Google Chrome is ditching its XSS detection tool | Hacker News Reader

Google Chrome is ditching its XSS detection tool

nakedsecurity.sophos.com

140 pointswglb27 comments

Disclaimer: I'm working on the Trusted Types project in Google.

To clarify, Trusted Types are not a replacement for XSS auditor. They are both related to XSS, but are fundamentally different and even target different flavors of XSS.

Trusted Types are an opt-in browser API that helps developers prevent DOM-based (~client-side) XSS by mandating that developer-specified rules are applied to data that reaches risky functions (like eval or innerHTML). We're working on having it available as a proper W3C spec. More info at https://bit.ly/trusted-types or https://youtu.be/1KQngEZ8qH4

XSS auditor was an opt-out Chrome only feature that tried to stop reflected (~server-side) XSS payloads from executing after the injection has already happened. It was an now outdated concept. The idea was nice - prevent XSS without changing a bit of code in your application, but now we know this just doesn't work for xss.

> Trusted Types are an opt-in browser API that helps developers prevent DOM-based (~client-side) XSS by mandating that developer-specified rules are applied to data that reaches risky functions (like eval or innerHTML). We're working on having it available as a proper W3C spec. More info at https://bit.ly/trusted-types or https://youtu.be/1KQngEZ8qH4*
Can you expand a little on what Trusted Types gives you if you already have a strict CSP which prevents unsafe-eval/ unsafe-inline and e.g. has a 'self' base-uri set?
I'm assuming it would prevent you injecting other HTML if a DOM-based vulnerability existed.

Trusted Types aim to prevent the injection, XSS-y CSP directives (script-src etc) act as an XSS exploit mitigation that fires after the injection is already there. So, for example, even if the JS execution is stopped, the attacker may still exfiltrate the data via form tags etc.
CSP was traditionally deployed by security folks only and is a bit disconnected from the regular dev workflow. For example, if your application does not conform to CSP (i.e. you have an XSS), you can know that only when you deploy it, and the violations keep coming.
TT are part of your JS program, and are much closer to how developers prevent other bugs in their programs. For example, since the API uses types, you can even set up your build for the application not to compile if innerHTML is used with a string.
Additionally, TT allow the application to be a bit more precise - e.g. maybe you can't refactor the application not to use eval() ever (this is surprisingly common), but would rather make sure that this one eval() instance is secure, and disallow all others. TT solve that elegantly. Your reviewed eval starts using eval(TrustedScript), and all other evals - should they exist - are blocked.
In our experience rolling out CSP, its nonced version (w/ no script-dynamic, no unsafe-*) works well for server-side injections, whereas TT cover the client-side better.
https://github.com/WICG/trusted-types/wiki/FAQ#do-i-need-tru...

Thanks for a very comprehensive answer!

We have APIs to put content in DOM elements without it being interpreted as html, and we have APIs to build DOM elements without using strings. Given that trusted types don't actually prevent xss from occuring and only make your data flow more explicit, why not just recommend to developers that they use the other APIs, instead of adding this new one?

I don't think you can build applications without using the dangerous sinks at all yet. Some common scenarios we know are common:
window.open href Setting text on a script element Setting src of a script element form.action innerHTML
Applications do use these sinks quite often, and some of them cannot be just get ridden of (e.g. href or script.src). Even removing eval takes ages. Complex applications parse HTML from the users, load scripts dynamically, and such.
That said - TT allow you to have such enforcement too - just set a "Content-Security-Policy: trusted-types;" header and all dangerous sinks can never be called. We call it Perfect Types, but it's not yet practical to build client side applications in that setting.

Your response makes it sound like this is intended more for cases where a site uses innerHTML and its too much of a burden to rewrite, so you allow only that innerHTML, but that allowed innerHTML is still susceptible to XSS, so what is the point?
As a side note, having a header to disallow innerHTML and etc is definitely a good step, but the rest of this seems superfluous or even leading to a false sense of security.

Good. Whilst the XSS auditor was able to protect against quite a wide range of payloads for reflected vulns, I think it caused more harm than good. Quite often I'd have to try and argue that an issue was real because the PoC didn't work in Chrome. Often I was able to find a way you could cause the server to modify the payload such that the browser was unable to match up the input and output, and then the payload would execute - but I think the legacy of this feature is that it has caused developers to take rXSS much less seriously.

Interesting! I recently found that Googlebot was susceptible to XSS [1], but it was mitigated by the recent updates (they updated Googlebot from Chrome 41 to the most recent version).
I guess a lot will depend on the new Trusted Types API. If it is opt-in then I imagine there will be period whilst it is adopted - what happens during this time?
[1] https://www.tomanthony.co.uk/blog/xss-attacks-googlebot-inde...

