Service Workers: an Introduction | Hacker News Reader

Service Workers: an Introduction

developers.google.com

126 pointsmnazim22 comments

W3C: "Service workers create the opportunity for a bad actor to turn a bad day into a bad eternity."[1] With service workers, cross-site scripting vulnerabilities are forever.

Service workers install by drive-by, which is troubling. In Firefox, check "about:serviceworkers" to see what you have installed. Take a look. You probably have far more service workers active than you thought. There are supposed to be enough cross-site scripting restrictions to keep service workers contained to their origin domain, but some holes have been found. There was a successful service worker attack on Dropbox, since fixed. "If you run a site that serves user files with secret URLs from a shared domain, you need to look out for the Service-Worker: script HTTP header; if you see it, run for the hills."[2]

If an attacker can get one page loaded from a site being attacked, they then own all traffic between the user and the site. This has lots of attack potential.

[1] https://www.w3.org/TR/service-workers/#security-consideratio... [2] https://alf.nu/ServiceWorker

2 quick hopefully informative nits:

It's better to consult the living standard of the editor's draft over the TR ("TR is for the TRash" as they say). The security section has been fleshed out a lot, for example: https://w3c.github.io/ServiceWorker/#security-considerations

In Firefox, about:serviceworkers is in the process of being replaced by about:debugging. The bug is https://bugzilla.mozilla.org/show_bug.cgi?id=1220747 if you want to follow-along, but start re-training your muscle memory now! :)

So Mozilla is making it harder for end users to find and kill service workers?

The main difference between the TR and the suggested document is a weak "privacy" section, suggesting that the data stored by service workers locally should be flushed on user request. Mozilla currently does not allow service workers to run in incognito windows, which sort of complies with that.

> So Mozilla is making it harder for end users to find and kill service workers?

about:serviceworkers was added as a hacky debugging/introspection tool with the expectation that the developer tools team would integrate the functionality into more full-featured, supported tooling. about:debugging is that tooling.

For example, about:debugging's "Workers" tab (on nightly): - Displays all registered Service Workers as well as whether they are stopped, enables starting them if desired, and attaching a debugger to them. - Also displays any open (Dedicated DOM) Worker instances as well as SharedWorker instances and provides debugging support. - Listens for changes and updates the UI in real-time covering whether the ServiceWorker is stopped/running/installing/etc. (about:serviceworkers renders in a single snapshot).

I'm parsing your underlying concern to be that ServiceWorkers are a powerful new capability for websites and that you want to make sure they're not being brushed under the rug as a "don't worry about it, nothing could possi-blye go wrong" (https://frinkiac.com/caption/S06E04/442491) situation.

It's indeed the case that ServiceWorkers are powerful and they open up new avenues of potential misbehavior. While the fundamental event-driven design of SW's where they're only spawned by explicit user action (browsing the site providing the service worker) or user-approved site actions (push notifications which require explicit user approval), there's still the potential for bad actors. Implementers share these concerns, which is where efforts like https://github.com/beverloo/budget-api and (potential? not sure if this is actually implemented?) heuristics like Chrome generating a (desktop) notification if a ServiceWorker gets woken up by a push and doesn't generate its own notification come from.

And I believe it's in these goals of having browsers proactively inhibit bad behavior and make users aware of the behavior of sites they have implicitly or explicitly granted permissions to, that the best defense of users is found. Which is to say, there will always be UI that helps users find and kill service workers, but the design goals are likely to center around a) debugging, b) performance/"what on earth is making my browser/computer run so slow!", or c) storage/"where did all my disk space go?" rather than one of manual gardening of service workers like they are weeds that spring up and users are responsible for cleaning up after.

Chrome also has chrome://serviceworker-internals.

I have no idea why Hampton Farms saw it fit to install a service worker. Furthermore, this isn't gated by a permission dialog like location services or plugin access?

> this isn't gated by a permission dialog like location services or plugin access?

No. The justification is that you'd need to prompt for additional permissions to do almost anything with a service worker, other than react to fetch event.

Note that this is the low-level API. For most people, I think, the recommended approach is to use libraries instead. Google itself has released sw-toolbox and sw-precache. There’s also a web-push server module for push messages.

You gotta see the humor in Google being of the opinion that a "service worker is a script". I mean, we've all kinda expected that Larry Page is under the assumption that we're all just subroutines running in his "matrix", but now it's official.

I clicked the link half-expecting to find an article about pink-collar jobs, but nope, just another badly named technical concept.

This seems like a security nightmare. I sincerely hope there's a way to just turn off "offline experience" in Chrome before service workers become widespread. The browser is a dangerous place to execute software. I suppose native apps can be just as insecure, but I worry about an ecosystem with many neighbors (web apps, extensions, plugins, add-ons, etc) running in the same process. It's much more likely one web app can steal my passwords from LastPass, for example, without knowing it. The chain of vectors just feels really dangerous.

I think the line:

> It's a JavaScript Worker, so it can't access the DOM directly. Instead, a service worker can communicate with the pages it controls by responding to messages sent via the postMessage interface, and those pages can manipulate the DOM if needed.

gives me some hope that it's intended to be isolated from all other tabs (including the one that started it!).

There will be bugs, of course, but it looks to me like they're starting from a secure position, which gives me hope.

Sandboxing means that plugins are not running in the same process at all.

I don't think those things run in the same process in Chrome, and Firefox is moving towards a multiprocess model as well.

Can someone explain why this was downvoted?

From my understanding, the service worker doesn't live indefinitely in the background - it reacts to events. So without additional permission prompts, it will only react when you load a page that's within the service worker scope. So I'm not sure what additional security concerns there are.

Oops. I thought this might be an attempt at rapprochement with their janitorial staffs. Perhaps the wrong nomenclature for our new era.

Regarding HTTPs: there is https://test.bi

Isn't this an egregious violation of most of the TOS / agreements for paid SSL certs?

I haven't read the specs, can you spawn multiple service workers on a single page ?

If yes, what are the security measures to stop a rogue site to blow up your browser ?

Only one service worker per scope, the newest kicks out any older ones.

How can I protect my browser against these things?

this seems unnecessary as you need a html file and browser to bootstrap anyway. There is more to it though, it can continue to run after the browser has exited and intercept push mesages.

Reply on news.ycombinator.com