LWN.net Logo

The proposed Fedora key-migration plan

For those who wonder how the Fedora project plans to migrate its users to a new set of package signing keys, a proposed plan has been posted. It involves an update to the fedora-release package (signed with the old key) which swaps in a new key and repository location, and a slow movement of older packages to the new repository. It should work, as long as one is sure that the old key can be trusted for a little longer.
(Log in to post comments)

The proposed Fedora key-migration plan

Posted Aug 28, 2008 23:35 UTC (Thu) by pragmatine (guest, #39557) [Link]

This is pretty bogus. If they don't trust the old key enough to want to replace it, you can't then install the new key using the old ones as its trust base. This only makes the new key as trusted (or less than) the old key, which since we know they don't trust anymore is a false sense of security.

Closing the window of attack

Posted Aug 29, 2008 1:37 UTC (Fri) by tialaramex (subscriber, #21167) [Link]

I don't agree. The factor you're missing is /time/ which is often the most critical factor in design of cryptographic systems.

Often an attacker can eventually obtain a secret (e.g. crack the Fedora signing key's passphrase), but they obtain it too late to be useful and the cryptosystem has functioned as desired. Nobody in the business of keeping serious secrets expects them to last forever, but they only have to last long enough. On June 5th 1944 the detailed plans for the Normandy landings were among the most important secrets in the world. Over the following days and weeks they became merely a historical curiosity.

Updating everyone to new keys closes the window of attack, giving the attackers much less time to capitalise effectively if they do in fact have, for example, an encrypted signing key for which they are trying to guess the passphrase. It's the same reason it makes sense to occasionally change your password. If someone has been slowly trying to figure out the password by watching you type it in, or they have been trying to break the PHK MD5 hash they managed to steal from an old backup, all their slow work is undone by simply changing the password, and they must begin again.

Closing the window of attack

Posted Aug 29, 2008 9:14 UTC (Fri) by pragmatine (guest, #39557) [Link]

that assumes the passphrase of the old key hasn't already been cracked though. if it has all bets are off.

Closing the window of attack

Posted Aug 29, 2008 23:10 UTC (Fri) by Sutoka (guest, #43890) [Link]

For an administrator, if the password of the key has already been cracked then you should try to verify the new key from as many different sources as possible first.

It's also possible the key *wasn't* stolen, and that this is just a precautionary measure to make sure the packages signed by the attacker don't get installed on the user's system as if they were valid.

I wonder if there's a version of the fedora-release package signed with the new key, allowing the admin to first install the key manually then use the newer version of the RPM (thus not having to trust the old key at all).

In the future, Fedora might wanna have multiple keys. One *normal* key they do all their signing with (like now), then another emergency key in case, well, this happens. The emergency key's only purpose would be to make it easier to replace the standard key (i.e. this fedora-release package would be signed with the emergency key). Obviously Fedora would want to make sure the key was encrypted and not stored somewhere that it'd be inconvenient to get to (preferably not accessible by a computer).

Closing the window of attack

Posted Aug 30, 2008 0:02 UTC (Sat) by bojan (subscriber, #14302) [Link]

> In the future, Fedora might wanna have multiple keys.

If Fedora packages were signed by multiple independent signatories, yum could be designed to accept (for instance) only packages signed like this:

- Fedora key
- at least 2 other keys from independent signatories

or

- 5 keys from independent signatories

In that case, a compromise of a single key would be easily avoided, as long as you had enough signatories in the pool to make up the numbers (and with vast number of people involved in Fedora, this is definitely possible).

The proposed Fedora key-migration plan

Posted Aug 29, 2008 1:27 UTC (Fri) by jamesh (guest, #1159) [Link]

If they kill the old key in people's RPM database (step 7), won't that prevent people from verifying the contents of their local RPM database? Or is there some other way to import the new signatures for the existing installed packages?

The proposed Fedora key-migration plan

Posted Aug 29, 2008 16:46 UTC (Fri) by pizza (subscriber, #46) [Link]

If I recall correctly, the signatures are only used to verify the actual packages (ie RPMs) prior to installation. Once installed, the package signatures serve no further use as the packages are discarded.

Verification of the files installed by various packages takes place against RPM's local database of all installed files -- but that can't tell you if the original package was trustworthy or not.

Copyright © 2008, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds