[Zope] Re: ZEO Client space was:

[Bill Anderson]

On 24 May 2001 21:25:14 -0700, Michel Pelletier wrote:
> On Thu, 24 May 2001, Tim Cook wrote:
> 
> > Michel Pelletier wrote:
> >
> > > XMLRPC could do this, but I'm afraid it would be seriously
> > > limited.  XMLRPC is very useful, but it does not map as well as it could
> > > onto the python object model.  I would prefer to see a GUI client running
> > > in the same process space as a ZEO client, this way the GUI client could
> > > do everything that Zope could do and re-use alot of the existing
> > > interfaces (like security, acquisition, script objects, etc).
> > >
> >
> > Michel,
> > I took this off list because it was off topic for the thread. But
> > feel free to Cc it back if you feel it's important.  I think it
> > was covered a few weeks ago on the -dev list but couldn't locate
> > it.
> 
> It's not off topic so I'm cc:ing the list.
> 
> > Anyway, the question is: I thought I had read that using ZEO
> > basically by-passed the Zope security model? Maybe I was thinking
> > of direct ZODB access?
> 
> This is a good question.

A ZEO Client has full access to the underlying objetcs, that's kinda the
point of it. ZEO being independant of Zope, it has no clue about the
Zope security mechanism. I think it would be possible to make it aware,
but others disagree.
> 
> Using ZEO does bypass the Zope security model, but then again, Zope
> bypasses the Zope security model.  Zope is, essentially, a client that
> implements an application on top of a persistent database.  Why can't your
> client do the same thing?  Zope enforces security upon its users using
> tested, re-usable interfaces.

Here is the rub, and why ssh/ssl is irrelevant and insecure too.

You are trustin the client implicitly, as 'root' on the ZODB. ALL the
security is implemented on the CLIENT side with Zope, and and ZEo client
fo rtha matter. For example, try this out:

Set up a Zope ZSS. Set up a Zope ZSS client on the machine. Set up a
second one. In the first one, make your emergency user 'bob', password
'bar'. On the second, make the emergency user 'Fred', password 'spam'.

Fred has root-level access to the ZODB. I don't care if you use SSH
tunnelling or SSL, you are still giving the user god-like abilities to
do whatever they please with your database. if they wish, they can
replace their client's auth mechanism with one that returns true all the
time, without bothering to actually look at the authentication. They can
give themselves whatever roles they desire.

This makes it, an all or nothing approach. Either you give the user
'root' to your Zope database, or you don;t allow them ZEO access. IMO,
that is bad, and dangerous.

...

> By going with xml-rpc, you'll be shoving *very* rich interface semantics
> of a full blown windowing sytem down the tiny semantic straw of xml-rpc,
> to try and manipulate interfaces that were originally indented to be used
> by medium to poor semantic HTML interfaces.  If you use a ZEO client,
> there are no bottle necks.  You could render the object system in 3D
> OpenGL if you wanted to, and still enforce the user.

I agree, xml-rpc is not exactly the way to go for this type of usage.
SOAP perhaps, but not xmlrpc.

> 
> Jerome's ZShell script shows how this can be done safely.  His shell is
> essentialy just another interface to the Zope core code.  It would be very
> dangerous if it did not use Zope's security API to restrict its actions.
> 
> In zshell's case, the interface is really simple, a command line. It is
> probably not as safe to to grant access to this command line over
> unprotected networks. and it would be unsafe to expose a GUIfied ZEO
> client to unprotected networks, but this is easily mitigatable with things
> like ssh or ssl.

If ZShell connects via ZEO, it is insecure. You are correct that it
would be bad to allow over untrusted networks, as is any ZEO that needs
security of data/data modification. As I have mentioned, ssl/ssh are
mostly irrelevant to database security, as even though the data may be
non-usbale to network outsiders, those with access can literally do
anything. Whmever you give access to via ZEo has unlimited authority on
the database.

For example, the oops in the ZShell 1.0 release illustrates this nicely.
To wit:
"""
This means that unfortunately the setperms command doesn't check if the
user has the "Change permissions" permission before changing
permissions,
and changes them.
"""

So, you give a user you presumably trust, ssh/ssl access to use ZShell.
They comment this code out, or the code that even botehrs to check for
permssion to do anything, and now they now '0WNZ Y0ur z0p3 sist3m'.

I think ZEO needs a forced authentication path hook. It wouldn't need to
implement the actual security, but would provide a method where upon the
app designer could say 'when accessing objects, first run through this
method to determine access level'. But that's a separate thread, for a
different list. :)