Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 25 Jul 2013 16:10:58 -0400
From: Daniel Kahn Gillmor <>
CC: Kurt Seifried <>, 
 Yves-Alexis Perez <>
Subject: Re: CVE Request: evolution mail client GPG key selection

On 07/25/2013 03:19 PM, Kurt Seifried wrote:
> One problem moving it to the backend is I may trust your key to sign
> email, but not to sign code and vice versa. At least for RPM this is
> handled internally (you install the key into RPM) so that's one less
> worry.

gpg doesn't claim to know *anything* about what you "trust" keys to do
other than whether you're willing to rely on that key to certify other keys.

gpg can tell you "this message was signed by dkg" but it cannot (and
should not) say "dkg is allowed to upgrade your libc".

basically, gpg's job is to handle the authentication side of things
(which is all that an MUA cares about) but *not* to handle the
authorization side of things.

Authorization is rightly scoped by different uses of the infrastructure,
as you suggest.  Some work is being done on this too: Stef Walter's
current developments of p11-kit aim at allowing tools in a similar
domain (initially, web site X.509 certification and potentially code
signing) to share authorization information.

> Yup. Key management sucks even with good software, and we don't have
> good software for this. Witness CRL/OCSP and then all the vendors
> shipping browser updates to specifically blacklist compromised
> certificates.

indeed.  Same sort of problem, same shitty non-scalable solutions :(

> Yeah the use of KeyIDs is just stupid (although I get why they did it,
> I still think it was stupid), they're to short. We should be using
> fingerprints only. Sigh.

even if it uses full fingerprints, the fact that i can associate your
primary key as a (revoked) subkey of mine means i can force the
keyservers to include one of my keys when anyone goes to refresh yours.

> True, and trust me if my key ever gets compromised (and I know about
> it) I'll be phoning certain people and not relying on key revocation
> to let them know.

That's a good idea, but i hope you don't see the two tactics as mutually
exclusive.  Personal notification is guaranteed to miss some people,
especially for people whose key is used publicly (like yours is).

> The problem here is the all or nothing trust setting. I may trust your
> key enough to use it to encrypt email to you, but not to accept signed
> email from you for example. To steal a phrase from Moxie Marlinspike
> "no trust agility".

Wait, are you saying that you throw away e-mails from me because they're
signed?  clearly not :)  And if my key is good enough for you to encrypt
mail to (that is, you believe it's a key i control, why don't you think
it's a key i control on the other direction?

sure, you might decide that you don't want to give me keys to your house
(or let me approve software on your machine, or whatever) based on the
fact that i send you a signed e-mail -- but that presumably has more to
do with who i am than whether it's my key or not.

> from enigmail's perspective why is that key present at all, so I can
> see defaulting to trusting it, basically the policy sounds like "if
> the user bothers to grab a key lets use it" which is probably not the
> safest implicit policy, but does get things working.

If that's what enigmail is thinking, it seems like a dangerous mistake
to me.  In regular daily use, it is implausible that all keys in a
user's keyring are actually held by trustworthy parties.

It does "get things working", but it leads to messages being encrypted
to arbitrary recipients.

I could see more nuanced approaches that still wouldn't require any user
intervention but would be better than the status quo, like:

 * if more than one key exists with the given user ID, and one of the
keys has a higher validity than the others, choose that one.

I personally run enigmail with the "Always trust people's key" unchecked.



Download attachment "signature.asc" of type "application/pgp-signature" (1028 bytes)

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.