innd and rejects of dates

[Julien ÉLIE]

Hi Thomas,

>> A Date: in 2000 or a Date: in 2020 would then be accepted by innd
>> for an Injection-Date: in 2010.  And an invalid date too.
>> Is it wise to let such articles pass through?
>
> What about replay of signed control messages, i.e. old checkgroups?
>
> Currently you can't reinject an old checkgruops successfully because
> its Message-ID is in history or it'll be rejected because its Date is
> too old; both are signed headers. If INN will accept old articles with
> a current Injection-Date - which was and is not signed, and even not
> signable as it will be appended by the injecting agent (?) -, it will
> be possible to replay a checkgroups from five year ago which will be
> executed due to its valid signature. What do I miss here?

That's a very interesting and thoughtful remark.

It is indeed very problematic for checkgroups that do not have their
Injection-Date: signed...  All of them are currently in that case.
And most of them do not use a serial number (which, anyway, is currently
not taken into account by INN).

But even with the serial number, the problem remains because newgroup
and rmgroup signed control articles may also exist...


Regarding the Injection-Date: header field, it is signable by the posting
agent because it can be added at posting time (contrary to the Injection-Info:
header field, necessarily added at injection time).




So basically, it implies that the check on the Date: header field
MUST still be done on newgroup, rmgroup and checkgroups articles.
(Not all control messages.)
Is it OK to do that?


Should something be done on moderated newsgroups?  As far as I see,
PGP Moose does not sign the Date: header field...  So reinjections
can occur, cannot them?  (Even now.)

-- 
Julien ÉLIE

« Hey, I had to let awk be better at *something*... » (Larry Wall)