Re: CVE-2013-4287 Algorithmic complexity vulnerability in RubyGems 2.0.7 and older

[Eric Hodel <drbrain () segment7 net>]

On Sep 18, 2013, at 15:05, Alexander Cherepanov <cherepan () mccme ru> wrote:
> On 2013-09-18 04:11, Eric Hodel wrote:
> > Here's a new patch to go with the new (unassigned) CVE.  This new patch replaces regular expression matches that are 
> > susceptible to backtracking with a parser-like approach.
>
> According to your patch 'versions have only one "-" (per semver)'. This
> means that "*" after "(#{VERSION_PATTERN})" in ANCHORED_VERSION_PATTERN
> is a bug. It should be "?". If you fix it then there should be no
> problem with VERSION_PATTERN at all. AFAICT VERSION_PATTERN gives you a
> linear complexity. Hence there is no need to suppress backtracking…

Good catch, I think this is a bug, but I'll need to check with the person who added that.

> > This patch applies to RubyGems 2.1.x releases.  I will create patches for RubyGems 1.8.23.1, 1.8.26, 2.0.9 and 2.1.4 
> > if it there is no obvious flaw seen in it.
> >
> > I would like to release this fix by Monday, 23 September as I will be traveling mid-week.
> >
> > The vulnerable regular expression constants are still present, but I can't think of a way to construct them that 
> > does not allow backtracking.
>
> ...but if you really want to suppress backtracking (say, for
> optimization) it is easy: either atomic grouping for every repetition
> (exactly the way you have already done but for other repetitions also)
> or add extra "+" after each "+" and "*". That's according to
> http://www.ruby-doc.org/core-2.0.0/Regexp.html .

Thank you.  I glossed over the * in ANCHORED_VERSION_PATTERN, and this fixes the problem with minimal change (something 
I would prefer for a security fix).

Here is a complete updated patch including the backtracking and extra "-" fixes:

Attachment: CVE-2013-XXXX.2.patch
Description:

The same script as my previous message can be used to verify it.