Unify StubLinker stub memory management with JIT code allocator

[anurag6569201]

StubLinker-generated stubs used a dedicated LoaderAllocator::m_pStubHeap, separate from the executable memory infrastructure shared by JITed code and other stub kinds. This PR unifies the two paths.

• LoaderAllocator: removed m_pStubHeap and its init/terminate/DAC/size-reporting plumbing.
• StubLinker/Stub: StubLinker::Link and Stub::NewStub now take a LoaderAllocator* instead of a LoaderHeap*, allocating stub memory through ExecutionManager::GetEEJitManager()->AllocCodeFragmentBlock.
• StubCodeBlockKind: added STUB_CODE_BLOCK_STUBLINK and STUB_CODE_BLOCK_SHUFFLE_THUNK to distinguish these stubs in the code-fragment allocator.
• StubLinkStubManager: removed its own range list; RangeSectionStubManager now recognizes the new stub kinds and forwards stub identification/tracing to StubLinkStubManager (mirroring the existing pattern for jump stubs and call-counting stubs).
• DAC/SOS: updated datadescriptor.inc and debug/daccess/request.cpp to drop the removed stub heap field.
• Call sites: updated stubcache.cpp, comdelegate.cpp, prestub.cpp, assembly.cpp, appdomain.hpp/.inl, and class.h to pass LoaderAllocator* instead of a stub LoaderHeap*.

// Before
Stub *Link(LoaderHeap *heap, DWORD flags, const char *stubType);

// After
Stub *Link(LoaderAllocator *pLoaderAllocator, DWORD flags, const char *stubType);

Source merge-base: c36f112e7ad108b8f52e5d7028d2ecc09a4015f2
Source head: bb70c03b305a8f424e9924bb2defc66bcd7424b1

[shipwright-agent]

⛔ Shipwright · Blocked

Recommendation: do not merge PR #1 · Tier T3
Checks: 0 total · 0 needing attention

Next step: resolve the blocking findings before merge.

Findings (11)

• CRITICAL The StubCodeBlockKind enum values are renumbered (e.g. · src/coreclr/vm/codeman.h:98
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL In SetupShuffleThunk, the InterlockedCompareExchangeT now stores a raw PCODE and on contention the losing thread simply discards its newly created thunk without any DecRef or clean · src/coreclr/vm/comdelegate.cpp:931
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL StubLinkStubManager::CheckIsStub_Internal now unconditionally returns FALSE, and DoEnumMemoryRegions no longer enumerates the range list. · src/coreclr/vm/stubmgr.cpp:1259
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL GetStubTargetMethod reads a PTR_MethodDesc from immediately before the stub start address (pStubData - sizeof(PTR_MethodDesc)) without any validation that the memory layout actuall · src/coreclr/vm/stubmgr.cpp:1268
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL The removal of StubHolder and the Stub class's refcount-based lifetime management means all call sites that previously relied on automatic DecRef on scope exit now leak or double-f · src/coreclr/inc/holder.h:1128
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL The change from Stub::RecoverStub to GetStubTargetMethod removes the debug-only signature validation (kUsedStub) and the assertion that the entry point matches the stub. · src/coreclr/vm/stubmgr.cpp:1268
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• CRITICAL The removal of the range list enumeration in StubLinkStubManager::DoEnumMemoryRegions means DAC memory enumeration no longer includes stub ranges. · src/coreclr/vm/stubmgr.cpp:2267
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• HIGH The enum renumbering in StubCodeBlockKind is not accompanied by a comment explaining why the values changed or whether the numeric values are part of a serialized contract. · src/coreclr/vm/codeman.h:98
  • Fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.
• …and 3 more findings in the check details.

Fireworks usage: 59,680 input · 1,408 output · 61,088 total tokens · $0.0141 · 23s · 0 fix iteration(s)

Open the Shipwright check for full evidence and the audit bundle. Use /shipwright rerun to verify again.

[shipwright-agent]

Shipwright · CRITICAL

GetStubTargetMethod reads a PTR_MethodDesc from immediately before the stub start address (pStubData - sizeof(PTR_MethodDesc)) without any validation that the memory layout actuall

Impact: GetStubTargetMethod reads a PTR_MethodDesc from immediately before the stub start address (pStubData - sizeof(PTR_MethodDesc)) without any validation that the memory layout actually contains a MethodDesc pointer. If the stub kind is not WRAPPER_STUB or the allocation layout differs, this dereferences arbitrary memory and can crash or corrupt state.

Suggested fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.

[shipwright-agent]

Shipwright · CRITICAL

The change from Stub::RecoverStub to GetStubTargetMethod removes the debug-only signature validation (kUsedStub) and the assertion that the entry point matches the stub.

Impact: The change from Stub::RecoverStub to GetStubTargetMethod removes the debug-only signature validation (kUsedStub) and the assertion that the entry point matches the stub. In DAC builds, the old code explicitly checked m_signature and GetEntryPoint before dereferencing. The new code blindly reads a pointer before the stub address, creating a memory-safety vulnerability where a crafted or corrupted dump can cause arbit…

Suggested fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.

[shipwright-agent]

Shipwright · HIGH

The new GetCodeSize helper uses a ptrdiff_t subtraction and asserts codeSize > 0, but the callers pass symbol addresses that may not be ordered correctly on all platforms or build

Impact: The new GetCodeSize helper uses a ptrdiff_t subtraction and asserts codeSize > 0, but the callers pass symbol addresses that may not be ordered correctly on all platforms or build configurations. If the linker places JIT_WriteBarrier_End before JIT_WriteBarrier, the assert fires in release builds or the size is silently wrong.

Suggested fix: Review the cited evidence, fix the risk if confirmed, and rerun Shipwright.