1.: Cross Site Request Forgery (XSRF) on almost every front/back-end requests
2.: reflected/stored XSS in search, ProductId/Product Internal name and so on
3.: Session Hijacking
1.: As can be verified with your favorite proxy tool (we use Burp), POST request
parameters are never "fortified" to prevent XSRF: no random token protection can be seen.
For those who don't know what a XSRF is: briefly it is a request that me, the attacker, force you (the victim)
- In GET requests it will be a link like http://x.x.x.x/account/doTransfer?from=666&to=667, where 666 is
a potential victim account and 667 the attacker one.
- In POST requests it will be an auto-submit form or a XMLHttpRequest
(if we would like to be more sophisticated).
I can force a victim to execute such a request in various methods, whose description is out from the scope of this ISSUE:
malicious mail link, link in chat programs, malicious pages, man in the middle attacks, malicious Flash/Applets/ActiveX, and so on.
The quick-and dirty code to make the XSRF attack looks as the following innocuous one:
<form method="POST" id="xsrf" name="xsrf"
<input type=hidden name="isCreate" value="true">
<input type=hidden name="productId" value="hack02">
<input type=hidden name="productTypeId" value="DIGITAL_GOOD">
<input type=hidden name="internalName" value="hack02">
Of course the product-creation mechanism is not finished (we need price, content and ProductName),
but is just to let you understand.
When this JS code will be present in a malicious page (opened by a new tab of the same browser - not Chrome ahah),
his content will be automatically executed and the POST request will be sent to the application: the product with Id=hack02
will be persisted inside the DB. Of course a valid party must be logged in the catalog module, in a way
that the global JSESSIONID cookie value will be the same in every tab of the browser.
Clearly we can do more than this...
2.: As most of the Ofbiz forms are vulnerable to XSS, some reflected and some stored,
exploit them is quite easy: we will exploited only stored ones.
We can for instance replace the value of internalName (that even if it is a needed
parameter is quite un-useful and so prone to store our malicious code) with something
<input type=hidden name="internalName"
The malicious code will display every cookie information in a pop-up, that only the victim
will see: obviously we don't want this.
3.: We can then create a little cookie-grabber servlet that listen for GET request from
our victims, extract the useful parameters and store them in a file or DB, in a way
that wen can hijack the session of the admin/manager.
The internalName value is prone to store our malicious code also because his maxlength
is 255 characters: this gives us a great advantage when creating a complex injection code,
if we don't want to inject a link to the malicious script like
The malicious code will look as the following one:
Of course the code can be a lot shorter, and the "already-exploited-check" can be removed.
After we have a valid JSESSIONID, if we open a browser, go to the grabbed URL (remember document.URL) that will be an
authentication-required resource, the login page will ask us for valid credentials.
In Opera (or Firefox with AnEC Cookie Editor plugin) we can see that a new cookie has been
given to us, because we don't have one. If we modify the JSESSIONID value with the grabbed
one, and we make the previous request another time (just refresh on the login page), then
we are riding the same victim session. If we are lucky and it's an admin, we can do
whatever we want on his/her behalf.
Mitigation can be made in two ways:
- Infrastructure: a web application firewall like ModSecurity can be deployed in front of Tomcat, in enterprise deployments such as
Apache --> mod_ajp --> Tomcat . This will don't fix XSRF attacks, but will mitigate XSS and Session Hijacking.
XSS--> input validation on form parameters and GET/POST request values must be implemented. I was thinking
to do it in org.ofbiz.base.util.UtilValidate, re-using code from Owasp ESAPI project (a really good one), or re-using the ModSecurity
Reg-expression patterns to filter out bad input.
XSRF--> build a class that will implement javax.servlet.Filter and will add to every GET/POST request a random token that will be unique
and will change every time. In this way (if the entropy is enough and the algorithm good, it will be quite impossible to guess it).
Said all of that, we really support Ofbiz!