XSS (Cross Site Scripting) Cheat Sheet
Esp: for filter evasion
By RSnake
Note from the author: XSS is Cross Site Scripting. If you don't know how XSS (Cross Site Scripting) works, this page probably won't help you. This page is for people who already understand the basics of XSS attacks but want a deep understanding of the nuances regarding filter evasion. This page will also not show you how to mitigate XSS vectors or how to write the actual cookie/credential stealing/replay/session riding portion of the attack. It will simply show the underlying methodology and you can infer the rest. Also, please note my XSS page has been replicated by the OWASP 2.0 Guide in the Appendix section with my permission. However, because this is a living document I suggest you continue to use this site to stay up to date.
Also, please note that most of these cross site scripting vectors have been tested in the browsers listed at the bottom of the page, however, if you have specific concerns about outdated or obscure versions please download them from Evolt. Please see the XML format of the XSS Cheat Sheet if you intend to use CAL9000 or other automated tools. If you have an RSS reader feel free to subscribe to the Web Application Security RSS feed below, or join the forum:
XSS (Cross Site Scripting):
- XSS locator. Inject this string, and in most cases where a script is vulnerable with no special XSS vector requirements the word "XSS" will pop up. You'll need to replace the "&" with "%26" if you are submitting this XSS string via HTTP GET or it will be ignored and everything after it will be interpreted as another variable (use the URL encoding calculator below to encode the entire string). Tip: if you're in a rush and need to quickly check a page, often times injecting the depreciated "<PLAINTEXT>" tag will be enough to check to see if something is vulnerable to XSS by messing up the output appreciably:
XSS locator 2. If you don't have much space and know there is no vulnerable JavaScript on the page, this string is a nice compact XSS injection check. View source after injecting it and look for <XSS verses <XSS to see if it is vulnerable:
No filter evasion. This is a normal XSS JavaScript injection, and most likely to get caught but I suggest trying it first (the quotes are not required in any modern browser so they are omitted here):
Image XSS using the JavaScript directive:
Case insensitive XSS attack vector:
HTML entities (the semicolons are required for this to work):
Grave accent obfuscation (If you need to use both double and single quotes you can use a grave accent to encapsulate the JavaScript string - this is also useful because lots of cross site scripting filters don't know about grave accents):
Malformed IMG tags. Originally found by Begeek (but cleaned up and shortened to work in all browsers), this XSS vector uses the relaxed rendering engine to create our XSS vector within an IMG tag that should be encapsulated within quotes. I assume this was originally meant to correct sloppy coding. This would make it significantly more difficult to correctly parse apart an HTML tag:
fromCharCode (if no quotes of any kind are allowed you can eval() a fromCharCode in JavaScript to create any XSS vector you need):
UTF-8 Unicode encoding (all of the XSS examples that use a javascript: directive inside of an <IMG tag will not work in Firefox or Netscape 8.1+ in the Gecko rendering engine mode). Use the XSS calculator for more information:
Long UTF-8 Unicode encoding without semicolons (this is often effective in XSS that attempts to look for "&#XX;", since most people don't know about padding - up to 7 numeric characters total). This is also useful against people who decode against strings like $tmp_string =~ s/.*/&#(/d+);.*/$1/; which incorrectly assumes a semicolon is required to terminate a html encoded string (I've seen this in the wild):
Hex encoding without semicolons (this is also a viable XSS attack against the above string $tmp_string =~ s/.*/&#(/d+);.*/$1/; which assumes that there is a numeric character following the pound symbol - which is not true with hex HTML characters). Use the XSS calculator for more information:
Embedded tab to break up the cross site scripting attack:
Embedded encoded tab to break up XSS. For some reason Opera does not allow the encoded tab, but it does allow the previous tab XSS and encoded newline and carriage returns below:
Embeded newline to break up XSS. Some websites claim that any of the chars 09-13 (decimal) will work for this attack. That is incorrect. Only 09 (horizontal tab), 10 (newline) and 13 (carriage return) work. See the ascii chart for more details. The following four XSS examples illustrate this vector:
Embedded carriage return to break up XSS (Note: with the above I am making these strings longer than they have to be because the zeros could be omitted. Often I've seen filters that assume the hex and dec encoding has to be two or three characters. The real rule is 1-7 characters.):
Multiline Injected JavaScript using ASCII carriage returns (same as above only a more extreme example of this XSS vector) these are not spaces just one of the three characters as described above:
