1878716 - Something screwy with window.open and canvas randomization

Status: RESOLVED FIXED

(Core :: Privacy: Anti-Tracking, defect, P2)

Description

[Tom Ritter [:tjr]]

If I open https://canvasblocker.kkapsner.de/test/test.html in two browsers, with FPP enabled in both - the window.open values will be the same between browsers. Which shouldn't be happening because randomization.

A different issue is reported in https://github.com/arkenfox/user.js/issues/1716#issuecomment-1925180783 which shows that window.open is getting the same hash as a window without FPP enabled - however I could not reproduce that.

Maybe (hopefully?) these are symptoms of the same issue and fixing one will fix both...

Comment 1

[Tom Ritter [:tjr]]

So far I've traced this to the fact that GenerateCanvasKeyFromImageData fails because aCookieJarSettings->GetFingerprintingRandomizationKey fails. There's aCookieJarSettings (it's not null), but we never called SetFingerprintingRandomizationKey on it, so it has no key.

Comment 2

[Tom Ritter [:tjr]]

We are taking this branch because we don't have an inprocessparent for this popup it seems.

Comment 3

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Copy the randomization key from the opener r?timhuang

Comment 4

[Tom Ritter [:tjr]]

I confirmed that the attached patch fixes it, but I'll need feedback from Tim next week. This is basically the same problem we had with null-principal documents, opened windows, the RFP overrides.... just now applied to the randomization key.

It seems like our two options are to either copy from the opener window into the opened window (but this would probably be inconsistent with a bunch of other stuff not-RFP related that we don't copy, like the rest of CJS) or we generate a new randomization key in the else block, so the opened window has a different randomization key than the opener.

Comment 5

[Tom Ritter [:tjr]]

For the record, the opened window is in the same process.

Comment 6

[Tim Huang[:timhuang]]

It makes more sense if we copy the random key from the opener window. The random key is decided when we open the channel for the document in the parent process. However, a null principal document could have never gone through the parent process. So, we should copy the key from the opener window.

Comment 7

[Tom Ritter [:tjr]]

Coming back to this, my patch seems to been screwed up - I can't remember or tell if my new block was inside if (inProcessParent) or after it. But it appears that it only works if it was after it, so I guess that's where it must have been.

It seems like we could copy a lot of the same properties we copy from inProcessParent from the opener. This patch only copies the CanvasRandomizationKey. But there's also CookieBehavior, PartitionKey, IsFirstPartyIsolated, IsOnContentBlockingAllowList, and ShouldResistFingerprinting. I think I'm going to need to look at this more because I have a feeling we're going to want to copy more of these...

Comment 8

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: De-duplicate canvas comparison code in tests r?timhuang

Comment 9

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Fix some Hardware Concurrency Tests r?timhuang

It turns out that we had the override string wrong, but we
didn't notice because we happened to be only testing things
where it was supposed to be disabled anyway.

There was a test where this would have not been the case
(simplePBMFPPTest) - but this was only be tested when we were
exempting it with ETP.

I added this scenario to a place where it would be tested
positively, and confirmed that it failed, then when I fixed
the override, started working again.

Comment 10

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add those tests to all the other files r?timhuang

Comment 11

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test for Canvas Randomization r?timhuang

This test ensures the expected number of pixels are randomized
in different scenarios of preferences

Comment 12

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test for popups r?timhuang

Comment 13

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases between a parent and child r?timhuang

For the first test, the parent is a page and the child is an iframe, but
eventually we will have tests for lots of different children types.

We test that the number of pixels we expect to be randomized are, and
we also compare the differences between the parent and child.

This allows us to ensure that the FPP key is being propagated correctly.

Comment 14

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in a popup that is same or cross domain r?timhuang

Comment 15

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Populate the key in another scenario that occurs for about:blank popups under certain conditions r?timhuang

Comment 16

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in an about:blank popup r?timhuang

Comment 17

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Rename a shared test to hopefully be less confusing r?timhuang

This test is a little odd because (a) its name is super long and unhelpful
(b) it's testing a very important sitation we've regressed on before
(c) it disables the protections, so you actually don't expect anything
to be enabled.

We're going to need a very similar version of this test where the
protections are enabled, so let's pre-emptively rename this one to
prepare for that.

