reddit hackernews mail facebook facebook linkedin


Cross-Site Request Forgery aka CSRF is an attack unintentionally triggered by the user himself. It sends HTTP requests to execute unexpected actions in different ways: trough img tag to perform GET requests or with Ajax requests when POST is required. You can learn basic CSRF in DVWA.

To perform this CSRF you firstly need to log in, then you must visit a malicious site who will perform a stealth HTTP request who will submit the change password form with specific values.


The original request can be found by using a local proxy like Burp Suite or analyzing HTTP headers with a browser extension like Live HTTP Headers. The payload:

    <title>My malicious website</title>
    <p>It works like a charm!</p>
    <img src="" width="1" height="1" />

The attributes width and height are only used to hide the broken image. This code displays:

dvwa csrf low

The hash of the password changed in my database from 5f4dcc3b5aa765d61d8327deb882cf99 to ab4f63f9ac65152575886860dde480a1, so it worked perfectly… We can now take a look at the Apache logs.

Access log of the malicious server:

dvwa csrf malicious

Access log of the DVWA server:

dvwa csrf access

However an error has been generated because we tried to display an image which is not:

dvwa csrf error


In this level a protection has been implemented. The developper checks the referer to be sure that the incoming request comes from the same local server. However this control is defective. If I rename my script on my malicious server from payload.html to it will work. Plus the regexp is mal formed and doesn’t protect special chars like dot ‘.’ so 127908071.html or aa127U0B0K1zz.php are also valid filenames.


Finally this level is well protected. The old password is required (which should be only known by the concerned user himself) and cannot be predicted in an automated script.

There is differents ways to secure sensitive actions. A well known method is to use a random token placed in the form and checked after the submission. Another approach is to configure the sessions to close with the browser and with a limited timelife. That way the time window when the vulnerability is effective is shorter.

External resources