On Thu, Jun 11, 2009 at 03:23:26PM -0500, Shawn Walker wrote:
> [F]or the same reasons you noted about the CA  certificates, I think
> the ability to override peer verification  behaviour is necessary.

I'm still not convinced that it's a good idea to allow users to
voluntarily use less security.  In the cases where a cert is
self-signed, or the client cannot verify the chain properly, I think
we're better off having the users expicitly decide to trust a particular
server.

> And I absolutely hate that behaviour of modern browsers, such as  
> FireFox.  I can't count how many websites Sun has up, or other  
> legitimate websites I've been to that I have to go through FireFox's  
> "your papers please!" page.  But do note that FireFox does give you an  
> escape route, it lets you continue anyway if you explicitly tell it to  
> do so.  For the same reasons, and more, I think we need to do the same  
> thing.

In retrospect, using a web browser was a stupid example since it doesn't
frame the problem in terms of our customers use cases.  To configure a
https repository, the client has to explicitly invoke set-publisher,
which runs a test to make sure that it can connect to the requested
publisher.  If that fails, we get an error message that specifies
what has gone wrong.  If it's a trust issue, I think it's better to
obtain the certificate that we want to trust, instead of flipping a
switch and saying, "eh forget about it."

Once we have the server's certificate, we can verify that it's the same
one that we used last time.  A failure would be analogous to the type of
warning that OpenSSH generates when your the host key of the host you're
trying to connect to doesn't match the stored key.  Presumably, clients
are using encryption and authentication to ensure that they're talking
to someone they trust.  If they don't care about this, then using https
really seems like overkill.  At least in our model.  (We're not sending
credit-card numbers over the wire or anything like that...)

-j
_______________________________________________
pkg-discuss mailing list
[email protected]
http://mail.opensolaris.org/mailman/listinfo/pkg-discuss

Reply via email to