Null breaks up JavaScript directive. Okay, I lied, null chars also work as XSS vectors but not like above, you need to inject them directly using something like Burp Proxy or use %00 in the URL string or if you want to write your own injection tool you can either use vim (^V^@ will produce a null) or the following program to generate it into a text file. Okay, I lied again, older versions of Opera (circa 7.11 on Windows) were vulnerable to one additional char 173 (the soft hypen control char). But the null char %00 is much more useful and helped me bypass certain real world filters with a variation on this example:
Null breaks up cross site scripting vector. Here is a little known XSS attack vector using null characters. You can actually break up the HTML itself using the same nulls as shown above. I've seen this vector bypass some of the most restrictive XSS filters to date:
Spaces and meta chars before the JavaScript in images for XSS (this is useful if the pattern match doesn't take into account spaces in the word "javascript:" -which is correct since that won't render- and makes the false assumption that you can't have a space between the quote and the "javascript:" keyword. The actual reality is you can have any char from 1-32 in decimal):
Non-alpha-non-digit XSS. While I was reading the Firefox HTML parser I found that it assumes a non-alpha-non-digit is not valid after an HTML keyword and therefor considers it to be a whitespace or non-valid token after an HTML tag. The problem is that some XSS filters assume that the tag they are looking for is broken up by whitespace. For example "<SCRIPT/s" != "<SCRIPT/XSS/s":
Non-alpha-non-digit part 2 XSS. yawnmouth brought my attention to this vector, based on the same idea as above, however, I expanded on it, using my fuzzer. The Gecko rendering engine allows for any character other than letters, numbers or encapsulation chars (like quotes, angle brackets, etc...) between the event handler and the equals sign, making it easier to bypass cross site scripting blocks. Note that this does not apply to the grave accent char as seen here:
Extraneous open brackets. Submitted by Franz Sedlmaier, this XSS vector could defeat certain detection engines that work by first using matching pairs of open and close angle brackets and then by doing a comparison of the tag inside, instead of a more efficient algorythm like Boyer-Moore that looks for entire string matches of the open angle bracket and associated tag (post de-obfuscation, of course). The double slash comments out the ending extraneous bracket to supress a JavaScript error:
No closing script tags. In Firefox and Netscape 8.1 in the Gecko rendering engine mode you don't actually need the "></SCRIPT>" portion of this Cross Site Scripting vector. Firefox assumes it's safe to close the HTML tag and add closing tags for you. How thoughtful! Unlike the next one, which doesn't effect Firefox, this does not require any additional HTML below it. You can add quotes if you need to, but they're not needed generally, although beware, I have no idea what the HTML will end up looking like once this is injected:
Protocol resolution in script tags. This particular variant was submitted by Łukasz Pilorz and was based partially off of Ozh's protocol resolution bypass below. This cross site scripting example works in IE, Netscape in IE rendering mode and Opera if you add in a </SCRIPT> tag at the end. However, this is especially useful where space is an issue, and of course, the shorter your domain, the better. The ".j" is valid, regardless of the MIME type because the browser knows it in context of a SCRIPT tag.
Half open HTML/JavaScript XSS vector. Unlike Firefox the IE rendering engine doesn't add extra data to your page, but it does allow the javascript: directive in images. This is useful as a vector because it doesn't require a close angle bracket. This assumes there is any HTML tag below where you are injecting this cross site scripting vector. Even though there is no close ">" tag the tags below it will close it. A note: this does mess up the HTML, depending on what HTML is beneath it. It gets around the following NIDS regex: /((/%3D)|(=))[^/n]*((/%3C)|<)[^/n]+((/%3E)|>)/ because it doesn't require the end ">". As a side note, this was also affective against a real world XSS filter I came across using an open ended <IFRAME tag instead of an <IMG tag:
Double open angle brackets. This is an odd one that Steven Christey brought to my attention. At first I misclassified this as the same XSS vector as above but it's surprisingly different. Using an open angle bracket at the end of the vector instead of a close angle bracket causes different behavior in Netscape Gecko rendering. Without it, Firefox will work but Netscape won't:
XSS with no single quotes or double quotes or semicolons:
Escaping JavaScript escapes. When the application is written to output some user information inside of a JavaScript like the following: <SCRIPT>var a="$ENV{QUERY_STRING}";</SCRIPT> and you want to inject your own JavaScript into it but the server side application escapes certain quotes you can circumvent that by escaping their escape character. When this is gets injected it will read <SCRIPT>var a="//";alert('XSS');//";</SCRIPT> which ends up un-escaping the double quote and causing the Cross Site Scripting vector to fire. The XSS locator uses this method.:
End title tag. This is a simple XSS vector that closes <TITLE> tags, which can encapsulate the malicious cross site scripting attack:
