On Mon, Apr 8, 2024 at 1:20 AM Yoav Weiss (@Shopify) <[email protected]> wrote:
> I'm not sure same-origin isolation won't end up being a desired feature > on its own. I heard developers asking for stronger isolation primitives on > more than one occasion. I'll talk to folks and think about it some more. > To be clear, this would not be secure isolation between two parts of the same origin. The documents would still share the same cookies, localStorage, permissions, etc. It would only allow them to run in different processes from each other. That's useful for restricting what else runs in the same process (e.g., to limit what can be leaked via Spectre + SABs), but it is not a new isolation primitive for data within an origin. That sort of primitive is generally discouraged, per the 2008 Beware of Finer-Grained Origins <https://www.adambarth.com/papers/2008/jackson-barth-b.pdf> paper. Charlie On Thu, Apr 11, 2024 at 8:49 AM Camille Lamy <[email protected]> wrote: > > > On Monday, April 8, 2024 at 6:12:23 PM UTC+2 [email protected] wrote: > > There is a huge demand for protecting data that's shared with users Any > help in strong binding data to origin and blocking sharing would a big win. > > > I believe that is a different problem from the one we're trying to solve > with this API. Would you happen to have more details about the issue you're > referring to? > > > > thx ..Tom (mobile) > > On Mon, Apr 8, 2024, 1:20 AM Yoav Weiss (@Shopify) <[email protected]> > wrote: > > This is very interesting! > > Do I understand correctly and the main reason this would be easier to > deploy is because embedded iframes and popups won't need to deploy COEP in > this model? > > > Yes. They also wouldn't need to deploy COOP, and so would be able to > interact with cross-origin popups. > > > > On Fri, Apr 5, 2024 at 12:14 PM Camille Lamy <[email protected]> wrote: > > Yes the user agent keying is deterministic, and we're adding reporting to > warn developers if they end up having two same origin documents that could > normally have DOM access but can't due to Document-Isolation-Policy. > > > I'm not sure same-origin isolation won't end up being a desired feature > on its own. I heard developers asking for stronger isolation primitives on > more than one occasion. I'll talk to folks and think about it some more. > > > Our recommendation would be to adopt the header on all documents of an > origin, which removes the concerns around script access. As a followup, we > might resurrect the Origin-Policy work to help with this issue. > > > Origin-Policy would definitely help avoid mistakes here.. > > > > For COOP and COEP, you're correct to note that they are not available due > to the platform limitations on Android WebView. Because of this, the > crossOriginIsolated spec already has a notion of crossOriginIsolation being > either logical (ie no API access) or effective (ie API access). We're > building on this existing notion. > > In terms of platform support, our goal is to first release on desktop, in > order to finally end the ungated SAB reverse Origin Trial. Then we'll > extend to Android (but not Android WebView). For Android, the situation is > a bit different from full Site Isolation, because here the isolation and > resulting increase memory consumption is driven by the website as opposed > to the platform. We might not implement full functionality on low-end > Android, but then none of the developers interested in the API want to have > it run on low-end Android. Basically, this gives access to > SharedArrayBuffers, which are mostly useful to cut calculation time in > heavy web apps, that wouldn't run on devices with limited hardware. > > Hope that helps! > Camille > > On Thursday, April 4, 2024 at 9:45:04 PM UTC+2 Charlie Reis wrote: > > I seem to recall that Android Chrome is also limited here, but maybe that > has changed and my knowledge is outdated. > > > Correct, we don't usually create out-of-process iframes on Android Chrome > if the device has less than 2G of RAM. Otherwise we allow it (e.g., for > partial > Site Isolation > <https://chromium.googlesource.com/chromium/src/+/main/docs/process_model_and_site_isolation.md#Partial-Site-Isolation>). > I'm not sure if COOP+COEP has any restrictions on low-end Android devices, > since that mode requires multiple processes but not out-of-process > iframes. For Document-Isolation-Policy, I believe there's some notes about > low-end Android devices in the explainer, maybe suggesting that it's less > needed on such devices? I'll let Camille clarify. > > Charlie > > On Thu, Apr 4, 2024 at 12:27 PM Vladimir Levin <[email protected]> > wrote: > > > > On Thu, Apr 4, 2024 at 12:11 PM Charlie Reis <[email protected]> wrote: > > My understanding is that at least this behavior is deterministic, right? > That is, either the same-origin frames will be able to script each other or > they won't and