October 2021 - webkit-dev - lists.webkit.org

Smart Pointer Analysis Tool for WebKit
by Ryosuke Niwa 01 Mar '23

01 Mar '23

Hi all, I’ve been working with Geoff (ggaren) and Jan Korous from Apple's compiler team to create a static analyzer which detects dangerous use of ref counted objects, and we’re looking for folks who are interested in trying out this tool and helping us refine the tool. Please let me know if you’re interested in using this tool & try applying code changes to WebKit. Our goal is to eventually integrate this tool into buildbot and EWS so that we can catch dangerous code before they get committed. *What is Dangerous?* So *what is* a *dangerous* use of ref counted objects you may ask? It’s any use of which we can’t trivially conclude that it doesn’t lead to a use-after-free. For now, we don’t detect dangerous use of non-ref counted objects including ones that can vend WeakPtr. It’s on us, humans, to decide which objects need to be ref counted or need to be CanMakeWeakPtr. Consider the following example. This code most certainly will lead to a use-after-free of “parent” in the third line because the code doesn’t keep the parent alive. Because isContentEditable updates style in some cases, it could lead to arbitrary script execution which could remove the parent from the document. Node* parent = element->parentElement(); if (parent->isContentEditable()) parent->scrollIntoView(); In general, relying on a complex data structure such as DOM tree to keep RefCounted objects alive while we call a non-trivial function is not safe. All it takes for the code to have a use-after-free is for someone to start updating style, layout, etc… inside the function either directly or indirectly. And we don’t want to make WebKit development really hard by forcing anyone who modifies a function to check every caller of the function and their callers, etc… to make sure it’s safe to do so. For this reason, it’s *dangerous* to store a raw pointer or a reference to a ref counted object as a local variable and use it after calling a non-trivial function. We did a similar analysis of a number of other patterns and usage of ref counted objects in WebKit and came up with the following basic rules for using ref counted objects in a safe manner. We’re hoping that these rules would be eventually incorporated into our coding style guideline: https://webkit.org/code-style-guidelines/ *Rules for Using Ref Counted Objects* 1. Every data member to a ref counted object must use either Ref, RefPtr, or WeakPtr. webkit.NoUncountedMemberChecker <https://github.com/llvm/llvm-project/blob/master/clang/docs/analyzer/checke…> 2. Every ref counted base class, if it has a derived class, must define a virtual destructor. webkit.RefCntblBaseVirtualDtor <https://github.com/llvm/llvm-project/blob/master/clang/docs/analyzer/checke…> 3. Every ref counted object passed to a non-trivial function as an argument (including "this" pointer) should be stored as a Ref or RefPtr in the caller’s local scope unless it's an argument to the caller itself by transitive property [1]. alpha.webkit.UncountedCallArgsChecker <https://github.com/llvm/llvm-project/blob/master/clang/docs/analyzer/checke…> 4. Every ref counted object must be captured using Ref or RefPtr for lambda. webkit.UncountedLambdaCapturesChecker <https://github.com/llvm/llvm-project/blob/master/clang/docs/analyzer/checke…> 5. Local variables - we’re still working on this ( https://reviews.llvm.org/D83259) Below, I've dissected each one of these rules with the real warning emitted by the analysis tool in development. Please let me know if you have any comments / concerns. ------------------------------ (1) is pretty trivial. Every ref counted data member should be stored using Ref, RefPtr, or WeakPtr since it would be not trivially obvious that life cycles of two or more objects are correctly tied or managed together. ------------------------------ (2) is also pretty easy to understand. In the following example, if someone destroys an instance of B using Ref<A>, then it would result in an undefined behavior. struct A : public RefCounted<A> { Vector<int> someData; }; struct B : public A { Vector<int> otherData; }; ------------------------------ For (3), passing a ref counted object as a raw pointer or reference to a function as an argument, the tool may generate a warning like this: Source/WebCore/html/FormAssociatedElement.cpp:181:13: warning: [WebKit] call argument is a raw pointers to a ref-countable type [-Wfunction-arg-ptr-to-refcntbl] setForm(findAssociatedForm(&asHTMLElement(), originalForm.get())); ^ This warns that void setForm(HTMLFormElement*) is called with the result of findAssociatedForm, which returns HTMLFormElement* without storing it anywhere. If setForm can somehow cause HTMLFormElement to be deleted, then this can result in the use-after-free in setForm. void FormAssociatedElement::resetFormOwner() { RefPtr<HTMLFormElement> originalForm = m_form.get(); setForm(findAssociatedForm(&asHTMLElement(), originalForm.get())); HTMLElement& element = asHTMLElement(); auto* newForm = m_form.get(); if (newForm && newForm != originalForm && newForm->isConnected()) element.document().didAssociateFormControl(element); } Why, you may ask, we don't put HTMLFormElement* in a Ref or RefPtr in setForm instead? This is because a lot of code has calls to a function with a local variable. Because the callee doesn’t know whether the caller has already stored each argument in Ref / RefPtr or not, we need to be safe and store them again in Ref / RefPtr, resulting in an unnecessary ref-churn. Additionally, if we took the approach of the callee being responsible for keeping every argument alive, then every non-trivial member function of a ref counted object must have "protectedThis" at the beginning of the function, which would be particularly wasteful if those member functions are sometimes called by other member functions. Once this rule is applied everywhere, for example, we can get rid of “protectedThis” from our codebase because it would be redundant. Furthermore, this rule transitively allows a raw pointer or a reference passed in as an argument to be used as arguments to call another function without storing it in Ref or RefPtr. For example, the following function satisfies this rule because both oldDocument and newDocument are raw references passed to this function as arguments, and therefore the caller of this function must have already stored as Ref / RefPtr as local variables. void HTMLMediaElement::didMoveToNewDocument(Document& oldDocument, Document& newDocument) { ASSERT_WITH_SECURITY_IMPLICATION(&document() == &newDocument); if (m_shouldDelayLoadEvent) { oldDocument.decrementLoadEventDelayCount(); newDocument.incrementLoadEventDelayCount(); } unregisterWithDocument(oldDocument); registerWithDocument(newDocument); HTMLElement::didMoveToNewDocument(oldDocument, newDocument); updateShouldAutoplay(); } ------------------------------ For (4), capturing a ref counted object as a raw pointer or a reference, the tool may generate a warning like this: Source/WebKit/UIProcess/API/APIHTTPCookieStore.cpp:115:74: warning: [WebKit] captured a raw pointer to a ref-countable type [-Wlambda-capture-ptr-to-refcntbl] cookieManager->deleteCookie(m_owningDataStore->sessionID(), cookie, [pool = WTFMove(pool), completionHandler = WTFMove(completionHandler)]() mutable { ^ This is warning about webPage object getting captured using a raw reference in the following function: void HTTPCookieStore::deleteCookie(const WebCore::Cookie& cookie, CompletionHandler<void()>&& completionHandler) { auto* pool = m_owningDataStore->processPoolForCookieStorageOperations(); if (!pool) { if (m_owningDataStore->sessionID() == PAL::SessionID::defaultSessionID() && !cookie.session) deleteCookieFromDefaultUIProcessCookieStore(cookie); else m_owningDataStore->removePendingCookie(cookie); RunLoop::main().dispatch([completionHandler = WTFMove(completionHandler)] () mutable { completionHandler(); }); return; } auto* cookieManager = pool->supplement<WebKit::WebCookieManagerProxy>(); cookieManager->deleteCookie(m_owningDataStore->sessionID(), cookie, [pool = WTFMove(pool), completionHandler = WTFMove(completionHandler)]() mutable { completionHandler(); }); } Here, we should have stored WebProcessPool in Ref instead. - R. Niwa

