RE: phpBB Worm
Interestingly enough the worm was probably developed on *nix and than
checked and corrected to work on Windows:
eval{
while(my @a = getpwent()) { push(@dirs, $a[7]);}
};
push(@dirs, '/ ');
the getpwent function is not supported on Windows. Actually the entire
loop that gets users home directories from the /etc/passwd file is very
*nix centric.
The author found that out, added the eval statement to prevent the
script from crashing on Windows and added the root directory in order to
have at least one entry on windows. This last line actually makes the
entire loop less important.
Additionally, on Windows the worm would affect files on a single disk.
As to which disk exactly, it probably depends on the web server
attacked, and how PHP and Perl are installed and used with the web
server. In some cases, if the web sites and the software do not reside
on the same disk, the worm payload will not work.
Ofer Shezaf, CTO
Breach Security, Inc.
Deployable Application Security
Tel: +972.9.956.0036 ext.212
Cell: +972.54.443.1119
ofers@xxxxxxxxxx
http://www.breach.com
-----Original Message-----
From: Anders Henke [mailto:anders@xxxxxxxxxx]
Sent: Thursday, December 23, 2004 2:31 PM
To: ycw1bh302@xxxxxxxxxxxxxx
Cc: bugtraq@xxxxxxxxxxxxxxxxx
Subject: Re: phpBB Worm
On Dec 22nd 2004, ycw1bh302@xxxxxxxxxxxxxx wrote:
> 1. Why has the worm been as effective on Windows servers as on *nix
servers? At the very least, shouldn't the difference in file and
directory naming cause a problem? I looked at the decoded Perl script,
but I'm not a Perl expert, so I couldn't understand all of it. And what
about the difference in file permissions?
Perl does provide cross-platform-functions for e.g. file access and
there's usually not much of a difference for running a well-written
perl script on Unix as well as on Windows other than the first line
(usually '#!c:\perl\perl.exe -w' on Windows and '#! /usr/bin/perl -w'
on Unix).
However, most Windows-Webservers other than Apache do run any .pl-Script
using the to-be-installed perl interpreter and don't care on the
bang-line.
The documentation found in 'perldoc perlport' does give a closer view
on the few differences when writing cross-plattform perl scripts.
> 2. More importantly, why wasn't the worm's destructive ability
limited by file permissions, especially on *nix servers? If, for
example, an HTML file on the server was uploaded by user bob, and has
permissions of 755, how can the Perl script delete that file? Shouldn't
the Perl script be created with the Perl process's permissions, which
was invoked by PHP, which should have the Web server's permissions,
which should be, at least on most *nix servers, the nobody user?
On shared servers with ISPs caring about security, user CGIs are using
the
suexec mechanism in order to run each customer within his own user's
space.
The downside of using suexec is that PHP as a CGI doesn't offer a small
number of special features some people do believe to be essential, as
well
as some people do write code in a way that making it work on PHP as CGI
is close to 'virtually impossible'. The PHP-Module also allows one to
set PHP-configuration settings via .htaccess; those configuration
changes are also ignored by CGI-PHP and can severely affect the way
an PHP-written application works (or doesn't work).
> This is a big issue on shared servers, or virtual hosts, whatever you
want to call them. Our site is on a shared server, and our site does
not even run phpBB, but most of our HTML files were replaced with the
worm's content. Obviously, then, another site on the server must have
an old version of phpBB. But why could the worm, coming in through
another site, modify files created by other users? Even if the worm's
script ran as the owner of the vulnerable viewtopic.php file, how could
it then modify non-world-writable files created by other users?
Right - if everyone were using e.g. suexec, this would be the case.
As a web host, you've got to chose to run either CGI-PHP or PHP as
module.
Your 'power'-users are calling for the module, the admin keeping
maintenance on an already overloaded server does also all for the module
(the module relieves the web server from forking a seperate process for
running a php-script), only those security-related ones are rejecting
both
mod_perl as well as mod_php and favour 'true' CGIs via suexec.
If your scripts support the fastcgi extension, one might use mod_fastcgi
with suexec support; however, this means one has to setup three
softwares
(fastcgi, suexec, php) and make them work together instead of the
often-recommended 'add mod_php'-Oneliner. As a result, you're spending
much work on a secure system, but your users are still calling for
mod_php
and in case any part of your setup breaks, your whole system is
unusable.
> I have long been concerned with the security of PHP scripts,
especially on shared servers. Since PHP almost always runs as an Apache
module, and Apache usually runs as nobody, one must make files and
directories world-writable for PHP scripts to be able to write to them.
But that means that any process on the server, including anyone's PHP
script, can modify the files.
Yes, you've got the point.
Apache 2 has the ability to run modules per VirtualHost within a
different
user context (perchild MPM).
-According to the Apache documentation, this module is non-functional,
not yet finished and development is not currently active.
-PHP is certainly one of the most interesting modules for this feature,
however, the last time I looked, exactly PHP didn't support it and
Apache
required to have at least one process running per virtualhost (which in
turn would render servers hosting thousands of sites unusable).
-Still today, the php documentation warns from using Apache 2.0 with PHP
in productive environment.
>From a security aspect, the only way for running PHP securely
(with 'secure' from the view of the administrator), CGI is currently
the only way to do so.
Regards,
Anders
--
Schlund + Partner AG Security and System Administration
Brauerstrasse 48 v://49.721.91374.50
D-76135 Karlsruhe f://49.721.91374.225