Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Sun, 25 Sep 2011 18:38:40 +0200
From: Rasmus Lerdorf <rasmus@....net>
To: Pierre Joye <pierre.php@...il.com>
CC: Zeev Suraski <zeev@...d.com>, Vincent Danen <vdanen@...hat.com>, 
 "oss-security@...ts.openwall.com" <oss-security@...ts.openwall.com>,
 "security@....net" <security@....net>, 
 Stas Malyshev <smalyshev@...arcrm.com>
Subject: Re: CVE request: is_a() function may allow arbitrary code execution
 in PHP 5.3.7/5.3.8

On 09/25/2011 04:10 PM, Pierre Joye wrote:
> On Sun, Sep 25, 2011 at 3:47 PM, Zeev Suraski <zeev@...d.com> wrote:
> 
>> There aren't any security issues in PHP in that context.  Assigning a CVE to PHP in that context would create the impression that there is indeed an issue in PHP here.
>> It's not a matter of who's 'guilty' in terms of positioning - but in terms of where the actual security issue resides.  And it does not reside in PHP.
>>
>> So I agree with Stas, it doesn't make sense to have a CVE here.  Otherwise, almost every change we make, including bug fixes, could somehow result in some faulty piece of code somewhere becoming vulnerable to something.
> 
> The whole point is that some code was not having any issue before this
> change. If the check was done earlier using is_a then this unexpected
> behavior will happen, and that actually causes a security issue in
> existing working code. The example in the blog post is very good one,
> it clearly shows that the impact on existing code is not only about
> wrongly implemented autoloader, or someone not disabling
> allow_url_fopen (I can imagine local file include being an issue as
> well under some circumstances).
> 
> All in all, there is no shame or bad image to get a new CVE for
> something like that, I even see it as a good thing as it will:

I didn't read the thread from the beginning, but is there an actual
exploit here? Presumably the autoloader code in question isn't doing an
fopen/eval to execute the code and since allow_url_include is disabled
by default, remote includes aren't an issue in the default install. So
are we talking about the tiny number of people who have explicitly
enabled allow_url_include and are running the code with this bad autoloader?

-Rasmus

Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.