A page calls transferControlToOffscreen() and posts the
canvas to a document on another site, so the receiver lives in a different
renderer process. The canvas carries the frame sink ids of the sending
process. When the receiver draws, it asks viz for that frame sink, the
browser sees a client id that does not belong to this renderer, and kills it
with RESULT_CODE_KILLED_BAD_MESSAGE ("Invalid client ID").
The fix drops the foreign ids while deserializing, so the canvas becomes a standalone one. Nothing reaches the sender's placeholder canvas -- it never could -- but no renderer dies. Reported in 2022, and again in 2026 as #471204157; the CL carries the tests from both investigations.
The screenshots come from hjanuschka/cl-8626307-offscreencanvas. The page pings the cross-site iframe every 300ms and shows whether it still answers.
python3 serve.py &
chrome --host-resolver-rules="MAP * 127.0.0.1" http://a.test:8765/parent.html
OffscreenCanvasCrossProcessBrowserTest.CrossSiteTransferDoesNotKillRenderer
drives the same flow between two sites and fails with
"RenderFrame deleted." when the fix is compiled out.V8ScriptValueSerializerTest.TransferOffscreenCanvasFromAnotherProcess
pins the deserializer: foreign ids come back as a standalone canvas.TransferOffscreenCanvasSameProcess pins the other
direction: matching ids keep the placeholder link.fast/canvas/offscreen-frame-transfer.html guards the
normal same-process transfer at the web level.V8ScriptValueSerializerTest and
OffscreenCanvas unit tests pass (124).Platform::GetFrameSinkClientId() just to
compare ids. The alternative is reading the client id out of a throwaway
GenerateFrameSinkId(), which burns a sink id per transfer.