Struct reading and disassembly script overhaul, various misc. loading fixes, bump to .NET 9 (#13)

* Bump projects to .net 9 and update nugets

* add VersionedSerialization + source generator

* migrate versioning to StructVersion class, add handling/detection for 29.2/31.2

* add new struct definitions

* rename serialization methods and add BinaryObjectStreamReader for interop

* Rework metadata struct loading to use new struct versioning

* move 29/31.1/.2 to use tags (-2022,-2023) instead of minor versions

* fix metadata usage validity checks

* rework code registration offsetting a bit and add second 29/31.1 condition

* tweak .1 condition (again)

* 29/31.2 was a psyop

* also remove 29.2 from the readme

* remove loading of packed dlls - this was a very unsafe feature

* support auto-recovering type indices from type handles
fixes loading of memory-dumped v29+ libraries since those replacee their class indices on load with a pointer to the corresponding type

* support loading PEs without an export table

* also read UnresolvedVirtualCallCount on regular v31

* Disable plugin loading for now

* Overhaul disassembler script + add Binary Ninja target (#12)

* Overhaul diassembler scripts:
- No longer defines top level functions
- Split into three classes: StatusHandler (like before), DisassemblerInterface (for interfacing with the used program API), ScriptContext (for definiting general functions that use the disassembler interface)
- Add type annotations to all class methods and remove 2.7 compatibility stuff (Ghidra now supports Python 3 so this is unnecessary anymore)
- Disassembler backends are now responsible for launching metadata/script processing, to better support disassembler differences
- String handling is back in the base ScriptContext class, disassembler interfaces opt into the fake string segment creation and fall back to the old method if it isn't supported

* Add Binary Ninja disassembler script backend
This uses the new backend-controlled execution to launch metadata processing on a background thread to keep the ui responsive

* make binary ninja script use own _BINARYNINJA_ define and add define helpers to header

* Update README to account for new script and binary ninja backend

* implement fake string segment functions for binary ninja but don't advertise support

* also cache API function types in binary ninja backend

* fix ida script and disable folders again

* Fix metadata usage issues caused by it being a value type now

* make TryMapVATR overrideable and implement it for ELFs

* Make field offset reading use TryMapVATR to reduce exceptions

* Fix NRE in Assembly ctor on < v24.2

* Update actions workflow to produce cross-platform CLI binaries, update readme to reflect .net 9 changes

* workflow: only restore packages for projects that are being built

* workflow: tweak caching and fix gui compilation

* workflow: remove double .zip in CLI artifact name

* 29/31.2 don't actually exist, this logic is not needed
This commit is contained in:
Luke
2024-11-14 14:32:11 +01:00
committed by GitHub
parent 5b0476fcc5
commit b05c03964a
130 changed files with 5117 additions and 4371 deletions