Comment 18

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a version of the test with protections enabled r?timhuang

Comment 19

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in a blob popup r?timhuang

Comment 20

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in a data popup r?timhuang

Comment 21

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in an about:blank iframe r?timhuang

Comment 22

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in a blob iframe r?timhuang

Comment 23

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that confirms the expected default behavior in PBM r?timhuang

Comment 24

[Tom Ritter [:tjr]]

Attachment: Bug 1878716: Add a test that compares the canvases in a data iframe r?timhuang

Comment 26

[Tom Ritter [:tjr]]

Uhg, got several test failures on try from the last test run i will need to dig into

Comment 27

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/ed5d79d4629f
Copy the randomization key from the opener r=timhuang
https://hg.mozilla.org/integration/autoland/rev/8c994e63a5df
De-duplicate canvas comparison code in tests r=timhuang
https://hg.mozilla.org/integration/autoland/rev/734befa016cd
Fix some Hardware Concurrency Tests r=timhuang
https://hg.mozilla.org/integration/autoland/rev/a236ed107799
Add those tests to all the other files r=timhuang
https://hg.mozilla.org/integration/autoland/rev/7f4801b8afbd
Rename a shared test to hopefully be less confusing r=timhuang
https://hg.mozilla.org/integration/autoland/rev/f628d8a1eb58
Add a version of the test with protections enabled r=timhuang
https://hg.mozilla.org/integration/autoland/rev/5f0a42f88d73
Add a test for Canvas Randomization r=timhuang
https://hg.mozilla.org/integration/autoland/rev/c503aad0ca16
Add a test for popups r=timhuang
https://hg.mozilla.org/integration/autoland/rev/f87e9a61ca51
Add a test that compares the canvases between a parent and child r=timhuang
https://hg.mozilla.org/integration/autoland/rev/3978af2aa2fd
Add a test that compares the canvases in a popup that is same or cross domain r=timhuang
https://hg.mozilla.org/integration/autoland/rev/9b0ee8d0d4f6
Populate the key in another scenario that occurs for about:blank popups under certain conditions r=timhuang,necko-reviewers,jesup
https://hg.mozilla.org/integration/autoland/rev/6f3394ea8888
Add a test that compares the canvases in an about:blank popup r=timhuang
https://hg.mozilla.org/integration/autoland/rev/ac1a6456b77e
Add a test that compares the canvases in a blob popup r=timhuang
https://hg.mozilla.org/integration/autoland/rev/fceb109159c2
Add a test that compares the canvases in a data popup r=timhuang
https://hg.mozilla.org/integration/autoland/rev/99bdce781621
Add a test that compares the canvases in an about:blank iframe r=timhuang
https://hg.mozilla.org/integration/autoland/rev/cba20f6a8f60
Add a test that compares the canvases in a blob iframe r=timhuang
https://hg.mozilla.org/integration/autoland/rev/cef71eec96a0
Add a test that confirms the expected default behavior in PBM r=timhuang
https://hg.mozilla.org/integration/autoland/rev/76724efd86c9
Add a test that compares the canvases in a data iframe r=timhuang

Comment 28

[Cosmin Sabou [:CosminS]]

https://hg.mozilla.org/mozilla-central/rev/ed5d79d4629f
https://hg.mozilla.org/mozilla-central/rev/8c994e63a5df
https://hg.mozilla.org/mozilla-central/rev/734befa016cd
https://hg.mozilla.org/mozilla-central/rev/a236ed107799
https://hg.mozilla.org/mozilla-central/rev/7f4801b8afbd
https://hg.mozilla.org/mozilla-central/rev/f628d8a1eb58
https://hg.mozilla.org/mozilla-central/rev/5f0a42f88d73
https://hg.mozilla.org/mozilla-central/rev/c503aad0ca16
https://hg.mozilla.org/mozilla-central/rev/f87e9a61ca51
https://hg.mozilla.org/mozilla-central/rev/3978af2aa2fd
https://hg.mozilla.org/mozilla-central/rev/9b0ee8d0d4f6
https://hg.mozilla.org/mozilla-central/rev/6f3394ea8888
https://hg.mozilla.org/mozilla-central/rev/ac1a6456b77e
https://hg.mozilla.org/mozilla-central/rev/fceb109159c2
https://hg.mozilla.org/mozilla-central/rev/99bdce781621
https://hg.mozilla.org/mozilla-central/rev/cba20f6a8f60
https://hg.mozilla.org/mozilla-central/rev/cef71eec96a0
https://hg.mozilla.org/mozilla-central/rev/76724efd86c9