9 17

Request for Position: MSE-in-Workers
by Matt Wolenetz 27 Jul '22

27 Jul '22

Hi webkit-dev, This is a request for Webkit's position on the MSE-in-Workers feature: Explainer: https://github.com/wicg/media-source/blob/mse-in-workers-using-handle/mse-i… Main MSE spec issue tracking this feature: https://github.com/w3c/media-source/issues/175 ChromeStatus entry: https://chromestatus.com/features/5177263249162240 Draft specification: https://github.com/w3c/media-source/pull/282 TAG review: https://github.com/w3ctag/design-reviews/issues/656 Demo page (Chromium has experimental implementation): https://wolenetz.github.io/mse-in-workers-demo/mse-in-workers-demo.html Please let us know if you have any feedback! Thanks, Matt

1 1

Request for Position: User Preference Media Features Client Hints Headers
by Thomas Steiner 29 Jun '22

29 Jun '22

Hi WebKittens, We are interested in hearing your thoughts on User Preference Media Features Client Hints Headers, a set of HTTP Client Hints headers around user preference media features as defined by Media Queries Level 5. Specification Title: User Preference Media Features Client Hints Headers Specification or proposal URL: https://tomayac.github.io/user-preference-media-features-headers/ Explainer URL: https://github.com/tomayac/user-preference-media-features-headers/blob/main… TAG review: https://github.com/w3ctag/design-reviews/issues/632 Mozilla Standards Position: https://github.com/mozilla/standards-positions/issues/526 Thank you very much for your time and consideration! Cheers, Tom

1 5

Request for feedback on CompressionStream and DecompressionStream
by Adam Rice 02 May '22

02 May '22

I am trying to gauge feedback on compression streams with a view to shipping them in Chromium. Very briefly, they are a way to do gzip and gunzip in the browser. Less briefly, the explainer https://github.com/WICG/compression/blob/master/explainer.md goes into some detail of the how and why. The specification https://wicg.github.io/compression/ gives verbose detail of how. You may also find the W3C TAG review https://github.com/w3ctag/design-reviews/issues/410 interesting. Any and all feedback is welcome! FAQ: Q. Will this support brotli, zstd, etc.? A. Not yet. It's on the radar. Thanks, Adam Rice, Chromium

3 3

Request for Position: Multi-Screen Window Placement API
by Ben Morss 12 Apr '22

12 Apr '22

Hi all, We'd love to get WebKit's position on the new Multi-Screen Window Placement API. The proposal is to make incremental improvements to existing screen information and window placement APIs, allowing web applications to be intentional about the experiences they offer to users of multi-screen devices. We'd like to make it simple to: - detect when screen properties changes - discover the presence of extended screen areas - request extended screen information - place a window on a specific screen - show a window fullscreen on a specific screen This is useful for applications like the following: - a slideshow app shows slides full-screen on a projector, while displaying speaker notes on a laptop - a creativity app pops up secondary windows, like palettes, on a separate screen - a videoconferencing app shows video on a large screen, and controls on a smaller screen - a medical app shows images on multiple displays - a financial app shows dashboards on multiple monitors Future work could involve support for foldable devices, showing a window that spans multiple screens, etc. For more details, please see these links: - Explainer: https://github.com/webscreens/window-placement/blob/main/EXPLAINER.md - Draft Community Group Report: https://webscreens.github.io/window-placement/ - Additional ideas: https://github.com/webscreens/window-placement/blob/main/additional_explora… Thanks for your feedback on this! All best, Ben

1 1

Re: [webkit-dev] Request For Position on CSS containment
by cathiechen 24 Dec '21

24 Dec '21

Hi webkit-dev, We made a lot of progress regarding CSS containment [1]. Rob and I have finished the layout containment and size containment [2].\o/ And the patches of paint containment and style containment are ready for review now [3]. So we think now it's time to move on to content-visibility: (https://www.w3.org/TR/css-contain-2/#content-visibility) Since content-visibility depends on paint and style containment, we will do some specification research first, then prototype it based on Rob's patches. Thank you for your feedback! [1] https://bugs.webkit.org/show_bug.cgi?id=172026 [2] https://bugs.webkit.org/show_bug.cgi?id=223569, https://bugs.webkit.org/show_bug.cgi?id=223570 [3] https://bugs.webkit.org/show_bug.cgi?id=224742, https://bugs.webkit.org/show_bug.cgi?id=226458 Best, Rob and Cathie

3 5

Running the GTK port on macOS with Docker
by Philippe Normand 23 Nov '21

23 Nov '21

Hi, The WPE/GTK ports nowadays rely on a SDK that provides all the tools needed for development, it's used on the bots as well. Currently we run it with Flatpak, but technically, Docker can also be used. I've actually checked that a GTK MiniBrowser build downloaded from the bots can run on macOS with Docker, that involves setting up XQuartz, it's not great, but for quick testing, I wonder if that could be useful for the Apple folks? The GTK port could also be built on macOS using this docker setup, the SDK includes the GCC and clang versions used on the bots. You can already do this with a VM, but then if you want to use the SDK it adds another lever of virtualization, which is not great. So directly using the SDK through Docker seems more appealing, to me at least :) WPE can't run, I haven't managed to start Weston with its X11 backend in XQuartz... Ideally I'd prefer a native Wayland compositor for macOS but that doesn't seem to exist yet. Anyway, please let me know if some folks are interested. Philippe

