Intent to Ship: Fenced Frames - Functionality Updates

[Liam Brady]

Contact emails

shiva...@chromium.org, d...@chromium.org, jka...@chromium.org, lbr...@google.com 

Explainer(s)

Send Automatic Beacons Once

https://github.com/WICG/turtledove/pull/718 

Serializable Fenced Frames Configs - Minor Change, No explainer available.

Note: With this change, FencedFrameConfig objects will be serializable and can be sent through "postMessage()" and other similar calls. 

Creative Macros in Fenced Frames Ads Reporting (FFAR)

https://github.com/WICG/turtledove/pull/763

Spec(s)

Send Automatic Beacons Once

https://github.com/WICG/fenced-frame/pull/109 

Serializable Fenced Frames Configs

https://github.com/WICG/fenced-frame/pull/111 

Creative Macros in Fenced Frames Ads Reporting (FFAR)

Protected Audience: https://github.com/WICG/turtledove/pull/762/files 

Fenced Frames: https://github.com/WICG/fenced-frame/pull/113

 

Summary

We launched Fenced Frames as a part of Chrome 115. We would like to add the following three functionalities to Fenced Frames.

1. Send Automatic Beacons Once

A common feature in ad frames is the "why this ad?" link. Since that link is separate from the ad itself, clicking "why this ad?", and its subsequent navigation, should be considered different from clicking directly on the advertisement itself. With our current automatic beacon design, however, once an automatic beacon is set (usually in the click handler for the ad link), that beacon will send out for any subsequent top-level navigations, including if "why this ad?" is clicked. This can result in erroneous impressions being sent out.

As a solution, this feature introduces a new "once" member to the FenceEvent dictionary passed into "window.fence.setReportEventDataForAutomaticBeacons()". If set to true, the saved automatic beacon data will be cleared out after the next beacon is sent, ensuring that automatic beacons with that data are only sent once. This means that further clicks to non-ad parts of the frame that result in top-level navigations will not send out erroneous beacons.

(This feature already shipped in M116. That was our mistake. At the time, we thought we would only need a PSA for it, and shipped it without it being behind a flag. We apologize for the mistake.)

2. Serializable FencedFramesConfigs

With this change, FencedFrameConfig objects will be serializable and can be sent through "postMessage()" and other similar calls. Serialization allows for a case where the frame that runs an ad auction is not the same frame that ends up embedding the winning ad in a fenced frame. FencedFrameConfigs cannot be serialized to storage, nor can they be sent in a message that crosses a fenced frame boundary. A FencedFrameConfig object is only valid in the traversable navigable it was originally created in, and, if sent outside to a different context, will not be able to navigate, since the new traversable navigable's fenced frame config mapping will not contain the internal config needed to do the navigation.

3. Creative macros in Fenced Frames Ads Reporting (FFAR)

This feature extends the Fenced Frame Ads Reporting (FFAR) API to support macro substitution in reporting URLs and allows reports to be sent to up to ten other origins that have enrolled with the Privacy Sandbox and allow-listed by the DSP. Use case: In online ad auctions for ad space, advertisers buying through DSPs in several situations use other adtech providers to monitor performance and keep track of how their advertising dollars are spent. (issue link)

Blink component

Blink>FencedFrames

TAG reviews and status

Fenced frames existing TAG review appended with these spec changes https://github.com/w3ctag/design-reviews/issues/838#issuecomment-1693631006

Link to Origin Trial feedback summary

No Origin Trial performed

Is this feature supported on all six Blink platforms (Windows, Mac, Linux, Chrome OS, Android, and Android WebView)?

Supported on all the above platforms except Android WebView.

Debuggability

Additional debugging capabilities are not necessary for these feature changes.

Risks

Compatibility

There are no compatibility risks, as described below:

1. Send Automatic Beacons Once: This is backward compatible with the existing API since the default value of “once” is false which is the same behavior as the previous behavior.

2. Serializable FencedFramesConfigs: This is added functionality and backward compatible with the existing FencedFramesConfig.

3. Creative macros in Fenced Frames Ads Reporting (FFAR): This is adding a new API and a backward compatible change to reportEvent.

