Brainfuck beware: JavaScript is after you!

tl;dr I just made a tool to transform any javascript code into an equivalent sequence of ()[]{}!+ characters. You can try it here, or grab it from github or npm. Keep on reading if you want to know how it works.

Non alphanumeric JavaScript

What do you know about non-alphanumeric XSS?

The other day one of my friends asked me that question on IRC, pointing me to some articles on sla.ckers.org where they tried to create some scripts like alert(1) with non-alphanumeric characters.

As a security researcher and a penetration tester, he insisted that extending that concept to any javascript source would be really useful for bypassing IDSs, IPSs and WAFs. So challange accepted!

Alphabet

Many alphabets could do the job, but just for fun, I tried to keep it as small as possible, using only the following characters:

[ and ] to access array elements, objects properties, get numbers and cast elements to strings.
( and ) to call functions and avoid parsing errors.
+ to append strings, sum and cast elements to numbers.
! to cast elements to booleans.
{ and } to get NaN and the infamous string "[object Object]"

Numbers

To start our journey to the world of brackets, lets represent the numbers with our new alphabet.

0 is easily obtained by casting an empty array like this +[]. In a similar way, we can cast the empty array to boolean to get true, and then to 1 with +!![]. Those numbers, along with + would be enough to get every natural.

But if we take advantage of JavaScript coercion of types, we can reduce the size of the sequence of the numbers in two ways.

First, if we add a number and a boolean, both operands would be casted to numbers. So instead of using sums of ones to generate larger values, we can add just a 1 and a sequence of trues (we can use more than one true at a time beacuse addition is left-to-right assosiative). For instance, here is 4: !+[]+!![]+!![]+!![].

The second idea is to get strings representing large numbers and cast them in order to get a shorter sequence of symbols. Once we obtained all the possible digits like we did above with 4, we can get the desired string by adding the first digit to [] (to make it a character), and combinig all of them with + (with the necessary parens). Once again, the left-to-right assosiativite would save us lots of chars. Finally, we only need to cast that. Doing this, 12 would look like this: +((+!![]+[])+(!+[]+!![])).

The second idea is to reuse what we’ve done above in order to get a shorter sequence of symbols. The main purpose of doing this is to represent bigger numbers without the need to sum 1 each time to get to our number, so instead we get it’s string representation and cast it to number. For example, representing 12 adding ones would be: (!+[]+!![]+!![]+!![]+!![]+!![]+!![]+!![]+!![]+!![]+!![]+!![]), but by resuing 1 and 2 we can be represen it like this: +((+!![]+[])+(!+[]+!![])). Here, we have casted the first digit to string, added the second, and then, converted everything to a number. Speaking in terms of code, on the first case we did a simple sum: (1+1+1+1+1+1+1+1+1+1+1+1); and on the second one we concatenated two strings and casted them into a number like this: +("1"+2).

Having said that, here is a table of all the possible digits:

+[]
+!![]
!+[]+!![]
!+[]+!![]+!![]
!+[]+!![]+!![]+!![]
!+[]+!![]+!![]+!![]+!![]
!+[]+!![]+!![]+!![]+!![]+!![]
!+[]+!![]+!![]+!![]+!![]+!![]+!![]
!+[]+!![]+!![]+!![]+!![]+!![]+!![]+!![]
!+[]+!![]+!![]+!![]+!![]+!![]+!![]+!![]+!![]

Base elements and strings

Now that we have numbers, lets go for more interesting elements from which we can obtain characters:

true as we have already seen, can be obtained from !![]
false from ![]
undefined by accessing to non-existing element to an array: [][+[]]
NaN is the result of trying to cast an object to number: +{}
"[object Object]" with []+{}

Casting them to string (when necessary) and accessing those like arrays will give us single characters, from which we can even get more strings! These are (the space), "[", "]", "a", "b", "c", "d", "e", "f", "i", "j", "l", "n", "N", "o", "O", "r", "s", "t" and "u". By combining them with numbers we can get "1e100" and "1e1000", which when casted to numbers would result in 1e+100 and Infinity. And by casting them back to strings we can manage to get "y", "I" and "+".

Gathering functions from available characters

By combining those characters, we can only get these JavaScript functions and type names: "call", "concat", "constructor", "join", "slice" and "sort".

Playing with our alphabet and these strings, we can get the following functions:

Function from array["sort"]["constructor"]
Array from array["constructor"]
Bolean from false["constructor"]
Number from 0["constructor"]
Object fom {}["constructor"]
String fom string["constructor"]
Function.prototype.call from f["call"]
String.prototype.concat from string["concat"]
Array.prototype.join from array["join"]
Array.prototype.slice from array["slice"]
Array.prototype.sort from array["sort"]

Unluckily, none of these functions would give us new characters, but don’t loose your hope yet!

Exploting the DOM for fun and characters

If we sacrifice some portabilty and constraint the scripts to webpages, we can take for granted that DOM elements would be available, and get the remaining characters.

One interesting function that becames available is window.unescape which would give us all the ASCII characters by calling window.unescape("%" + HEXA_ASCII_VAL).

All we are missing to get unescape is the "p" character. So once again we make a trade-off, sacrificing some more portability to get it. If we know that we are in a webpage served over HTTP or HTTPS we can asume that by casting window.location to string, and getting its third character we would obtain the precious "p".

