On 26/10/11 04:38, Denis A. Altoé Falqueto wrote:
On Tue, Oct 25, 2011 at 3:44 PM, Leonid Isaev<lisaev@xxxxxxxxxxxx> wrote:
OK, remarks here:
1. Web of trust is something relevant when people actually know each other
either directly or indirectly (e.g. through mutual friends). When developers
are concerned, for any distro, this concept looses its meaning, because you
have no way of knowing them and just have to trust them. This is why the likes
fedora and debian don'teven have this TrustAll option (at least I am unaware of
such) -- the keys are trusted always. You either have to blindly trust devs key
at the gpg level, or use TrustAll.
2. Don't just import keys when pacman asks you, because this opens you to an
attack you described. Instead, import keys from the website manually and then
be cautious when pacman says that a key is invalid/missing.
3. Due to (1) TrustAll is likely to stay, but you can always replace Optional
with Required in due time.
The trust problem is complex, indeed, but we can at least mitigate it
doing the following (it's what I do):
1. set TrustedOnly, instead of TrustAll
2. import the keys when pacman asks
3. # pacman-key --edit-key<email or id for key>. That will open a gpg session.
4. go to http://www.archlinux.org/developers/ and/or
http://www.archlinux.org/trustedusers/ to check the new signatures
5. sign the key, checking if the fingerprint is correct, according to
the websites from step 4
5. perform save to apply the changes
That way, one can be a little more secure when trusting the keys. The
point is always checking with different places. Today, there are the
keyservers and the Arch developer info pages. Some day, there could be
more options (read-only wiki page, fixed BBS posts), so if one is
compromised, the others can serve as checkpoints for integrity.
IMHO, I don't like TrustAll very much (and the equivalents concepts in
other distributions). It takes the responsibility from the users, who
are the ultimate decision makers of their systems. But that is just my
opinion (not an invitation to a long pointless discussion). We have
options enough to satisfy everyone.
This is something that will be improved over time... Although it is not
finally decided yet, there will likely be a limited number of Arch
master keys that a user would have to verify, import and locally sign.
These would be very widely published so the user has confidence they are
importing the correct key. The master keys would then be used to sign
all the devs keys, so the dev keys become trusted through the web of
trust rather than having to manually trust them.
But all this takes time. We will get there eventually...
Allan
