Comment by commoner | Hacker News Reader

Comment by commoner | Hacker News Reader

Through the Tor Uplift Project,[1] Tor Browser's Fingerprinting Protection feature is now available in Firefox on desktop and Android. The feature makes Firefox hide some pieces of identifying information from the sites you interact with, such as your timezone (which is set to UTC), some of your fonts, your keyboard layout/language, and parts of your user agent (for example, your browser version is set to the latest ESR version).[2]

To enable Fingerprinting Protection in Firefox, go to about:config and set privacy.resistFingerprinting to true.

Some Firefox forks enable Fingerprinting Protection by default, including LibreWolf[3] (desktop) and Mull[4] (Android). If you are on Android, the release version of Firefox does not include access to about:config, and you'll need to either switch to Firefox Beta/Nightly or use a fork like Mull, Fennec F-Droid, or Iceraven to take advantage of this feature.

[1] https://wiki.mozilla.org/Security/Tor_Uplift

[2] https://support.mozilla.org/en-US/kb/firefox-protection-agai...

[3] https://librewolf.net

[4] https://f-droid.org/en/packages/us.spotco.fennec_dos/

I just altered that setting and now Firefox resets my zoom level for every page I read to 100%. Makes HN unreadable as my default zoom for this site is 170% and having to set it every page I visit becomes old very quickly!

Interesting side effect though.

Yes, disabling site-specific zoom is one of the things that Resist Fingerprinting does.[1] If you want to use site-specific zoom while keeping Resist Fingerprinting enabled, the Zoom Page WE add-on[2] should allow this. I've just tried it and it worked for me.

[1] https://bugzilla.mozilla.org/show_bug.cgi?id=1369357

[2] https://addons.mozilla.org/en-US/firefox/addon/zoom-page-we

Why does a website need to know my zoom level, i.e., why is this information even made available?

It’s been a while but IIRC as long as there’s JS it can be gotten implicitly by probing and comparing dimensions of elements and viewport on the page.

We could question if that’s really necessary as well but the ship has kind of sailed on that one.

IMHO, even when sand-boxed, allowing a fully Turing-complete language with such a vast selection of available APIs to run on page load per default is what kills privacy.

People should be trained to allow script execution only when they trust the site, and there should be levels: Zero, Fully Isolated, Trusted.

OK now time to wait for someone to tell me this will be too much to ask from users. It wouldn't be an invalid point either, we can't even train people to have some common sense when in control of tons of steel going fast loaded with highly flammable liquids... So, there's that.

I don't know.

It's not even per page. I don't care if I trust that page, I don't want any FB scripts to run. There are so many external libraries included (loaded from a CDN they do not control) that I don't trust any developer to know with 100% certainty what their app includes.

Maybe something like randomizing the variables in a realistic way across many sessions where the real session is controlled and viewed directly by the user.

I don't think it would work when there's a state across views or server though, but maybe that's something you avoid when using Tor anyway?

My hunch is that it’s not explicitly provided to the website by the browser, but there are ways to measure what the zoom level is using JavaScript on the page.

One could use element.getBoundingClientRect and similar APIs to measure what size certain elements are rendered at and compare that with their default size for instance.

The resulting zoom level can then be used as a signal for fingerprinting.

To render content based on device screen scaling. window.devicePixelRatio provides page zoom level combined with OS DPI setting. Things like canvas do not use scaling by default, so to render content appropriately you need to know by how much you have to scale it. Other use cases exit too.

I find that question valid for almost everything that's used for fingerprinting. Eg. Why is timezone available by default?

I wrote more but I sound like old man shouts at cloud. I've got ad and JavaScript blockers, but so much of the web is created sitting upon a mesh of invasive bullshit that it breaks easily.

Shakes fist at cloud anyway.

so that dates displayed can be correct by default, without having the user select their timezone from a dropdown with a million different options

Why not offer the "dropdown" at a layer the browser manipulates?

IE, web component where you can set a format string and then the browser renders it substituting the info it has?

Then sites that need to know the date can ask for it.

There's a lot we could do by splitting stuff into permissions and some sort of standard templates.

and then you can read it out to indirectly get the time zone anyway (albeit a less specific version of it, since you wouldn't have the DST info). also, sometimes you need to actually have the users time zone for things like e.g. scheduled blog posts, so that they will show up when the user expects them to, not their scheduled time in UTC or whatever.

