On Sat, Sep 20, 2008 at 6:08 PM, Henning Schmiedehausen <[EMAIL PROTECTED]> wrote: > On Sat, 2008-09-20 at 10:08 +0100, Robert Burrell Donkin wrote: >> On Fri, Sep 19, 2008 at 6:11 PM, Justin Erenkrantz >> <[EMAIL PROTECTED]> wrote: >> > On Fri, Sep 19, 2008 at 6:12 AM, Hiram Chirino <[EMAIL PROTECTED]> wrote: >> >> How about we include the signatures in the source distros? That way >> >> if you trust your source, then you can trust the dependencies it >> >> downloads. >> > >> > Eww. That'd be a giant gaping security hole. >> >> not necessarily, depends how it's done >> >> signing works through trusting the people who own the keys. given >> sufficient signaturees (to prevent small conspiracies), where the >> signatures are downloaded from shouldn't matter. > > Hiram suggested to put the signatures into the source, which in turn is > also distributed from the repo. If you compromise the repo and change > the artifact, it is trivial to update the source artifact to contain a > matching signature.
AIUI it is not sufficient to gain access to the source. forging a signature without the private key is not feasible. therefore, a feasible attack means using a private key trusted to sign incubator releases that has been unknowingly compromised or using a rogue incubator release manager. it should be possible to provide defense against these attacks by using signatures from a sufficient number of keys. > This is a security hole. And I don't really care for some of the > proposed "high nineties" security solutions. Either a solution is secure > or it is not. Everything else is just FUD. public key cryptography cannot guarantee absolute security. security measure only protect against particular attack vectors. no solution is toyally secure. IMHO the best approach is defense in depth with good protection against all known attack vectors > The problem with the central repo is that you need an easy accessible > web of trust if you want validation. The Apache web of trust is > distributed and an overlay to the GPG web of trust. But if you live in > Juneau, Alaska, it is hard for you to access it and get a trust > relationship to it. WOT is about personal identity and is not role based. the apache WOT allows other members to trust that the owner of the private key is in fact the 'Robert Burrell Donkin' listed in the members list. release validation is quite a different notion of trust. just because a release has been signed by someone called 'Robert Donkin' does not mean that it can be trusted. perhaps it is some other individual or perhaps i signed something with another key rather than my 'CODE SIGNING KEY'. what is required is a role-based notion of trust: that a particular key is trusted to sign incubator releases for a particular podling. this does not require a link through the WOT. having said all that, i believe that apache as an organisation *really* needs a better WOT so that *we* can verify our own releases. - robert --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
