Re: [gentoo-dev] Regarding long delays on GLSA generation

[Alex Legler]

On 18.01.2014 18:38, Pacho Ramos wrote:
> […]
>
> Then, how are you finally going to fix this?

Thank you for finally showing interest in anything we do. Here is how we
are 'finally' going to fix this.

> Only for knowing, I still
> was seeing some delays and, then, I though situation was not improved.
> For example, since this year started, I have only seen 8 GLSAs filled:
> http://www.gentoo.org/security/en/glsa/

Er, we're 18 days in…

You shouldn't look at this purely by the numbers, you should see that we
have now again a steady stream of advisories going not. No gaps like
from 2013-01 to 2013-07. That is largely the effort of the
recently-joined members.

>
> Then, I thought something was still wrong as that rate didn't seem
> enough to me for handling upcoming security issues and the really old
> ones. Also, if you that 8 GLSAs, you will see the only one that has been
> done in a fast way is the ntp one, the other 7 took months (or years) to
> be handled.

So you observed correctly there's still plenty of delays. There are
three parts to an advisory that take time:
- Drafting: Collecting information, linking references, getting package
versions done right (slots are a huge pain still).

- Reviewing: Our current process asks for two independent positive
reviews from other team members before an advisory can be sent.

- Sending: The original author gets a .txt to email and have to check in
the .xml to CVS.

Going through these three steps requires at least three people, and the
(group of) people whose action is required shifts twice. That overall
process is spot #1 we are planning to improve. The current plan contains
requiring only one review and the reviewer sends the advisory directly.
So we go from author -> reviewer 1 -> reviewer 2 -> author to just
author -> reviewer.

Concerning the single steps here are other measures:
- Drafting: Implement a new GLSA format to
  * reduce the amount of editorial text written by us
  * support slots (makes specifying vulnerable ranges in slotted package
    much easier)
  * (cleanup old stuff no longer needed)

- Reviewing: Reduced editorial text means less to review.

- Sending: We want to improve our tooling to automatically send
advisories and push them to a git repository.

The new GLSA format was up for review on -security last week. Next up
will be getting it specified formally, implemented in our tooling,
glsa-check and a new security.g.o frontend. [1]
Then, we can adopt the new workflow.

>
> Then, instead of blaming on how should I have asked for clarification on
> this (well, looks like the main topic here is that I have asked about
> this in ML instead of the real problem :O), I think you should focus on
> explaining how are you fixing this problem.

Your original email didn't reflect actual interest in the details. Now
that we've established you do care, I hope my explanations helped you
out there.

> I have been long time wondering about this because:
> 1. I usually get lots of bugs from alias I am a member whose we go fast
> bumping, calling for stabilization and dropping vulnerable versions and,
> the, the bugs get stalled.
> 2. Once of the machines I maintain would benefit from being able to use
> glsacheck to only update vulnerable packages as not always have enough
> time for updating the full world
>
>
>

[1] Lots of code to be written here. .py+.rb, help wanted!

-- 
Alex Legler <a3li@g.o>
Gentoo Security/Ruby/Infrastructure