Nice write-up! It's a shame that your report wasn't handled with the usual prompt response & fix.

Scary headline; FTA:
> Don’t worry, though – another, hopefully better, protection measure is on the way.
> Another feature is in development to help: an application programming interface (API) called Trusted Types. Trusted types treats user input as untrustworthy by default and forces developers to take steps to sanitise it before it can be included in a web page.
A better headline may be "Google Chrome replacing XSS Auditor with Trusted Types"

Reminds me of Perl's taint mode

> A better headline may be "Google Chrome replacing XSS Auditor with Trusted Types"
As I think a Googler has mentioned above, the XSS auditor is for reflected XSS vulns (caused by the server unsafely outputting user input). Trusted Types protects against DOM-based vulns, which are more client-side.

> Websites should prevent this kind attack
I've had far too coworkers argue that we shouldn't "waste time" preventing XSS attacks _because_ Chrome already detects and prevents (most of) them.

Your coworkers aren't too smart

This comment doesn't really contribute anything — and it makes a common mistake of treating this as an issue of intelligence when it's more specific security experience and awareness. In particular, I've seen a pathology around XSS where part of the problem is that someone is pretty savvy and downplays the risks thinking “well, I'd never set things up so that would work”, ignoring all of the times they've made mistakes, used a library or vendor product, had projects change hands, etc.

The classic "we already have a wall (1 meter high, that folks do not understand very well)" so why need to build a new wall, on top the current wall or prepare for a potential invasion?"

It's interesting to hear, knowing the the lack of XSS Auditor on Firefox for years was a recurrent topic on Bugzilla. Even IE and Safari got similar protections. Curious to see if the XSSs I found over the years will now work on Chrome.

Well, the age of in-browser reflected xss filters is simply over.
This was a flawed idea for multiple reasons, and they are thankfully now gone from Chrome and Edge (https://portswigger.net/daily-swig/xss-protection-disappears...). I guess from modern browsers only Safari still has one. There are modern XSS defenses you can use, XSS filters just did not stand well the test of time.
I would state that it is good if your XSS works in Chrome, just like it works in Firefox. That XSS should be fixed by the website, instead of that website owners neglecting the fix and assuming they're protected because the alert doesn't fire in Chrome. Especially given that if the attacker spends a bit of time tailoring the payload, they may bypass the auditor.

> That XSS should be fixed by the website, instead of that website owners neglecting the fix and assuming they're protected because the alert doesn't fire in Chrome.
What if a website doesn't fix its XSS vulnerabilities and continues to spew attacker-controlled content? I don't think it will help users to base browser security on "shoulds".
I've been taught that security should be built in layers. Removing a functioning albeit not perfect layer for no good reason is baffling to me.

From a cursory glance, the replacement (Trusted Types), just seems like a very cumbersome runtime type-safety engine kind of thing, shoehorned into a JS/DOM context.
Is that a correct understanding?
If so... Why use this cumbersome API instead of just relying on existing and well-implemented concepts as already present in Flow or Typescript?
That way you could have Flow or Typescript also compile to whatever form this required to have.
Or am I missing some key bits here?

This isn't a programming language type safety mechanism, it's a way to tell the browser to enforce that you can't run code that could potentially be vulnerable to XSS, except in specific places that you designate with policies. This example was illuminating for me: https://developers.google.com/web/updates/2019/02/trusted-ty...

Tomato/tomato.
You are effectively saying that only certain types of strings and DOM structures are allowed on certain properties and attributes.
This is perfectly encodable in a type-system. I don’t see a reason to dismiss that observation outright.

As far as I know typescripts add types to JavaScript which will enforce tests for certain classes of programming errors. Trusted types is a custom validator for what is allowed to be assigned to innerHTML and the like. That check is done during runtime. Typescript doesn't prevent untrusted code from being infected, but hopefully your validator will.

Exactly. That said, because the API itself is typed, it is possible to have the runtime enforcement also verified when statically type-checking your application code (e.g. that innerHTML is passed a TrustedHTML value, and not a string) in Flow, TypeScript or type-aware linters - like this: https://youtu.be/1KQngEZ8qH4?t=1330

I learned about Xss mitigation 3-4 years ago through google gruyere, a tutorial that came with a sandboxed environment to practice in [1]
Where did others learn about XSS mitigation? Any resources to recommend?
[1] https://google-gruyere.appspot.com/

The title is wrong. Google's not ditching anything, it seems.

Reply on news.ycombinator.com