pwn2own 2026 bug - JavaScript Promise Integration in WASM
Categories
(Core :: JavaScript: WebAssembly, defect, P3)
Tracking
()
| Tracking | Status | |
|---|---|---|
| firefox-esr115 | --- | unaffected |
| firefox-esr140 | --- | disabled |
| firefox150 | --- | disabled |
| firefox151 | --- | disabled |
| firefox152 | --- | fixed |
People
(Reporter: freddy, Unassigned)
Details
Attachments
(7 files)
TBD
| Reporter | ||
Comment 1•6 months ago
|
||
| Reporter | ||
Comment 2•6 months ago
|
||
| Reporter | ||
Comment 3•6 months ago
|
||
| Reporter | ||
Comment 4•6 months ago
|
||
| Reporter | ||
Comment 5•6 months ago
|
||
| Reporter | ||
Updated•6 months ago
|
Updated•6 months ago
|
Updated•6 months ago
|
Updated•6 months ago
|
Comment 6•6 months ago
|
||
The write up refers to the old JS-PI implementation and bug 2008358 that we fixed as part of it. I haven't been able to reproduce on the Nightly yet, so there's a chance we have fixed this.
Comment 7•6 months ago
|
||
Claude translated exploit to just use the JS shell builtins. Shell code replaced with -2.458343745902715e+255 (0xf4).
Comment 8•6 months ago
|
||
I can reproduce this issue in a 4-13-2026 nightly JS shell build, but not on a 5-12-2026 JS shell build (5f15185120f488edb4b4f2f38599fff543b9aa5a). Bug 2008358 was fixed by bug 2023217 on 4-16-2026.
From the writeup:
Ion's GVN pass merges `WasmArrayDataPointer` loads across JSPI suspension points because `wasmCallAliasSet()` explicitly excludes `WasmArrayDataPointer` from its clobber set, trusting the GC to rewrite any stale spill slots after an array moves. But `Instance::updateFrameForMovingGC` is never called for JSPI-suspended stacks — only `Instance::traceFrame` runs (via `TraceSuspendableStack`), and `traceFrame` skips `StackMap::Kind::ArrayDataPointer` entries. The combined result is that after a minor GC moves a WasmGC array while a JSPI stack is suspended, the Ion-cached `data_` pointer is stale: it still points at the array's old nursery location. Subsequent `array.get` or `array.set` operations on the resumed stack read from or write to freed memory.
Bug 2023217 fixed this by tracing suspended stacks for moving GCs and updating array data pointers [1]. That lines up with this issue being reproducible in 4-13-2026, but not 5-12-2026.
I'll test some more builds just to be triple sure, but this is has been fixed already.
Comment 9•6 months ago
|
||
Quick manual bisect:
Reproduces: https://ftp.mozilla.org/pub/firefox/nightly/2026/04/2026-04-15-21-21-14-mozilla-central/
No longer reproduces: https://ftp.mozilla.org/pub/firefox/nightly/2026/04/2026-04-16-09-37-20-mozilla-central/
That confirms that bug 2023217 (landed 4-16) fixed this.
We also enabled JS-PI in Nightly only last week, so this never was enabled with the bug in it.
Comment 10•6 months ago
|
||
Unable to reproduce bug 2039571 using build mozilla-central 20260514091837-3710783bfb55. Without a baseline, bugmon is unable to analyze this bug.
Removing bugmon keyword as no further action possible. Please review the bug and re-add the keyword for further analysis.
Updated•6 months ago
|
Updated•2 months ago
|
Description
•