967330 - Faulty: assert failure "updated region lies across rotation boundaries!" in ContentHostSingleBuffered::UpdateThebes

Status: RESOLVED FIXED

(Core :: Graphics, defect)

Description

[Benoit Jacob [:bjacob] (mostly away)]

Found by Christoph Diehl's "Faulty" fuzzer, see bug 777067

Comment 1

[Benoit Jacob [:bjacob] (mostly away)]

Argh, forgot to attach session. But this easy to reproduce, happens all the time. Hold on...

Comment 2

[Benoit Jacob [:bjacob] (mostly away)]

Attachment: Faulty session

Comment 3

[Benoit Jacob [:bjacob] (mostly away)]

Classification: PLayerTransactionParent protocol, bad assertion on message parameters, easy

Comment 4

[Benoit Jacob [:bjacob] (mostly away)]

Attachment: UpdateThebes full of grace

UpdateThebes needs to do graceful runtime error handling, because the region it receives can be an arbitrarily bad one from an untrusted client.

Comment 5

[Nicolas Silva [:nical]]

Comment on attachment 8373799 [details] [diff] [review]
UpdateThebes full of grace

Review of attachment 8373799 [details] [diff] [review]:
-----------------------------------------------------------------

::: gfx/layers/composite/ContentHost.cpp
@@ +520,5 @@
>  
>    if (!mTextureHost) {
>      mInitialised = false;
> +    return true; // FIXME should we return false? Returning true for now
> +  }              // to preserve existing behavior of NOT causing IPC errors.

It could be worth landing as is first and try to return false in a separate bug to see what we get. This is definitely something that we don't want to wind up hitting, but I suspect that sometimes it's just very hard to avoid it. For instance if suddenly we need to use a different Compositor/GLContext, we tend to get rid of textures and may miss a frame (which is probably ok when loosing a context). So far the strategy we seem to have followed on the compositor thread is to avoid crashing at all costs, even if it means breaking correctness for a frame. Also, for layer that may be affected by ImageBridge (so not in this case), it is very hard to ensure that your layer has a texture, because the first frame may arrive before or after the layer (textures almost always arrive before layers because the content thread is a busy guy). So when async video is involved, returning early may be a valid thing to do.

Comment 6

[Benoit Jacob [:bjacob] (mostly away)]

http://hg.mozilla.org/integration/mozilla-inbound/rev/88950b336660

Comment 7

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/88950b336660