Question

I’m creating a PHP website which involves users signing up, and I’m wondering about best practices for “email confirmation” codes.

New users must confirm their email addresses – I do this by generating a code and sending it to the user in an email, which he can then use to activate his account. Rather than storing this key in a database, I’m using a handy little workaround: the code is the result of:

md5("xxxxxxxxx".$username."xxxxxxxxxxx".$timestamp."xxxxxxxxx");

Where $timestamp refers to the user-creation time. On the whole I was quite pleased with this, but then I got to thinking, is this secure enough? And what about the possibility of collisions? And I also need to generate codes for password reset, etc. If I used a similar methodology, a collision could result in one user inadvertently resetting another user’s password. And that’s no good.

So how do you do these things? My thoughts was a table of the following format:

codePK (int, a-I), userID (int), type (int), code (varchar 32), date (timestamp)

Where ‘type’ would be 1, 2 or 3 meaning “activation”, “email change” or “password reset”. Is this a good way of doing it? Do you have a better way?

Using a method similar to the above, could I automatically delete anything over two days old without using cron-jobs? My host (nearlyfreespeech.net) does not support them. If at all possible I’d like to avoid having a cron-job on an external host which wget’s a script which deletes things, as that’s just messy =P.

Thanks!
Mala

Update:
To clarify: I’ve realized the only way to securely and safely go about this task is by using a database, which is what the original function was trying to avoid. My question is on how the table (or tables?) should be structured. Somebody suggested I do away with codePK and just make the code a PK. So in short, my question is: is this what you do?

Answer 1

When I need these kinds of tricks it is normally of one of two reasons, both mentioned by you:

1. As a key used for verification emails sent to the user
2. As a key used for password-reset links

Of course there would be numerous other occasions where you would consider using such a construction.

First of all, you should always use some kind of salt that is hidden and that only you know. Note that this salt should be different for each user. The salt could for example be calculated as sha256(something random). This salt should then be stored in the database along with the username and password (hashed with the salt).

What I would do when sending the password reset link is to create another salt (don’t give the user access to anything hashed with your salt. He knows his password, so using bruteforce he could potentially figure out your salt). The other salt, which essentially is only a hash of a random string (you might wanna go for md5 here, as you mentioned that the length is an issue), should you then save into your database.

Often you can just get away with adding an additional column to your users table. This, however, also has a few problems, mainly that once the password has been reset or the user has been activated, you will remove the key from the database, which results in most rows having null values, which in turn brings some other trouble.

What this essentially boils down to:

• Hash your users’ passwords using a unique-for-the-user salt (and perhaps a global, secret salt).
• Generate a key by hashing a number of random or pseudorandom sources like timestamps, mt_rand() or even random.org if you really want random stuff.
• Never use your global salt or the salt that is unique to the user for hashing anything that the user gets access to, including password reset keys, activation keys, etc.

Please not that I am by no means a security expert, and I have probably forgotten a number of things, and I may have mentioned some very bad practice things. Just my 5 cents 😉