Scenario:
There is nothing you can do, except use HTTPS.
It doesn't matter how many cookies you add or what data you hash; it can all be sniffed and sent back to the server.
If you're going to force a user to use a single UA throughout the life of their request, that can help: you don't need any special hashing business, because you're hashing it into $_SESSION
which neither the user nor the hijacker can access directly, so why bother hashing it? Might as well just store $_SESSION["reportedUA"] = $_SERVER["HTTP_USER_AGENT"]
on log-in and then check reportedUA
on each request.
That, too, is trivial to hijack, once you realise it's happening, as you need only sniff the reported UA when you sniff the session cookie, and start using that.
What next? IP address? Session hijacking might be happening from behind a NAT, in which case you're screwed. Your users might be using dial-up, in which case they're screwed.
This problem has no solution: there is no way. There couldn't be a way. If a hacker can see your session cookies, then they can mess you up, because there's no additional information or challenge related to something only the user knows (i.e. password) that's sent with each request.
The only way to make the session secure is to secure the entire session.
It is enough to store just user login (or user id) in the session.
To prevent session fixation/hijacking everything you need is just to implement simple algorythm (pseudocode):
if (!isset($_SESSION['hash']) {
$_SESSION['hash'] = md5(!empty($_SERVER['HTTP_USER_AGENT']) ? $_SERVER['HTTP_USER_AGENT'] : 'no ua');
} else if ($_SESSION['hash'] != md5(!empty($_SERVER['HTTP_USER_AGENT']) ? $_SERVER['HTTP_USER_AGENT'] : 'no ua')) {
session_regenerate_id();
$_SESSION = array();
$_SESSION['hash'] = md5(!empty($_SERVER['HTTP_USER_AGENT']) ? $_SERVER['HTTP_USER_AGENT'] : 'no ua');
}
You could move the hash calculation into some function to prevent of duplication, i've just shown a sketch of possible protection.
This is how I implemented this kind of protection in my kohana session class:
abstract class Session extends Kohana_Session
{
public function read($id = null)
{
parent::read($id);
$hash = $this->calculateHash();
$sessionHash = $this->get('session_fixation');
if (!$sessionHash) {
$this->set('session_fixation', $hash);
} elseif ($sessionHash != $hash) {
$this->regenerate();
$_SESSION = array();
$this->set('session_fixation', $hash);
}
}
private function calculateHash()
{
$ip = !empty($_SERVER['REMOTE_ADDR']) ? $_SERVER['REMOTE_ADDR'] : '127.0.0.1';
$ua = !empty($_SERVER['HTTP_USER_AGENT']) ? $_SERVER['HTTP_USER_AGENT'] : 'no ua';
$charset = !empty($_SERVER['HTTP_ACCEPT_CHARSET']) ? $_SERVER['HTTP_ACCEPT_CHARSET'] : 'no charset';
$ip = substr($ip, 0, strrpos($ip, '.') - 1);
return md5($ua . $ip . $charset);
}
}
Don't try to write your own session scheme, PHP will do it better.
yes you can add more information to your $_SESSION to help prevent session hijacking
for example I generate a fingerprint by combining a secret phrase or random data with the user agent and the session_id() and hash it all. To hijack a session the user would need to figure out a valid session_id, and the hash of the fingerprint. it will look like this. This is a good read
$_SESSION['fingerprint'] = md5('somethingSecret' . $_SERVER['HTTP_USER_AGENT']. session_id());
then you would validate the session like
$check_print = md5('somethingSecret' . $_SERVER['HTTP_USER_AGENT']. session_id());
if($check_print != $_SESSION['fingerprint'] || $_SESSION['authenticated']){
//invalid session
}
As of 15 November, the two answers I have received do not address my question, which was
"Is this [a single session variable] a strong enough security measure by itself?"
This question says yes, but there seems to be some dissension. Here is a summary of the various results:
1) A single session variable is not enough security since a session can be hijacked fairly easily.
2) Since this can occur, no session is truly safe, but it can be made safer with the addition of a fingerprint. This ensures a unique, repeat-able check each time a session needs validation. @zerkms recommends a hash of User-Agent
and a few others (refer to his code).
3) Salting the fingerprint is mostly useless since it obscures the data but is replicated on every client machine, therefore losing its unique-ness.
4) A database solution is useless since it is a client-side problem.
Not the definitive answer I was looking for, but I suppose it will have to do, for lack of anything better.
Reading that has helped/confused me further:
Session hijacking and PHP
Is HTTPS the only defense against Session Hijacking in an open network?
Is this a strong enough security measure by itself,
Set two session variables to validate eachother and/or
Implement database/hash validation
No, and the reason is this: Anything that your valid user can send to your server for authentication (Session ID, cookies, some hashed string, anything!) can be sniffed by others if it's not encrypted. Even if the server processes the data with md5 hashing, salt, double-session-variable checks, id or whatever, and stores that information, it is easily reproduced by the server when it receives the spoofed data again from some other source.
As many people have suggested, SSL is the only way to prevent this type of evesdropping.
It has occurred to me that, were the server to generate a new session id for each request, and allow the browser to reply with it only once, there could theoretically be only one hijacker request or post before the server and the authorized browser knew about it. Still unacceptable, though, 'cause one is enough to do serious damage.
Hey what about this:
Create a single-use GUID and random salt and encrypt it with a shared password using PHP - this is sent as the session id or a cookie.
The client receives the cookie, decrypts it with the shared password using javascript (there are many enc/dec utilities available)
Set the current cookie or session id to the GUID.
That would ensure that nobody could hijack the session unless they knew the password, which is never sent over the network.
SSL seems much easier, and is more secure still.
EDIT: Ok, it's been done - nevermind ;-)