BODY tag (I like this method because it doesn't require using any variants of "javascript:" or "<SCRIPT..." to accomplish the XSS attack). Dan Crowley additionally noted that you can put a space before the equals sign ("onload=" != "onload ="):
Event Handlers that can be used in similar XSS attacks to the one above (this is the most comprehensive list on the net, at the time of this writing). Please note I have excluded browser support from this section because each one may have different results in different browsers. Thanks to Rene Ledosquet for the HTML+TIME updates:
& JavaScript includes (works in Netscape 4.x):
LAYER (also only works in Netscape 4.x)
Remote style sheet (using something as simple as a remote style sheet you can include your XSS as the style parameter can be redefined using an embedded expression.) This only works in IE and Netscape 8.1+ in IE rendering engine mode. Notice that there is nothing on the page to show that there is included JavaScript. Note: With all of these remote style sheet examples they use the body tag, so it won't work unless there is some content on the page other than the vector itself, so you'll need to add a single letter to the page to make it work if it's an otherwise blank page:
Remote style sheet part 2 (this works the same as above, but uses a <STYLE> tag instead of a <LINK> tag). A slight variation on this vector was used to hack Google Desktop. As a side note, you can remove the end </STYLE> tag if there is HTML immediately after the vector to close it. This is useful if you cannot have either an equals sign or a slash in your cross site scripting attack, which has come up at least once in the real world:
Remote style sheet part 3. This only works in Opera but is fairly tricky. According to RFC2616 setting a link header is not part of the HTTP1.1 spec, however some browsers still allow it (like Firefox and Opera). The trick here is that I am setting a header (which is basically no different than in the HTTP header saying Link: <http://ha.ckers.org/xss.css>; REL=stylesheet) and the remote style sheet with my cross site scripting vector is running the JavaScript, which is not supported in FireFox:
Remote style sheet part 4. This only works in Gecko rendering engines and works by binding an XUL file to the parent page. I think the irony here is that Netscape assumes that Gecko is safer and therefor is vulnerable to this for the vast majority of sites:
Local htc file. This is a little different than the above two cross site scripting vectors because it uses an .htc file which must be on the same server as the XSS vector. The example file works by pulling in the JavaScript and running it as part of the style attribute:
List-style-image. Fairly esoteric issue dealing with embedding images for bulleted lists. This will only work in the IE rendering engine because of the JavaScript directive. Not a particularly useful cross site scripting vector:
VBscript in an image:
Mocha (older versions of Netscape only):
Livescript (older versions of Netscape only):
US-ASCII encoding (found by Kurt Huwig). This uses malformed ASCII encoding with 7 bits instead of 8. This XSS may bypass many content filters but only works if the host transmits in US-ASCII encoding, or if you set the encoding yourself. This is more useful against web application firewall cross site scripting evasion than it is server side filter evasion. Apache Tomcat is the only known server that transmits in US-ASCII encoding:
META (the odd thing about meta refresh is that it doesn't send a referrer in the header - so it can be used for certain types of attacks where you need to get rid of referring URLs):
META using data: directive URL scheme. This is nice because it also doesn't have anything visibly that has the word SCRIPT or the JavaScript directive in it, because it utilizes base64 encoding. Please see RFC 2397 for more details or go here or here to encode your own. You can also use the XSS calculator below if you just want to encode raw HTML or JavaScript as it has a Base64 encoding method:
META with additional URL parameter. If the target website attempts to see if the URL contains "http://" at the beginning you can evade it with the following technique (Submitted by Moritz Naumann):
IFRAME (if iframes are allowed there are a lot of other XSS problems as well):
FRAME (frames have the same sorts of XSS problems as iframes):
TABLE (who would have thought tables were XSS targets... except me, of course):
TD (just like above, TD's are vulnerable to BACKGROUNDs containing JavaScript XSS vectors):
DIV background-image with unicoded XSS exploit (this has been modified slightly to obfuscate the url parameter). The original vulnerability was found by Renaud Lifchitz as a vulnerability in Hotmail:
DIV background-image plus extra characters. I built a quick XSS fuzzer to detect any erroneous characters that are allowed after the open parenthesis but before the JavaScript directive in IE and Netscape 8.1 in secure site mode. These are in decimal but you can include hex and add padding of course. (Any of the following chars can be used: 1-32, 34, 39, 160, 8192-8.13, 12288, 65279):
DIV expression - a variant of this was effective against a real world cross site scripting filter using a newline between the colon and "expression":
