Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:100877
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: pass (pb1.pair.com: domain gmail.com designates 209.85.214.49 as permitted sender)
MIME-Version: 1.0
In-Reply-To: <BN6PR02MB3234EFDCADFC9F9CF7EF43AEBF4B0@BN6PR02MB3234.namprd02.prod.outlook.com>
References: <BN6PR02MB3234EFDCADFC9F9CF7EF43AEBF4B0@BN6PR02MB3234.namprd02.prod.outlook.com>
Date: Thu, 12 Oct 2017 17:01:42 +0200
Message-ID: <CAF+90c_Ft+sth7uo7L6277x-3T14OVe1jYF-GBaUX=Vkr2nEig@mail.gmail.com>
To: Dmitry Stogov <dmitry@zend.com>
Cc: PHP internals list <internals@lists.php.net>, Stanislav Malyshev <smalyshev@gmail.com>, 
	Nikita Popov <nikic@php.net>, Xinchen Hui <laruence@gmail.com>, Sara Golemon <pollita@php.net>, 
	Zeev Suraski <zeev@zend.com>, "rasmus@lerdorf.com" <rasmus@lerdorf.com>
Content-Type: multipart/alternative; boundary="94eb2c05e6e67f884e055b5ad12b"
Subject: Re: Fix for unserialise() "vulnerabilities"
From: nikita.ppv@gmail.com (Nikita Popov)

--94eb2c05e6e67f884e055b5ad12b
Content-Type: text/plain; charset="UTF-8"

On Thu, Oct 12, 2017 at 4:38 PM, Dmitry Stogov <dmitry@zend.com> wrote:

> Hi,
>
>
> I've found, that at least half of unserialise() security problems, occurs
> because of non-symmetric serialize/unserialize assumption, regarding
> references encoded with "r".
>
>
> serialize() assumes it's an object.
>
>
> https://github.com/php/php-src/blob/master/ext/standard/var.c#L828
>
>
> universalize() allows any value.
>
>
> https://github.com/php/php-src/blob/master/ext/standard/
> var_unserializer.re#L677
>
>
> This allows manual crafting of strings that may lead to creation of
> unexpected data structures.
>
> I propose to fix this just by fixing the symmetry.
>
>
> https://gist.github.com/dstogov/53382540bdfee7b6c7dadf142dc437ed
>
>
> This will prohibit, some manually crafted strings.
>
> Of course, this will break few "security" related tests. Especially:
>
>
> > Bug #70284 (Use after free vulnerability in unserialize() with GMP)
> [ext/gmp/tests/bug70284.phpt]
> > Bug #70211 (php 7 ZEND_HASH_IF_FULL_DO_RESIZE use after free)
> [ext/soap/tests/bug70211.phpt]
> > Bug #70172 - Use After Free Vulnerability in unserialize()
> [ext/standard/tests/serialize/bug70172.phpt]
> > Bug #70963 (Unserialize shows UNKNOW in result)
> [ext/standard/tests/serialize/bug70963.phpt]
> > Memleaks if unserialize return a self-referenced array/object
> [ext/standard/tests/serialize/unserialize_mem_leak.phpt]
> > Bug #72433: Use After Free Vulnerability in PHP's GC algorithm and
> unserialize [ext/standard/tests/strings/bug72433.phpt]
>
> Any objections? (this is for master only of course)
>

Hi,

I don't think this will really fix any vulnerabilities, because the core
issue are R references, not r references. If this prevents a vulnerability
using r, you can usually replicate something similar using R instead.

However, I still agree that it makes sense to restrict this. Especially
because unserialize() currently allows creating structures that are just
impossible in plain PHP, such as cyclic arrays without use of references
(GLOBALS notwithstanding).

The check looks too strict to me though. Shouldn't it first DEREF the value
before performing the OBJECT check? (E.g. for something like
"a:3:{i:0;O:8:"stdClass":0:{}i:1;R:2;i:2;r:2;}", in which case r:2 will be
a REF to OBJECT).

Regards,
Nikita

--94eb2c05e6e67f884e055b5ad12b--