View File

@@ -9,6 +9,9 @@ using System.Buffers;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using Il2CppInspector.Next;
using Il2CppInspector.Next.BinaryMetadata;
using VersionedSerialization;
namespace Il2CppInspector
{
@@ -120,7 +123,7 @@ namespace Il2CppInspector
// Find CodeRegistration
// >= 24.2
if (metadata.Version >= 24.2) {
if (metadata.Version >= MetadataVersions.V242) {
// < 27: mscorlib.dll is always the first CodeGenModule
// >= 27: mscorlib.dll is always the last CodeGenModule (Assembly-CSharp.dll is always the first but non-Unity builds don't have this DLL)
@@ -137,7 +140,7 @@ namespace Il2CppInspector
// Unwind from string pointer -> CodeGenModule -> CodeGenModules + x
foreach (var potentialCodeGenModules in FindAllPointerChains(imageBytes, va, 2))
{
if (metadata.Version >= 27)
if (metadata.Version >= MetadataVersions.V270)
{
for (int i = imagesCount - 1; i >= 0; i--)
{
@@ -145,7 +148,7 @@ namespace Il2CppInspector
potentialCodeGenModules - (ulong) i * ptrSize, 1))
{
var expectedImageCountPtr = potentialCodeRegistrationPtr - ptrSize;
var expectedImageCount = ptrSize == 4 ? Image.ReadMappedInt32(expectedImageCountPtr) : Image.ReadMappedInt64(expectedImageCountPtr);
var expectedImageCount = Image.ReadMappedWord(expectedImageCountPtr);
if (expectedImageCount == imagesCount)
return potentialCodeRegistrationPtr;
}
@@ -203,24 +206,42 @@ namespace Il2CppInspector
return (0, 0);
var codeGenEndPtr = codeRegVa + ptrSize;
// pCodeGenModules is the last field in CodeRegistration so we subtract the size of one pointer from the struct size
codeRegistration = codeRegVa - ((ulong) metadata.Sizeof(typeof(Il2CppCodeRegistration), Image.Version, Image.Bits / 8) - ptrSize);
codeRegistration = codeGenEndPtr - (ulong)Il2CppCodeRegistration.Size(Image.Version, Image.Bits == 32);
// In v24.3, windowsRuntimeFactoryTable collides with codeGenModules. So far no samples have had windowsRuntimeFactoryCount > 0;
// if this changes we'll have to get smarter about disambiguating these two.
var cr = Image.ReadMappedObject<Il2CppCodeRegistration>(codeRegistration);
var cr = Image.ReadMappedVersionedObject<Il2CppCodeRegistration>(codeRegistration);
if (Image.Version == 24.2 && cr.interopDataCount == 0) {
Image.Version = 24.3;
codeRegistration -= ptrSize * 2; // two extra words for WindowsRuntimeFactory
if (Image.Version == MetadataVersions.V242 && cr.InteropDataCount == 0) {
Image.Version = MetadataVersions.V243;
codeRegistration = codeGenEndPtr - (ulong)Il2CppCodeRegistration.Size(Image.Version, Image.Bits == 32);
}
if (Image.Version == 27 && cr.reversePInvokeWrapperCount > 0x30000)
if (Image.Version == MetadataVersions.V270 && cr.ReversePInvokeWrapperCount > 0x30000)
{
// If reversePInvokeWrapperCount is a pointer, then it's because we're actually on 27.1 and there's a genericAdjustorThunks pointer interfering.
// We need to bump version to 27.1 and back up one more pointer.
Image.Version = 27.1;
codeRegistration -= ptrSize;
Image.Version = MetadataVersions.V271;
codeRegistration = codeGenEndPtr - (ulong)Il2CppCodeRegistration.Size(Image.Version, Image.Bits == 32);
cr = Image.ReadMappedVersionedObject<Il2CppCodeRegistration>(codeRegistration);
}
// genericAdjustorThunks was inserted before invokerPointersCount in 24.5 and 27.1
// pointer expected if we need to bump version
if (Image.Version == MetadataVersions.V244 && cr.InvokerPointersCount > 0x50000)
{
Image.Version = MetadataVersions.V245;
codeRegistration = codeGenEndPtr - (ulong)Il2CppCodeRegistration.Size(Image.Version, Image.Bits == 32);
cr = Image.ReadMappedVersionedObject<Il2CppCodeRegistration>(codeRegistration);
}
if ((Image.Version == MetadataVersions.V290 || Image.Version == MetadataVersions.V310) &&
cr.GenericMethodPointersCount >= cr.GenericMethodPointers)
{
Image.Version = new StructVersion(Image.Version.Major, 0, MetadataVersions.Tag2022);
codeRegistration = codeGenEndPtr - (ulong)Il2CppCodeRegistration.Size(Image.Version, Image.Bits == 32);
}
}
@@ -228,18 +249,15 @@ namespace Il2CppInspector
// <= 24.1
else {
// The first item in CodeRegistration is the total number of method pointers
vas = FindAllMappedWords(imageBytes, (ulong) metadata.Methods.Count(m => (uint) m.methodIndex != 0xffff_ffff));
if (!vas.Any())
return (0, 0);
vas = FindAllMappedWords(imageBytes, (ulong) metadata.Methods.Count(m => (uint) m.MethodIndex != 0xffff_ffff));
// The count of method pointers will be followed some bytes later by
// the count of custom attribute generators; the distance between them
// depends on the il2cpp version so we just use ReadMappedObject to simplify the math
foreach (var va in vas) {
var cr = Image.ReadMappedObject<Il2CppCodeRegistration>(va);
var cr = Image.ReadMappedVersionedObject<Il2CppCodeRegistration>(va);
if (cr.customAttributeCount == metadata.AttributeTypeRanges.Length)
if (cr.CustomAttributeCount == metadata.AttributeTypeRanges.Length)
codeRegistration = va;
}
@@ -253,16 +271,17 @@ namespace Il2CppInspector
// Find TypeDefinitionsSizesCount (4th last field) then work back to the start of the struct
// This saves us from guessing where metadataUsagesCount is later
var mrSize = (ulong) metadata.Sizeof(typeof(Il2CppMetadataRegistration), Image.Version, Image.Bits / 8);
var mrSize = (ulong)Il2CppMetadataRegistration.Size(Image.Version, Image.Bits == 32);
var typesLength = (ulong) metadata.Types.Length;
vas = FindAllMappedWords(imageBytes, typesLength).Select(a => a - mrSize + ptrSize * 4);
// >= 19 && < 27
if (Image.Version < 27)
foreach (var va in vas) {
var mr = Image.ReadMappedObject<Il2CppMetadataRegistration>(va);
if (mr.metadataUsagesCount == (ulong) metadata.MetadataUsageLists.Length)
if (Image.Version < MetadataVersions.V270)
foreach (var va in vas)
{
var mr = Image.ReadMappedVersionedObject<Il2CppMetadataRegistration>(va);
if (mr.MetadataUsagesCount == (ulong) metadata.MetadataUsageLists.Length)
metadataRegistration = va;
}
@@ -271,22 +290,17 @@ namespace Il2CppInspector
// Synonyms: copying, piracy, theft, strealing, infringement of copyright
// >= 27
else {
// We're going to just sanity check all of the fields
// All counts should be under a certain threshold
// All pointers should be mappable to the binary
var mrFieldCount = mrSize / (ulong) (Image.Bits / 8);
foreach (var va in vas) {
var mrWords = Image.ReadMappedWordArray(va, (int) mrFieldCount);
// Even field indices are counts, odd field indices are pointers
bool ok = true;
for (var i = 0; i < mrWords.Length && ok; i++) {
ok = i % 2 == 0 || Image.TryMapVATR((ulong) mrWords[i], out _);
}
if (ok)
else
{
foreach (var va in vas)
{
var mr = Image.ReadMappedVersionedObject<Il2CppMetadataRegistration>(va);
if (mr.TypeDefinitionsSizesCount == metadata.Types.Length
&& mr.FieldOffsetsCount == metadata.Types.Length)
{
metadataRegistration = va;
break;
}
}
}
if (metadataRegistration == 0)