But how can we obtain the window.location object if we don’t have access to window yet? Luckly JavaScript, being so premissive, would give that object by doing this:

Function("return location")()

And with location now we can have three more characters "h", "p", "/", escape and unescape functions!

If we could get the character "%" we would be able to get the rest by calling unescape("%" + HEXA_ASCII_VALUE). Luckly, escaping "[" yields the string "%5B, and from that, we can obtain the percentage sign.

Now, we can reach any ASCII character like this:

[][(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[+!![]]+(!![]+[])[+!![]]+(!![]+[])[+[]]][([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+([]+{})[+!![]]+([][+[]]+[])[+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+(!![]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+[]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(!![]+[])[+[]]+([]+{})[+!![]]+(!![]+[])[+!![]]]((!![]+[])[+!![]]+(!![]+[])[!+[]+!![]+!![]]+(!![]+[])[+[]]+([][+[]]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+!![]]+([]+{})[!+[]+!![]+!![]+!![]+!![]+!![]+!![]]+([][+[]]+[])[+[]]+([][+[]]+[])[+!![]]+(!![]+[])[!+[]+!![]+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(+[]+{})[+!![]]+([]+[][(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[+!![]]+(!![]+[])[+!![]]+(!![]+[])[+[]]][([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+([]+{})[+!![]]+([][+[]]+[])[+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+(!![]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+[]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(!![]+[])[+[]]+([]+{})[+!![]]+(!![]+[])[+!![]]]((!![]+[])[+!![]]+(!![]+[])[!+[]+!![]+!![]]+(!![]+[])[+[]]+([][+[]]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+!![]]+([]+{})[!+[]+!![]+!![]+!![]+!![]+!![]+!![]]+(![]+[])[+[]+!![]+!![]]+([]+{})[+!![]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(+[]+{})[+!![]]+(!![]+[])[+[]]+([][+[]]+[])[!+[]+!![]+!![]+!![]+!![]]+([]+{})[+!![]]+([][+[]]+[])[+!![]])())[!+[]+!![]+!![]]+(!![]+[])[!+[]+!![]+!![]])()([][(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[+!![]]+(!![]+[])[+!![]]+(!![]+[])[+[]]][([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+([]+{})[+!![]]+([][+[]]+[])[+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+(!![]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+[]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(!![]+[])[+[]]+([]+{})[+!![]]+(!![]+[])[+!![]]]((!![]+[])[+!![]]+(!![]+[])[!+[]+!![]+!![]]+(!![]+[])[+[]]+([][+[]]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+!![]]+([]+{})[!+[]+!![]+!![]+!![]+!![]+!![]+!![]]+(!![]+[])[!+[]+!![]+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(+[]+{})[+!![]]+([]+[][(![]+[])[+[]+!![]+!![]+!![]]+([]+{})[+!![]]+(!![]+[])[+!![]]+(!![]+[])[+[]]][([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+([]+{})[+!![]]+([][+[]]+[])[+!![]]+(![]+[])[+[]+!![]+!![]+!![]]+(!![]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+[]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(!![]+[])[+[]]+([]+{})[+!![]]+(!![]+[])[+!![]]]((!![]+[])[+!![]]+(!![]+[])[!+[]+!![]+!![]]+(!![]+[])[+[]]+([][+[]]+[])[+[]]+(!![]+[])[+!![]]+([][+[]]+[])[+!![]]+([]+{})[!+[]+!![]+!![]+!![]+!![]+!![]+!![]]+(![]+[])[+[]+!![]+!![]]+([]+{})[+!![]]+([]+{})[!+[]+!+[]+!+[]+!+[]+!+[]]+(+[]+{})[+!![]]+(!![]+[])[+[]]+([][+[]]+[])[!+[]+!![]+!![]+!![]+!![]]+([]+{})[+!![]]+([][+[]]+[])[+!![]])())[!+[]+!![]+!![]]+(!![]+[])[!+[]+!![]+!![]])()(([]+{})[+[]])[+[]]+HEXA_VALUE)

Finally, all we need to transform a script into symbols, is reading it as a string, encoding it in our alphabet, and use Function as eval.

Hieroglyphy

With the findings in this article, I’ve made a tool for encoding scripts, strings and numbers into this alphabet. It’s available at github, so feel free to fork and modify it. You can also try it online here.

Room from improvement

Both this article and Hieroglyphy are just proof of concepts, there is plenty of room from improvments:

Once we were able to generate all ASCII characters, no effort was made to get the shortest representation of any of them.
When targeting modern browsers only, btoa would be a great help yielding lots of characters in shorter sequences.
Depending on the target, one may select a bigger alphabet for reducing the encoding size.
If we know the domain where the script would be run, more characters can be derived from it.

Acknowledgments

Thanks to Matt for giving me the initial idea, helping me with the development and the redaction of this article.

Don’t miss his future entry on how can you use Hieroglyphy for XSS!

Updates

As many people pointed out, there are other projects that encode javascript in a different way, but all the ones I have seen are either broken, incomplete, or they use a vast number of different characters.
David Herman from mozilla and Martin Kleppe from Ubilabs where both working on this at the same time. David’s version is targeted at ES5 compatible js engines, and doesn’t need a browser. Martin’s trades off some portability by using function’s toString in a commonly implemented non-standard way to use fewer characters.
Unluckly hieroglyphied code can’t run in quirksmode on IE. It’s impressive how MS takes backwards compatibility to the limit, disabling any improvement to the javascript engine (or maybe using another) in this mode.