fix(elf): preserve virtual relocation identity (#9990)

* fix(elf): keep compressed storage out of decoded metadata facts

* fix(elf): preserve virtual relocation identity

Keep linked initializer backing and relocation values in their owning virtual
or ET_REL section coordinates. Replace the IL2CPP physical-copy materializer
with compact virtual value facts while preserving file-backed capacity.

Reuse initializer repairs and original regressions from ShuaixinHuang PRs
repair retains independent alias values instead of blanket alias refusal.

Co-authored-by: 帅鑫 Huang <1836656514@qq.com>

* fix(elf): admit compressed string tables the profile does not read

SHF_COMPRESSED string tables are rejected only when they name sections
or symbols. An unreferenced string table is not decoded, so it no longer
makes an otherwise analyzable ELF unsupported.

* fix(il2cpp): admit virtual pointer workspace before retention

* fix(elf): reject null relocation target sections

* fix(il2cpp): avoid decoding unrelated Mach-O words

* fix(il2cpp): preserve uncertainty in candidate scans

* fix(il2cpp): bound uncertain array candidates

* fix(il2cpp): preserve unknown registration candidates

* fix(il2cpp): classify type kinds before bound data

* fix(il2cpp): bind candidate counts to matching names

---------

Co-authored-by: Cursor Agent <cursoragent@cursor.com>
Co-authored-by: 帅鑫 Huang <1836656514@qq.com>
This commit is contained in:
authored and GitHub committed 2026-11-11 18:13:18 +08:00
1 parent 3c0e9d8a02
commit 8052f8daf0
18 files changed
+1807 -419

No files matched your search

+4 -4
View File
@@ -213,8 +213,8 @@ rejects it. The file is read, never loaded or executed.
A position-independent `libil2cpp.so` stores zero in most pointer slots and
keeps the value in `R_AARCH64_RELATIVE` or `R_X86_64_RELATIVE` relocations
(`DT_RELA`). The reader writes those values into a private copy before
searching it, so addresses are the link-time addresses and RVAs are relative to
(`DT_RELA`). The reader indexes loader values by virtual address while keeping
the source bytes unchanged, so addresses are the link-time addresses and RVAs are relative to
the first loaded segment. Android packed
relocations (`DT_ANDROID_RELA`) are refused. No target code is executed.
@@ -451,11 +451,11 @@ The builds checked are listed after the module search rules above.
| `generic-method-registration-not-found` | input error | No CodeRegistration layout bounds every genericMethodTable row, or the generic tables are unreadable. |
| `ambiguous-generic-method-registration` | input error | More than one CodeRegistration layout bounds every genericMethodTable row. |
Format and selection failures name `binary_path`; capacity refusals remain typed resource errors. Failed linking preserves the metadata source identity, validated counts, selected binary path/architecture and its SHA-256 when observed, plus prior producer observations. It never publishes an incomplete native table as complete. The binary input is bounded to 1 GiB.
Format and selection failures name `binary_path`; capacity refusals remain typed resource errors. Failed linking preserves the metadata source identity, validated counts, selected binary path/architecture and its SHA-256 when observed, plus prior producer observations. It never publishes an incomplete native table as complete. The binary input is bounded to 1 GiB. The virtual-pointer index has the same 1 GiB estimated representation allowance, reserving 2 KiB per distinct virtual destination before insertion for retained Maps and finish-time lookup storage. Repeated declarations of one destination share that reservation. This conservative allocation accounting is separate from actual process RSS and the report allowance; exhaustion returns a typed `native-index-capacity` refusal before another destination is retained.
Scans yield cooperatively in bounded batches. Name-address retention and reverse-trie construction are each limited to 65,536 entries/nodes and fallback module-array candidates to 64 (`native-index-capacity`); exceeding capacity does not prove absence. Rendered type text has a 64 MiB UTF-8 cache-workspace guard, separate from the 64 MiB encoded-report admission, and 64 nested rendering frames (`native-type-text-capacity` / `type-nesting-capacity`). Genuine cycles are malformed; unsupported type kinds are explicit coverage failures. Native row/text admission runs before allocation or retention. Unused generic parameter names are read on demand from the same stable metadata snapshot. There is no cross-inspection file cache; validated immutable Evidence is reused for session views.
ELF RELATIVE writes into zero-fill are kept as a private virtual-pointer overlay. Unresolved supported-width symbol relocations, including RELA PLT entries, remain unknown only for the bytes their ELF ABI relocation type covers; an adjacent 32-bit relocation does not make a separate 64-bit method slot unknown. Stored zeros never establish no-body. COPY, TLS descriptor, REL, packed APS2 and other unsupported relocation encodings fail closed. Input bytes are unchanged.
ELF RELATIVE writes use a virtual-pointer overlay for both file-backed and zero-fill locations. Distinct virtual addresses that share source file bytes retain independent values; an untouched alias keeps its stored bytes. Unresolved supported-width symbol relocations, including RELA PLT entries, remain unknown only for the bytes their ELF ABI relocation type covers; an adjacent 32-bit relocation does not make a separate 64-bit method slot unknown. Stored zeros never establish no-body. COPY, TLS descriptor, REL, packed APS2 and other unsupported relocation encodings fail closed. Input bytes are unchanged.
ELF relocation tables and file-backed destinations may span adjacent load ranges when every range agrees on the file translation. Any overlapping zero-fill declaration makes those bytes ambiguous; REA refuses the table or relocation destination instead of choosing the stored file bytes. Missing file bytes remain distinct from conflicting owners: supported unresolved relocations outside file-backed memory retain their unknown overlay, while ambiguous destinations refuse linking. Registration coverage uses the same file/zero-fill ownership checks as pointer reads.