socket to communicate with it, then creates a multiprocessing.connection object over the socket. The connection object uses a shared key (randomly generated) and hmac to ensure that no other process will be allowed to communicate over the connection.

On Linux, this works perfectly. On windows, I get the error:

multiprocessing.AuthenticationError: digest received was wrong

The key is a string of randomly generated bits which is pickled before being sent to the subprocess by its stdin:

authkey = ''.join([chr(random.getrandbits(7)) for i in range(20)])

And I have checked carefully that the key matches on both ends of the connection this-a-way:

 print "key:", ' '.join([str(ord(x)) for x in authkey])

The server is started with:

 l = multiprocessing.connection.Listener(
         ('localhost', int(port)), authkey=authkey)

..and the client is started with:

 c = multiprocessing.connection.Client(
         ('localhost', int(port)), authkey=authkey)

Both processes are running on the same machine, with the same version of python.

Stranger yet, I have found that if I fix the key (say, authkey=’test’), then I still get the AuthenticationError the first time I run the program, but not on subsequent runs.

Answer 1

The solution appears to be to use os.urandom to generate the key rather than the method shown above. I do not know why this has any effect–in either case we’re passing random byte strings. The solution is also more correct from a security standpoint because getrandbits is not intended for cryptographic use.