Re: signing RPMs without a passphrase?

On Oct 9, 2008, at 12:00 PM, Lev Lvovsky wrote:

> thank you *Jeff*!
>
> The first response in the link provided just seemed a little off- 
> base to me.  There's nothing intrinsically more secure about me  
> typing in some passphrase vs. an automated procedure just skipping  
> the step - AFAIK, GPG is used to provide file signature verification  
> (along with mdt5 and whatever other hash algo. is employed).  But  
> it's also used to verify the entity that the RPM came from - an  
> identity which the installer chooses to trust, passphrase  
> notwithstanding.  Am I missing something  there?
>

There are two issues that you are attaching to signing:

    1) Point of origin
    2) untampered guarantee

They are really separate issues.

FWIW, RPMTAG_COOKIE was rpm's attempt to pin down origin
sufficiently well. That string has the fully qualified host name as
well as a time stamp.

Yes both DNS and time can be wrong or maliciously tampered with.

Yes there is information leakage of build system internals through
RPMTAG_COOKIE too.

But RPMTAG_COOKIE could be used to split point-of-origin from
untampered if some other content were supplied there instead.

And if anyone knows a signing algorithm that does not require
a pubkey distribution framework in order for end-user to verify  
integrity, I'll
be happy to just automate the integrity signing within rpmbuild
everywhere and always.

> I'll check out keyutils - thank you very much for your help Jeff!
>

Note that keyutils also has a way to provide an asynchronous callback,
with negative "NOKEY" return caching to avoid pointless overhead,
as well.

keyutils == nice stuff

73 de Jeff