I use a site that only does UTC, and time zone related complaints are the most common issue asked about on the forums

> IE, web component where you can set a format string and then the browser renders it substituting the info it has?

you can easily detect this

    - please display this localdatetime as a string
    - read it back and parse it
    - are they (almost) exactly multiples of 60 minutes apart?
    - if yes it's most probably your timezone

Do apps really need to read back strings they omitted?

they don't, but they can and by doing it they can check if the date has been changed by the browser because of time zone shift.

Yeah but why do you have to tell the server? The browser could display the date and time locally.

because websites are sometimes interactive and allow the user to schedule things to happen at certain times.

And displaying time/date locally would leak that information anyway if you wanted to do it in a way that works in various contexts it would need to in a website (e.g. canvas based apps)

even if you just let the user stylize the font of the date (which you clearly would need to), you tell your magic date input to only show the current hour, then use a font that has a certain width for each number, allowing you to then based on the width of that element figure out the hour, same for other things, obviously. It's easy to imagine some thing like that without thinking about all the details, but it's not really feasible once you think about how this would be implemented and how it could be circumvented. And that's in addition to not working in contexts where you schedule a blog post, zoom meeting, or whatever else might require the server to account for user time zone

This is the problem with the modern web, which has become an app distribution platform. When you treat the browser as an OS, you need to expose a lot of information for stuff to work.

It would be very interesting to develop a modern web based purely on declarative content (modern HTML/CSS). HTMX is an interesting take on this, although it's currently implemented as server-provided JS: i don't see a reason why such patterns couldn't be implemented by the browser itself.

> It would be very interesting to develop a modern web based purely on declarative content (modern HTML/CSS).

For sure. I think some scripting could also potentially be implementable without massive fingerprinting / privacy implications. E.g. pure compute scripts, form validation, etc. that has no practical way to smuggle any data out of your browser. Anything that sends a request would have to be statically derived (or explicit user input as into form).

Functional data pipelines without side-effects could do the trick indeed. It would also make it a lot easier to debug for performance issues, and the browser could be more clever about optimizations: for example if you've got a loop changing DOM elements, maybe you could wait for the loop to finish before starting a re-render... something that's impossible to do with JS-based rendering where global page state may change under your feet at any given time.

EDIT: Just for the sake of mentioning, simple/obvious computations for interactivity was the promise of GNU's libreJS project. I'm unaware of the current state of it, though.

> It would be very interesting to develop a modern web based purely on declarative content

Is not that the subset of the web that would work when javascript is disabled? Some already develop it in that direction - what is not declarative shall be unnecessary. Or are you suggesting something different?

The problem is the declarativeness of the web is very much limited for UI/UX purposes. There's been good steps taken with HTML5, although dropping XML-compliance was in my view a major mistake in terms of operability/simplicity.

I don't understand why we need to have dozens of CSS frameworks for "components" that have become common practice across the ecosystem. Pagination, "Hero" elements, intra-page tabs, breadcrumbs (and many others) should be HTML standard so that it's more accessible and users can come up with their own stylesheets. The breadcrumbs for example would enable your browser UI to show a "go up" button like your file browser does. Another interesting example would be element filtering: why can't a <form> with a local action property (like "#data") be used to filter a list of elements without JS?

As long as most UI of a page is dictated by dozens of piled-upon CSS hacks, user stylesheets will remain a wild dream. But given how little variety there is on the web these days, many things could be standardized part of the HTML spec so that CSS is only needed for customization (eg. colors, spacing) on simpler pages, while retaining the possibility for the server to suggest more complex CSS UIs as we currently do if you absolutely want to do that.

But why zoom level? Set that locally after the site sends the info.

> It would be very interesting to develop a modern web based purely on declarative content.

There is something like that, the Gemini protocol.

If you're talking about the gemtext format, unfortunately beyond titles, list, blockquotes, preformatted texts, and links, nothing else has been standardized.

