Null pointer dereference in JSC::GetByStatus

[Mikhail R. Gadelha]

In GetByStatus::computeForStubInfoWithoutExitSiteFeedback, there are several places that dereference the stubInfo argument when calling the GetByStatus constructor. Here's the backtrace:


#0  JSC::GetByStatus::GetByStatus (this=0xfffe9438, summary=JSC::StubInfoSummary::NoInformation, stubInfo=0x0) at ../../Source/JavaScriptCore/bytecode/GetByStatus.cpp:177
#1  0xf27896e6 in JSC::GetByStatus::computeForStubInfoWithoutExitSiteFeedback (locker=..., profiledBlock=0xec49b560, stubInfo=0x0, callExitSiteData=...)
    at ../../Source/JavaScriptCore/bytecode/GetByStatus.cpp:207
#2  0xf2788fa4 in JSC::GetByStatus::computeFor (profiledBlock=0xec49b560, map=..., bytecodeIndex=..., didExit=..., callExitSiteData=...) at ../../Source/JavaScriptCore/bytecode/GetByStatus.cpp:155
#3  0xf278aeb4 in JSC::GetByStatus::computeFor (profiledBlock=0xec49b560, baselineMap=..., icContextStack=..., codeOrigin=...) at ../../Source/JavaScriptCore/bytecode/GetByStatus.cpp:402
#4  0xf2c1d286 in JSC::DFG::ByteCodeParser::parseBlock (this=0xfffecd6c, limit=251) at ../../Source/JavaScriptCore/dfg/DFGByteCodeParser.cpp:6278
#5  0xf2c60b34 in JSC::DFG::ByteCodeParser::parseCodeBlock (this=0xfffecd6c) at ../../Source/JavaScriptCore/dfg/DFGByteCodeParser.cpp:8804
#6  0xf2c61c18 in JSC::DFG::ByteCodeParser::parse (this=0xfffecd6c) at ../../Source/JavaScriptCore/dfg/DFGByteCodeParser.cpp:9006
#7  0xf2c64000 in JSC::DFG::parse (graph=...) at ../../Source/JavaScriptCore/dfg/DFGByteCodeParser.cpp:9165
#8  0xf31f490a in JSC::DFG::Plan::compileInThreadImpl (this=0xeb01e8d0) at ../../Source/JavaScriptCore/dfg/DFGPlan.cpp:199
#9  0xf4055450 in JSC::JITPlan::compileInThread (this=0xeb01e8d0, thread=0x0) at ../../Source/JavaScriptCore/jit/JITPlan.cpp:165
#10 0xf40ed054 in JSC::JITWorklist::enqueue (this=0xedcc71b0, plan=...) at ../../Source/JavaScriptCore/jit/JITWorklist.cpp:83
#11 0xf2f5931a in JSC::DFG::compileImpl (vm=..., codeBlock=0xec49aca0, profiledDFGCodeBlock=0x0, mode=JSC::JITCompilationMode::DFG, osrEntryBytecodeIndex=..., mustHandleValues=..., callback=...)
    at ../../Source/JavaScriptCore/dfg/DFGDriver.cpp:90
#12 0xf2f5944a in JSC::DFG::compile (vm=..., codeBlock=0xec49aca0, profiledDFGCodeBlock=0x0, mode=JSC::JITCompilationMode::DFG, osrEntryBytecodeIndex=..., mustHandleValues=..., callback=...)
    at ../../Source/JavaScriptCore/dfg/DFGDriver.cpp:106
#13 0xf4045c08 in JSC::operationOptimize (vmPointer=0xecaf5000, bytecodeIndexBits=0) at ../../Source/JavaScriptCore/jit/JITOperations.cpp:2096
#14 0xecc320aa in ?? ()

[Radar WebKit Bug Importer]

<rdar://problem/82535652>

[Mikhail R. Gadelha]

Created attachment 436793 [details]
Patch

[Yusuke Suzuki]

Comment on attachment 436793 [details]
Patch

Can you upload a test for review too? (while, maybe, we should not land it together)

[Yusuke Suzuki]

(In reply to Yusuke Suzuki from comment #3)
> Comment on attachment 436793 [details]
> Patch
> 
> Can you upload a test for review too? (while, maybe, we should not land it
> together)

I'm assuming that you found this bug based on some tests.

[Mikhail R. Gadelha]

Created attachment 438719 [details]
Test

Sure, that's the testcase.

[Mikhail R. Gadelha]

It was found by our fuzzer, let me know if you need a reduced testcase.

I reproduced the null pointer dereference in mips, armv7 and x86_64. I didn't test it in arm 64.

[Yusuke Suzuki]

Comment on attachment 436793 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=436793&action=review

r=me

> Source/JavaScriptCore/bytecode/GetByStatus.cpp:184
> +        RELEASE_ASSERT(stubInfo);

Make it ASSERT. I don't think it is worth RELEASE_ASSERT.

> Source/JavaScriptCore/bytecode/GetByStatus.cpp:188
> +        RELEASE_ASSERT(stubInfo);

Ditto.

[Mikhail R. Gadelha]

Created attachment 438837 [details]
Patch

[Mark Lam]

Is this patch ready to land?

[Yusuke Suzuki]

This crash does not happen in usual build, so changing it to non security issue.

[EWS]

Committed r282950 (242040@main): <https://commits.webkit.org/242040@main>

All reviewed patches have been landed. Closing bug and clearing flags on attachment 438837 [details].