-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Robert Menschel writes:
> Hello Gerd,
> 
> Wednesday, February 9, 2005, 4:51:39 PM, you wrote:
> 
> >> Most people would find the SHA1 file sufficient, the more paranoid among us
> >> would do GPG.
> 
> GvE> sorry to intercept, but aren't the updated rule files that you
> GvE> plan to publish code that is instantly executed as fetched? ...
> 
> 1) IMO the rules files aren't "executed", but are regex filters (or
> equivalent) that check for spam. A compromised system might be tweaked
> to allow more spam through, but should not be zombied.

A malicious plugin (for example) will indeed be executed.  Malicious lines
in a *system-wide* config file could cause trouble by scribbling over
files; e.g. "bayes_path /etc/passwd" or similar...

> GvE> We've seen breakins at some well respected open source sites
> GvE> within the last months (e.g. debian). But the damage was minor
> GvE> because only few people downloaded the malicous code before it
> GvE> was found and removed. This will be different with an automatic
> GvE> update system...
> 
> My point is that if the rules files to be transmitted are
> automatically compiled from subsidiary rules files, and the GPG
> signature is applied by an automated system, without a human checking
> that each and every rule is indeed valid and correct, then there's
> nothing to stop a breakin from slipping their dirt into one of those
> subsidiary files, and the automated system will happily sign the
> corrupted file.
> 
> If a file goes from system A to B to C to D, the system at D can check
> A's signature to verify the file wasn't modified by B or C, but if A's
> signature is automatically applied on system A, there's NO assurance
> to the quality of the signed file.
> 
> Am I wrong about this?

Yep! ;)

There's levels of trust.  For example, a third-party mirror may be
more easily compromised than our locked-down rules-update server.
And let's say we have 30 mirrors (as the ASF do) -- or, worse,
Coral, which adds even more, and has to be considered a potentially
dangerous traffic medium -- that would mean 31 possible avenues
of attack.

Signing at the "bottleneck" -- on a locked-down updates server -- which
then dists out to machines on the public internet, and ensuring
downloaders always verify sigs, is a lot safer.

- --j.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Exmh CVS

iD8DBQFCDB96MJF5cimLx9ARAtlNAJ9u4LB+Rh1PWAFspl1039VT7nd+ZgCeNcSq
WcHy9jePbI80qjpu8IK0i+s=
=BOLJ
-----END PGP SIGNATURE-----

Reply via email to