It would be very simple to secure the pydoc server so 
that it doesn't accept connections from external boxes 
as well as provide for a way of extending connections to 
trusted hosts by keeping a list of valid IP addresses. 
This would make pydoc suitable for running on boxes 
that aren't behind firewalls, which currently it is not; 
most home machines don't have a firewall and are 
regularly port scanned by script kiddies...

Since pydoc does not log connections, you can't tell 
who is connecting to your machine or what they are 
trying to reach. My solution is to simply make the 
default pydoc server only accept connections from the 
host it was started on.

The change is for the DocServer class. a validIPList 
keeps track of the IP addresses that can legally connect 
to the server. The verify_request method is overridden to 
enforce this rule.

            import socket
            self.validIPList = ['127.0.0.1']
            self.validIPList.append(socket.gethostbyname
(socket.gethostname()))


        def verify_request(self, request, client_address):
            if client_address[0] in self.validIPList:
                return 1
            else:
                return 0

This patch does not provide a UI change to allow the 
user to easily add additional IP addresses. If that is 
desired because of the assumption that people typically 
run the pydoc server not for personal use, but for a group 
of machines to reach, then the simplest change would 
be to have a checkbox for "Allow any host to connect" 
and then have a self.allowAny member variable to reflect 
that checkbox state, so the verify_request becomes

    def verify_request(self, request, client_address):
        if self.allowAny or client_address[0] in 
self.validIPList:
            return 1
        else:
            return 0

ka

I think this is actually a good idea; but I don't think the implementation is really sufficient as it stands. Particularly, it's going to require that someone hand edit a file in Lib to adjust the behavior from the "default" of only allowing connections from localhost. A user interface is not required, but an easy to reach configuration file is, I think.

Instead, I think it should read a pydoc.cfg ConfigParser file-- and just apply the defaults if said file doesn't exist. (Where to put it? I don't know. ~/pydoc.cfg?)

Also, having to list specific IP addresses is going to greatly limit utility for those people who do want it more open. Some people might want to allow everyone in their subnet to access it, instead of just 'everyone' or 'specific people' as this patch implies. I don't think there's an easy way to do CIDR math in the Python library, but a simple regex in said configuration file would be plenty I imagine. Or even a list of strings you check to see if the ip address startswith.

In the current form, I'd recommend rejection. I don't know if the submitter is interested in any major updates after a few years, but if they are.. :)

Wouldn't it be easier to just bind the server to localhost? That way, the server should only listen on the loopback interface and not any of the external network interfaces. At around line 1974 of pydoc.py (Python 2.4.3)...

            host = (sys.platform == 'mac') and '127.0.0.1' or 'localhost'
            self.address = ('', port)
            self.url = 'http://%s:%d/' % (host, port)

Replace the '' with host in self.address by default, perhaps. Then, add a host parameter to the serve function and let this be used to override the above. Expose the parameter as a command line argument. I'll come up with a patch for this at some point, I suppose.

This looks weird, a security issue with a low priority???

As the pydoc server "advertises" that it is running from localhost in both CLI and GUI, it is best to bind the socket to 'localhost' instead of '' (which would bind it to all the interfaces).

So, a simple fix for this issue, which will remove the security concern:
             host = 'localhost'
-            self.address = ('', port)
+            self.address = (host, port)

If is to be run from user-defined interface with a new --host <interface> option, that it can be dealt with as new feature request.

This issue can be considered fixed with commits r84173 and r84174.

It looks like this bug was reintroduced in a5a3ae9be1fb.

Devin, please open a new issue describing the current problem you see.  Comments to long-closed issues will likely be overlooked.

Sure, thanks.

New issue: http://bugs.python.org/issue22421