Interoperability

there are no interoperability risks as no other browsers have decided to implement these features yet.

Is this feature fully tested by web-platform-tests? Link to test suite results from wpt.fyi.

Yes

Tests: https://github.com/web-platform-tests/wpt/tree/master/fenced-frame

Results: https://wpt.fyi/results/fenced-frame?label=experimental&label=master&aligned

Specifically, these features correspond to the following tests:

Send Automatic Beacons Once:

• automatic-beacon-two-events-clear.https.html (test) (result)

• automatic-beacon-two-events-persist.https.html (test) (result)

Serializable FencedFrameConfigs:

• deep-copy-config.https.html (test) (result)

Creative macros in Fenced Frames Ads Reporting (FFAR):

• fence-report-event-destination-url.https.html (test) (result)

Anticipated spec changes

None 

Link to entry on the Chrome Platform Status

https://chromestatus.com/feature/5103970808233984 

Links to previous Intent discussions

Intent to prototype: https://groups.google.com/a/chromium.org/g/blink-dev/c/Ko9UXQYPgUE/m/URRsB-qvAAAJ 

Intent to experiment:

https://groups.google.com/a/chromium.org/g/blink-dev/c/y6G3cvKXjlg/m/Lcpmpi_LAgAJ 

Intent to extend origin trial:

https://groups.google.com/a/chromium.org/g/blink-dev/c/SD8Ot2gpz4g/m/A9uA-_cGAwAJ 

https://groups.google.com/a/chromium.org/g/blink-dev/c/gpmaOi3of_w/m/SyMclFhMAAAJ 

https://groups.google.com/a/chromium.org/g/blink-dev/c/CBrV-2DrYFI/m/RTojC6kHAgAJ 

Intent to ship:

https://groups.google.com/a/chromium.org/g/blink-dev/c/tpw8wW0VenQ/m/mePLTiHlDQAJ

[Daniel Vogelheim]

Hi Liam,

This intent has come up in the OWP security triage, and I'm trying to figure out whether there's XSS potential in the 3rd sub-feature, "Creative macros in FFAR". This looks like a string-based pattern replacement where the result string will then be parsed by the browser. Similar things have lent themselves to XSS, e.g. when a string value contains meta characters that code downstream will then parse in unexpected ways. Unfortunately, I can't seem to find enough information about how exactly the replacement & subsequent usage works to make up my mind whether there's a concern or not. Could you help me out with a bit more information?

What I've found is this:
- In "<dfn>send a beacon</dfn>": "TODO: Substitute macros from |macro map| into |destination url|." (source)
- In "<dfn>asynchronously finish reporting</dfn>" I find where macro map being set, but then it says: "TODO: Pass |macroMap| and |allowedReportingOrigins| to [=Finalize a reporting destination=] when it is updated to take the parameters." (source) I can't find that usage.

Questions I have:
- Am I reading the right docs? Where else should I look?

- Is this meant as a simple string-based substitution? Is there any filtering of allowed characters, or so?

- What happens with the result values?
- Is the entity that sets the macro values always the same that has set the patterns the values are being used for, or could those be different entities?

Thanks!

--
You received this message because you are subscribed to the Google Groups "blink-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to blink-dev+...@chromium.org.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/62771371-5dbb-4d02-a232-a99ded5b293fn%40chromium.org.

[Liam Brady]

Note that these features are targeted at M117. Will defer to gtanzer@ to answer the FFAR questions.

[Garrett Tanzer]

Hi Daniel,

• There are a few relevant call sites in the overall reporting flow:
  • Declare allowlist of reporting destination origins
    • This happens in navigator.joinAdInterestGroup(), by an ad auction buyer
  • Declare macros (key:value correspondences)
    • This happens in the buyer reporting worklet, by the same ad auction buyer that declared the reporting destination origins
  • Perform report to custom url
    • This happens under the auction's winning ad creative's origin, which isn't necessarily the same as the ad auction buyer, but it is chosen by the ad auction buyer
• Here is the sequence of events:

1. 1. The browser ingests and validates the allowlist when the interest group is declared.
   2. The browser ingests the macro key:value mapping when the auction happens. The key/value strings have no structure that is validated. The browser adds "${" and "}" around the user-provided keys in the macro mapping.
   3. In the ad that results from the auction:
      1. The ad sends a URL to the browser, including macros like ${KEY}. The URL has to be a valid HTTPS url even with the macros unsubstituted. (See impl: https://source.chromium.org/chromium/chromium/src/+/main:content/browser/renderer_host/render_frame_host_impl.cc;l=8644?q=content%2Fbrowser%2Frenderer_host%2Frender_frame_host_impl.cc%20-f:out)
      2. The browser performs a simple string substitution to replace the keys with the values. The implementation is reused from navigator.deprecatedReplaceInURN, another part of Protected Audience. It doesn't substitute macros recursively, so you can't get an infinite loop. deprecatedReplaceInURN has apparently not been spec'd, which is was unable to quickly reuse that spec and left it as a TODO for now.
      3. The origin of the resulting URL is checked against the allowlist. If it doesn't match any of them, no action is performed. If the URL is invalid, this will create an opaque origin and therefore always fail the check. (See impl: https://source.chromium.org/chromium/chromium/src/+/main:content/browser/fenced_frame/fenced_frame_reporter.cc;l=454?q=fenced_frame_reporter.cc%20-f:out) We may add an explicit URL valid/HTTPS check after substitution for neatness/to be robust to any future changes to the allowlist checking.
      4. If the URL does pass the check, a GET fetch will be performed on it (the result is unused). There's no other parsing/usage of the URL other than performing a fetch.

So in summary, we were not very concerned about XSS because the entity choosing the macro pairings is the one we want to protect, and the entity providing the base URLs is semi-trusted. The less trusted entity is the site of the destination URL, which just receives a GET and doesn't get to control anything. Since the URL has to be valid both before and after the macro substitution, this limits how wacky the substitutions can get even if you don't trust one of those entities. The weirdest stuff you could do would be like "FOO}&key=value1&${BAR" -> "key=value2", so that it becomes "${FOO}&key=value1&${BAR}" -> "key=value2". We can add a check to exclude $,{,} characters in the macro key so even this isn't possible.

Hope this answers your questions,

Garrett

On Monday, August 28, 2023 at 9:35:34 AM UTC-4 Daniel Vogelheim wrote:

[Garrett Tanzer]

After some discussion offline, we're going to sanitize the macro keys and values with EscapeQueryParamValue so that macro substitution always stays within the original query parameter field. This should prevent XSS-y substitutions while keeping the API surface exactly the same for regular use.

This fix will be implemented in M118, and we've decided that it need not block launch in M117 due to the low severity (especially because the existing reportEvent surface grants the buyer reporting worklet complete control over the destination URL, so the ad creative is already known to trust the buyer wrt where its reports get sent).

[Mike Taylor]

I also had an offline discussion with Daniel to confirm shipping as-is in M117, and sanitizing in 118 is acceptable from a security POV (trust but verify, etc).

LGTM1 to ship.

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/2b14ffdf-1fdb-4fa7-9054-65356c007b50n%40chromium.org.

[Yoav Weiss]

LGTM2

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/04a15088-83e8-4ffc-8a6f-8b255c74e585%40chromium.org.

[Chris Harrelson]

LGTM3

To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAL5BFfU55BdzLaOxRho0DiWRF5pSTGLXQfj8Kig437exZO-o5A%40mail.gmail.com.

[Garrett Tanzer]

An update on this:

The extra input sanitization was merged for M119 behind a feature flag for M119 fenced frame changes. It rejects macros at declaration time whose keys/values use characters that are disallowed in URI components (e.g. &, ?, $, {, } that were most relevant here).

https://chromium-review.googlesource.com/c/chromium/src/+/4842629

This should prevent escaping out of URL parameters without breaking reasonable code from M117, and keeps behavior parallel to other APIs around Protected Audience like deprecatedReplaceInURN. (We discussed adding the same sanitization there, but the owners did not want to risk compat issues since the API has been around for longer.)