STYLE tags with broken up JavaScript for XSS (this XSS at times sends IE into an infinite loop of alerts):
STYLE attribute using a comment to break up expression (Thanks to Roman Ivanov for this one):
Anonymous HTML with STYLE attribute (IE and Netscape 8.1+ in IE rendering engine mode don't really care if the HTML tag you build exists or not, as long as it starts with an open angle bracket and a letter):
IMG STYLE with expression (this is really a hybrid of the above XSS vectors, but it really does show how hard STYLE tags can be to parse apart, like above this can send IE into a loop):
STYLE tag (Older versions of Netscape only):
STYLE tag using background-image:
Downlevel-Hidden block (only works in IE5.0 and later and Netscape 8.1 in IE rendering engine mode). Some websites consider anything inside a comment block to be safe and therefore does not need to be removed, which allows our Cross Site Scripting vector. Or the system could add comment tags around something to attempt to render it harmless. As we can see, that probably wouldn't do the job:
BASE tag. Works in IE and Netscape 8.1 in safe mode. You need the // to comment out the next characters so you won't get a JavaScript error and your XSS tag will render. Also, this relies on the fact that the website uses dynamically placed images like "images/image.jpg" rather than full paths. If the path includes a leading forward slash like "/images/image.jpg" you can remove one slash from this vector (as long as there are two to begin the comment this will work):
OBJECT tag (if they allow objects, you can also inject virus payloads to infect the users, etc. and same with the APPLET tag). The linked file is actually an HTML file that can contain your XSS:
Using an OBJECT tag you can embed XSS directly (this is unverified so no browser support is added):
Using an OBJECT tag you can embed a Flash movie that contains XSS:
Using the above ActionScript inside flash can obfuscate your XSS vector:
XML namespace. The htc file must be located on the same server as your XSS vector:
XML data island with CDATA obfuscation (this XSS attack works only in IE and Netscape 8.1 in IE rendering engine mode) - vector found by Sec Consult while auditing Yahoo:
XML data island with comment obfuscation (this is another take on the same exploit that doesn't use CDATA fields, but rather uses comments to break up the javascript directive):
Locally hosted XML with embedded JavaScript that is generated using an XML data island. This is the same as above but instead referrs to a locally hosted (must be on the same server) XML file that contains your cross site scripting vector. You can see the result here:
HTML+TIME in XML. This is how Grey Magic hacked Hotmail and Yahoo!. This only works in Internet Explorer and Netscape 8.1 in IE rendering engine mode and remember that you need to be between HTML and BODY tags for this to work:
Assuming you can only fit in a few characters and it filters against ".js" you can rename your JavaScript file to an image as an XSS vector:
SSI (Server Side Includes) requires SSI to be installed on the server to use this XSS vector. I probably don't need to mention this, but if you can run commands on the server there are no doubt much more serious issues:
PHP - requires PHP to be installed on the server to use this XSS vector. Again, if you can run any scripts remotely like this, there are probably much more dire issues:
IMG Embedded commands - this works when the webpage where this is injected (like a web-board) is behind password protection and that password protection works with other commands on the same domain. This can be used to delete users, add users (if the user who visits the page is an administrator), send credentials elsewhere, etc.... This is one of the lesser used but more useful XSS vectors:
IMG Embedded commands part II - this is more scary because there are absolutely no identifiers that make it look suspicious other than it is not hosted on your own domain. The vector uses a 302 or 304 (others work too) to redirect the image back to a command. So a normal <IMG SRC="http://badguy.com/a.jpg"> could actually be an attack vector to run commands as the user who views the image link. Here is the .htaccess (under Apache) line to accomplish the vector (thanks to Timo for part of this):
Cookie manipulation - admittidly this is pretty obscure but I have seen a few examples where <META is allowed and you can use it to overwrite cookies. There are other examples of sites where instead of fetching the username from a database it is stored inside of a cookie to be displayed only to the user who visits the page. With these two scenarios combined you can modify the victim's cookie which will be displayed back to them as JavaScript (you can also use this to log people out or change their user states, get them to log in as you, etc...):
UTF-7 encoding - if the page that the XSS resides on doesn't provide a page charset header, or any browser that is set to UTF-7 encoding can be exploited with the following (Thanks to Roman Ivanov for this one). Click here for an example (you don't need the charset statement if the user's browser is set to auto-detect and there is no overriding content-types on the page in Internet Explorer and Netscape 8.1 in IE rendering engine mode). This does not work in any modern bro