this will happen consistently (based on the agent cluster > key). > > > Yes, I think it would be deterministic based on the headers, so hopefully > education via error messages would help. > > An observation I had is that it seems that the Document-Isolation-Policy > is still at the mercy of the platform having the resources to > process-isolate frames. > > > Camille can probably confirm the details, but I believe that's right. > COOP+COEP depends on the platform being able to open a new window in a > different process, which I think all platforms but Android WebView can > support at this point (?). Document-Isolation-Policy would depend on > out-of-process iframes, which wouldn't work on Android WebView or iOS, at > least for the time being. On platforms that do support out-of-process > iframes, it would make crossOriginIsolated modes much easier to adopt, > though. > > > I seem to recall that Android Chrome is also limited here, but maybe that > has changed and my knowledge is outdated. > > > > Also I'm not sure if it would be possible for 3p iframes to starve > platform of such resources so that the top level frame would no longer be > able to create 1p frames that have access to COI-gated APIs > > > IIUC, I think each origin is limited in the number of processes it could > create in a given page (basically one with SAB access and one without), > which helps. > > > Ah that makes sense. There may still be some possibility with just > spamming 3p iframes but that likely exists today anyway > > Thanks! > Vlad > > > > Charlie > > On Thu, Apr 4, 2024 at 8:05 AM Vladimir Levin <[email protected]> wrote: > > This does sound a bit unfortunate. My understanding is that at least this > behavior is deterministic, right? That is, either the same-origin frames > will be able to script each other or they won't and this will happen > consistently (based on the agent cluster key). > > An observation I had is that it seems that the Document-Isolation-Policy > is still at the mercy of the platform having the resources to > process-isolate frames. It wasn't clear to me from the explainer whether > this is already a limitation with the COOP and COEP approaches, however > unwieldy those may be. This basically means that one of the listed use-case > of authors maintaining two copies of their widgets -- one with > SharedArrayBuffers, one without -- doesn't seem to be addressed. Also I'm > not sure if it would be possible for 3p iframes to starve platform of such > resources so that the top level frame would no longer be able to create 1p > frames that have access to COI-gated APIs > > (I also don't know what is the right forum in which to raise these issues) > > Thanks, > Vlad > > On Wed, Apr 3, 2024 at 1:54 PM Charlie Reis <[email protected]> wrote: > > Thanks for sharing this. I do think it's worth calling attention to this > paragraph > <https://github.com/explainers-by-googlers/document-isolation-policy?tab=readme-ov-file#browsing-context-group-switch-instead-of-agent-cluster-keying> > of the explainer, for one thing to consider about the proposal: > > The Document-Isolation-Policy proposal relies on agent cluster keying to > achieve isolation, instead of browsing context group switches. This means > that it introduces a situation where two same-origin documents might find > themselves in different agent clusters and be unable to have DOM access to > each other. This is unprecedented in the HTML spec. > > > In other words, two same-origin frames within the same page (or anywhere > in the same browsing context group) can end up in different processes, > unable to script each other. It could be that this is considered fine and > might be outweighed by the benefits of the proposal, though it does have > some implications for web developers and for the browser's implementation: > > - Web developers might be confused when some attempts to script a > same-origin frame fail, since this has always been possible within a given > browsing context group. Maybe this can be mitigated with a different type > of error message in the DevTools console? > - In Chromium's implementation, both the browser process and renderer > process make assumptions that same-origin frames within the same browsing > context group (also known as content::BrowsingInstance) must be in the same > process so that they can script each other. Dividing that up based on > Document-Isolation-Policy seems like it should be possible, though it would > add some complexity and might require some auditing of process model > > <https://chromium.googlesource.com/chromium/src/+/main/docs/process_model_and_site_isolation.md> > code. > > Maybe this is a manageable risk? > > Thanks, > Charlie > > > On Wed, Apr 3, 2024 at 5:41 AM Camille Lamy <[email protected]> wrote: > > Contact [email protected] > > Explainerhttps://github.com/explainers-by-googlers/document- > isolation-policy > > SpecificationNone > > Summary > > Document-Isolation-Policy allows a document to enable crossOriginIsolation > for itself, without having to deploy COOP or COEP, and regardless of the > crossOriginIsolation status of the page. The policy is backed by process > isolation. Additionally, the document non-CORS cross-origin subresources > will either be loaded without credentials or will need to have a CORP > header. > > > Blink componentBlink>SecurityFeature > <https://bugs.chromium.org/p/chromium/issues/list?q=component:Blink%3ESecurityFeature> > > Motivation > > Developers want to build applications that are fast using > SharedArrayBuffers (SAB), which can improve computation time by ~40%. But > SharedArrayBuffers allow to create high-precision timers that can be > exploited in a Spectre attack, allowing to leak cross-origin user data. To > mitigate the risk, SharedArrayBuffers are gated behind crossOriginIsolation > (COI). CrossOriginIsolation requires to deploy both > Cross-Origin-Opener-Policy (COOP) and Cross-Origin-Embedder-Policy (COEP). > Both have proven hard to deploy, COOP because it prevents communication > with cross-origin popups, and COEP because it imposes restrictions on > third-party embeds. Finally, the whole COOP + COEP model is focused on > providing access to SharedArrayBuffers to the top-level frame. Cross-origin > embeds can only use SABs if their embedder deploys crossOriginIsolation and > delegates the permission to use COI-gated APIs, making the availability of > SABs in third-party iframes very unreliable. Document-Isolation-Policy, is > proposing to solve these deployment concerns by relying on the browser > Out-of-Process-Iframe capability. It will provide a way to securely build > fast applications using SharedArrayBuffers while maintaining communication > with cross-origin popups and not requiring extra work to embed cross-origin > iframes. Finally, it will be available for embedded widgets. > > > Initial public proposalhttps://github.com/WICG/proposals/issues/145 > > TAG reviewNone > > TAG review statusPending > > Risks > > > Interoperability and Compatibility > > None > > > *Gecko*: No signal > > *WebKit*: No signal > > *Web developers*: No signals > > *Other signals*: > > WebView application risks > > Does this intent deprecate or change behavior of existing APIs, such that > it has potentially high risk for Android WebView-based applications? > > None > > > Debuggability > > None > > > Is this feature fully tested by web-platform-tests > <https://chromium.googlesource.com/chromium/src/+/main/docs/testing/web_platform_tests.md> > ?No > > Flag name on chrome://flagsNone > > Finch feature nameNone > > Non-finch justificationNone > > Requires code in //chrome?False > > Estimated milestones > > No milestones specified > > > Link to entry on the Chrome Platform Statushttps://chromestatus.com/ > feature/5141940204208128?gate=5097535879512064 > > This intent message was generated by Chrome Platform Status > <https://chromestatus.com/>. > > -- > 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 [email protected]. > To view this discussion on the web visit https://groups.google.com/a/ > chromium.org/d/msgid/blink-dev/CAMKsNvp7Xcgz%3DcMXJ7%2B% > 2BBgwhO2wOKEkaMiDk_wUY1nprvPG4HQ%40mail.gmail.com > <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAMKsNvp7Xcgz%3DcMXJ7%2B%2BBgwhO2wOKEkaMiDk_wUY1nprvPG4HQ%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > > -- > 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 [email protected]. > To view this discussion on the web visit https://groups.google.com/a/ > chromium.org/d/msgid/blink-dev/CAH%2B8MBZfpRCRsQMqvjy8NPmo9_ > v8WQcdu%3Ds06%2BVWb6a17dQ-jw%40mail.gmail.com > <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAH%2B8MBZfpRCRsQMqvjy8NPmo9_v8WQcdu%3Ds06%2BVWb6a17dQ-jw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > > -- > 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 [email protected]. > To view this discussion on the web visit https://groups.google.com/a/ > chromium.org/d/msgid/blink-dev/1db78c8e-0672-4f62-ac3e- > 22cb614ca2a6n%40chromium.org > <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/1db78c8e-0672-4f62-ac3e-22cb614ca2a6n%40chromium.org?utm_medium=email&utm_source=footer> > . > > -- > 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 [email protected]. > > To view this discussion on the web visit https://groups.google.com/a/ > chromium.org/d/msgid/blink-dev/CAOmohS%2BzyOX6amnva6t_ > HBsXPXAFoZEri7A78ka7-OwA66B%3Dmw%40mail.gmail.com > <https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAOmohS%2BzyOX6amnva6t_HBsXPXAFoZEri7A78ka7-OwA66B%3Dmw%40mail.gmail.com?utm_medium=email&utm_source=footer> > . > > -- 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 [email protected]. To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/blink-dev/CAH%2B8MBZLUq2gr%3D8iXK37cQPbzrG8CqqSBAV0iTy9MLJLkBoviA%40mail.gmail.com.