2 4

Fuzzy Reftest Plans, and Metadata Locations
by Sam Sneddon 01 Nov '21

01 Nov '21

Hi! As part of the ongoing work on GPU Process, we’re interested in adding support for reftest fuzzy matching (i.e., allowing a certain amount of tolerance when comparing the generated images). Our intention is to match the semantics of WPT’s reftests (https://web-platform-tests.org/writing-tests/reftests.html#fuzzy-matching) <meta name=fuzzy content="maxDifference=15;totalPixels=300"> There are cases where we’ll want to apply these to the tests unconditionally, for example where varying behaviour is expected across ports (such as anti-aliasing differences), and in these cases for WPT tests these annotations should probably be exported upstream. The current plan, and work is underway to do this, is to support this syntax via parsing the HTML in Python when there is a hash mismatch, which should minimise the performance impact versus always reading this metadata. However, this doesn’t entirely suffice. There are cases where we might want to allow more tolerance on one platform or another, or vary based on GPU model or driver. As such, this requires not only platform specific metadata (i.e., similar to that which we have in TestExpectations files today), but also expectations with finer granularity. As such I think there are a few options here: One option is to extend the meta content to encode conditional variants, though this doesn’t work for WPT tests (unless we get buy-in to upstream these annotations into the upstream repo, though that might be desirable for the sake of results on wpt.fyi). We would need to be confident that this wouldn’t become unwieldy however; we wouldn’t want to end up with something like (if:port=Apple)maxDifference=1;totalPixels=10,(if:platform=iOS)maxDifference=10;totalPixels=20,(if:port=GTK)maxDifference=10;totalPixels=300. Another option is to extend TestExpectations to store more specific data (though again this might become unwieldy, as we’re unlikely to add new “platforms” based on every variable we might want to distinguish results on). This also means the metadata is far away from the test itself, and the TestExpectations files would continue to grow even further (and we already have 34k lines of TestExpectations data!). TestExpectations is also a rather horrible file format to modify the parser of. There is also test-options.json which has most of the same downsides as TestExpectations, albeit without the pain in modifying the parser. Finally, we could add per-test or per-directory files alongside the tests. (Due to how things work, these could presumably also be in directories in platform/.) This I think is probably the best option as it keeps the metadata near the test, without needing to modify the test (which, per above, is problematic for WPT as we move to automatically exporting changes). One could imagine either a __dir__-metadata.json (to use a similar name to how WPT names directory-level metadata files) or a -expected-fuzzy.json file alongside each test. Your opinions would be warmly welcomed! Thanks, Sam

5 4

-Wreturn-type and -Wredundant-move reminders
by Michael Catanzaro 29 Oct '21

29 Oct '21

Hi devs, A reminder about this common idiom: switch (...) { case Foo: return ...; case Bar: return ...; } RELEASE_ASSERT_NOT_REACHED(); When it's intended that the code always returns inside the switch statement, the RELEASE_ASSERT_NOT_REACHED() is required to avoid tripping GCC's -Wreturn-type. If you forget it, I or somebody else will wind up adding it later to avoid the warning. Clang does not warn here, and this is the most common type of warning I clean up, so please don't forget! :) This warning is useful in other situations, and it seems nicer to placate GCC than to disable it. I'll sneak in another reminder: "return WTFMove()" is almost always not needed. Clang warns only if the move prevents return value optimization, but GCC will always warn if it detects the move is unneeded. Have a nice day, Michael

2 2

Aditya Keerthi is now a WebKit reviewer!
by Tim Horton 28 Oct '21

28 Oct '21

Hello! I am delighted to announce that Aditya Keerthi is now a WebKit reviewer. Now it's time for all of you to send him tricky patches and see how he does :) Congratulations, Aditya! --Tim

1 0