I understand the appeal of simplicity but if you ask me that's a huge step backwards compared to HTML5. No <form>, no <section>/<article... It's like markdown but with another syntax :-/

i would very much like for htmx to not have to exist and just have the functionality subsumed into the HTML spec

it wouldn't be much work

Hey thanks for taking the time to reply. Have you maybe got in touch with hacker-friendly browsers such as nyxt? There may be some interested people over there.

Also, is there some good venues to discuss the semantic/declarative web with you htmx folks and hopefully people from other like-minded projects? IRC? XMPP? Matrix?

Sorry for the delayed reponse: nope, never talked w/ the nyxt folks. I tried to post a topic on the working group thingie but they, understandably, weren't very receptive.

We use discord for chat right now:

https://htmx.org/discord

Well the good thing about nyxt is it's super extensible so a PoC doesn't require proper "reception" on their side.

Do you maybe have a gateway/bridge to a libre network such as IRC/XMPP/Matrix? I find HTMX pretty interesting but i wouldn't touch discord with a 10-foot pole, if only because my limited computing resources won't allow for such a resource-hungry app to run in the background.

It seems like matterbridge supports discord backend but i don't have a discord account to try it with. If you're not willing to host matterbridge, i'm already hosting one and i would just need credentials to try and connect it to Discord. If you're willing to give that a try, feel free to mail me at my username @ thunix.net.

https://github.com/42wim/matterbridge

This is something that's confused me about Tor Browser. How helpful could zoom level possibly be towards fingerprinting a user to a degree where it's of any use for targeted advertising and the like?

Seeing that a user has a site's zoom set to 90% seems to be close to worthless in terms of narrowing down what cohort they're in, let alone identifying them individually. What am I missing here?

> Seeing that a user has a site's zoom set to 90% seems to be close to worthless in terms of narrowing down what cohort they're in, let alone identifying them individually.

How so? 90% is by far not the normal zoom level people browser with, so it is a perfectly valid data point to use for fingerprinting. Every single bit of data they can get makes your whole fingerprint more unique.

It adds a little bit of information that might help identify you. It may be less than a bit (e.g. if it's set to 100%), but it's still something.

You can see the informational content of various fingerprints here: https://coveryourtracks.eff.org/

Another thing that has less than a bit of content is having cookies enabled. Nearly everyone has cookies enabled, for better or worse, so having them on doesn't add much to a fingerprint. Having them off adds far more. But both add something.

> How helpful could zoom level possibly be towards fingerprinting a user

It's just another bit of information. Collect enough bits and eventually you'll have a likely-unique id. It doesn't matter what that information is as long as it somehow is about you (and not e.g. random). If you want to fingerprint, you just try to grab every bit of information you can, no matter how irrelevant it is taken on its own.

It doesn't need to say anything about a cohort to be useful, it just needs to enable identifying so that they can track you around and eventually combine other information they discover about you in a profile. And it doesn't even need to be 100% accurate; "that person coming from a Telia-owned IP visiting this site again at 2 AM GMT+2 using Firefox Nightly on Linux, with 1440p display and 120% zoom level and no fonts installed" could be two or three guys if I'm lucky but it's probably close enough to not matter for someone who just wants to sell me garbage.

If your zoom level were the only thing, then indeed it would be useless. Problem is, browsers leak lots of bits. It's better to try plug all leaks you can than it is to ask whether that particular leak alone is harmful enough.

Ah when you put it like that, I see the logic.

> close to worthless ... What am I missing here?

Differential calculus. One negligibility times by a huge amount equals one discreet amount. One bit here, one bit there, you get a fingerprint of a thoundred bits.

The UTC cloaking has a lot of downsides. Websites displaying false times, but not reliably, as they are mostly not indicating the timezone, and you never know if they are displaying local time of the site's location or try to get clever and use the visitor's time. No privacy win if you are browsing on locally relevant web websites, because your timezone is implied anyway. e.g. most of Europe has the same one. All of this is justifiable for TOR use, but it is the single most annoying problem making it too inconvenient for regular use.

Am utterly dumb downside is that the browser history - in your own local History window - has times shown in UTC.

There's also SmartCookieWeb Preview:

https://github.com/CookieJarApps/SmartCookieWeb-preview

With all the things that end up being surprising settings that fingerprint me, why is there not an “absolutely no auto-playing video” setting?

Reply on news.ycombinator.com