1 |
On Sun, 8 Jan 2017 10:40:15 -0500 |
2 |
Mike Gilbert <floppym@g.o> wrote: |
3 |
|
4 |
> The content of gentoo-news.git should already be covered by the |
5 |
> detached signatures that are required to be present for each file. |
6 |
> What is the benefit to requiring the commits themselves be signed? |
7 |
|
8 |
Oh, I didn't know about those file signatures. But I think signing the |
9 |
commits would make sense nonetheless, as this offers some advantages: |
10 |
|
11 |
* Commit signatures are easy to verify: Everyone who is interested in |
12 |
verifying their /usr/portage image will already have an infrastructure |
13 |
in place to verify commit signatures, because that's how things are |
14 |
done for repo/gentoo.git. |
15 |
|
16 |
* The detached news signatures are nontrivial to verify (in an |
17 |
automated fashion): Just looping over all news files in the repo and |
18 |
verifying their signatures is not an option, because some of the |
19 |
signatures on older news items can't be verified anymore (expired |
20 |
keys, signatures by retired devs, etc.). Hence, one will have to |
21 |
write some code to verify just the new news items introduced after a |
22 |
git pull. |
23 |
|
24 |
* Commit signatures have slightly better security guarantees: If we |
25 |
only verify the detached signatures, attackers can still mess around |
26 |
with the commit graph; in particular, an MITM attacker could silently |
27 |
drop some of the news during a pull. With commit signatures, the only |
28 |
way for the attacker to achieve this is to pretend there aren't any |
29 |
new commits at all (something the user would probably notice after a |
30 |
while). |
31 |
|
32 |
At the same time, I don't see any disadvantages to requiring commit |
33 |
signatures; does anyone else? |
34 |
|
35 |
Regards, |
36 |
Luis Ressel |