🚀 go-pugleaf

Russ Allbery <rra@stanford.edu> wrote:
>"Adam H. Kerman" <ahk@chinet.com> writes:

>>Did I misunderstand what Russ stated earlier?

>>The control.ctl entry and the public key on your Web site continue to
>>refer to the long defunct newsgroup ba.news.config. Isn't correcting all
>>that a necessary prerequisite to getting the key to work?

>Given that nothing that verifies Usenet control messages does so on the
>basis of the contents of a web site, no, it's not.

I wasn't assuming that verification uses another medium, but that
providing News administrators with correct configuration information is
a prerequisite.

>Step one is getting a combination of PGP key and control.ctl entry that
>generates a signed control message which then verifies for someone other
>than rone.  Once we figure out which key and control.ctl entry accomplish
>that, then we can document it, on that web site and elsewhere.

>The problem is, there are two different keys, and those keys have multiple
>different user IDs depending on which copy of the key you have because a
>change of user ID to match the name of the newsgroup was incompletely
>propagated.  It looked like the 632E03A9 key with a user ID of
>ba.news.config was the path of least resistance, but I may have been
>wrong.  (The correspondence between a user ID and an actual newsgroup is
>unnecessary as long as all the configuration lines up.)  However, I can
>certainly change the ftp.isc.org configuration to use something different
>instead.

I didn't realize the newsgroup name was being used as the user, so
yeah, I suppose the user could be anything at all. But if you're
gonna use a newsgroup name as the user, using the name of a newsgroup
that hierarchy administrator hasn't recognized for 11 years just seems
to be asking for trouble.

>It would be nice if, once we get this sorted out, we can tell people to
>update only *one* thing, either the PGP keyring or the control.ctl file,
>and not ask them to have to update both of them.  We might have marginally
>more success at getting this untangled.

Thinking about this for a moment, I suspect that not having the keyring
in control.ctl is not any kind of stumbling block for a News
administrator to implement this verification method. As implementation
isn't a prerequisite for running a News server, and as there are real
advantages to having separate files, please don't include the keyring in
control.ctl. control.ctl is reasonably human readable. As most of it is
documentation, it's meant to be human readable. The keyring just doesn't
belong there.

Here's an example of the advantage of keeping control.ctl as human readable
as possible. I subscribe to a number of servers. When I realized
something was wrong with the newgroup message, I looked at control.ctl,
found the reference to the hierarchy's Web site, found a Web page in
checkgroups format, and requested that one of my servers process the
checkgroups manually to make changes (which also solved the 11 year old
ba.news.* reorganization locally).

It seems to me that following newgroup messages and checkgroups as they
come in and then processing them manually for hierarchies whose set of
recognized newsgroups don't change for more than a decade is a perfectly
reasonable way to administer a News server. It may not even be desirable
to process changes in regional hierarchies automatically. Changes are
infrequent and unexpected and something a News administrator might like
to eyeball first. Keeping control.ctl more human readable makes it easier
to handle changes manually.

>Maybe it would be better to go with the F0BD2595 key and change
>control.ctl to expect ba.config.  That is more consistent with the
>current newsgroup names, even though that doesn't matter from a technical
>perspective.

Which key is more practical to use is a question for someone like Julien
or a working News administrator, not I.