Merge branch '_xeroxz' into 'master'

updated vmexit so that it gens an exit stub... also handed vm enters...

See merge request vmp2/vmdevirt!8

include/devirt_t.hpp

@@ -22,8 +22,8 @@
 
 #include "X86TargetMachine.h"
 #include "llvm/Pass.h"
-#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Passes/OptimizationLevel.h"
+#include "llvm/Passes/PassBuilder.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/Host.h"
 #include "llvm/Support/TargetRegistry.h"
@@ -48,17 +48,17 @@ namespace vm
         friend class lifters_t;
 
       public:
-        explicit devirt_t( llvm::LLVMContext *llvm_ctx, llvm::Module *llvm_module );
+        explicit devirt_t( llvm::LLVMContext *llvm_ctx, llvm::Module *llvm_module, vmp2::v4::file_header * vmp2_file );
         llvm::Function *lift( std::uintptr_t rtn_begin, std::vector< vm::instrs::code_block_t > vmp2_code_blocks );
         bool compile( std::vector< std::uint8_t > &obj );
 
       private:
         llvm::LLVMContext *llvm_ctx;
         llvm::Module *llvm_module;
+        vmp2::v4::file_header *vmp2_file;
         std::shared_ptr< llvm::IRBuilder<> > ir_builder;
         std::vector< std::shared_ptr< vm::vmp_rtn_t > > vmp_rtns;
 
-        void create_routine( void );
         void push( std::uint8_t byte_size, llvm::Value *input_val );
         llvm::Value *pop( std::uint8_t byte_size );
 

include/devirt_utils.hpp

@@ -1,4 +1,9 @@
 #include <vmprofiler.hpp>
+#include <coff/image.hpp>
+
+#define VM_ENTER_NAME "vmenter_"
+#define VM_EXIT_NAME "vmexit_"
+#define VM_RTN_NAME "rtn_"
 
 namespace devirt
 {
@@ -15,4 +20,12 @@ namespace devirt
             std::vector< std::pair< std::uint32_t, std::vector< vm::instrs::code_block_t > > > &virt_rtns,
             std::vector< std::uint8_t > &vmp2file );
     } // namespace util
+
+    /// <summary>
+    /// append devirtualized functions to the original binary... patches vm enter jmps to devirtualized code...
+    /// </summary>
+    /// <param name="obj">compiled obj file... generated by llvm...</param>
+    /// <param name="bin">original binary in a vector... this binary gets patched...</param>
+    /// <returns>returns true if new .devirt section has been appended and all linking was successful...</returns>
+    bool append( std::vector< std::uint8_t > &obj, std::vector< std::uint8_t > &bin );
 } // namespace devirt

include/vmp_rtn_t.hpp

@@ -1,6 +1,7 @@
 #pragma once
 #include <vmprofiler.hpp>
 
+#include "llvm/IR/InlineAsm.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
@@ -22,12 +23,14 @@ namespace vm
         llvm::Function *llvm_fptr;
         llvm::AllocaInst *flags, *stack;
         llvm::Module *llvm_module;
+        vmp2::v4::file_header *vmp2_file;
 
         std::uintptr_t rtn_begin;
         std::shared_ptr< llvm::IRBuilder<> > ir_builder;
         std::vector< llvm::AllocaInst * > virtual_registers;
         std::vector< std::pair< std::uintptr_t, llvm::BasicBlock * > > llvm_code_blocks;
         std::vector< vm::instrs::code_block_t > vmp2_code_blocks;
+
         void create_virtual_registers( void );
         void create_routine( void );
 
@@ -36,6 +39,7 @@ namespace vm
 
       public:
         explicit vmp_rtn_t( std::uintptr_t rtn_begin, std::vector< vm::instrs::code_block_t > vmp2_code_blocks,
-                            std::shared_ptr< llvm::IRBuilder<> > &ir_builder, llvm::Module *llvm_module );
+                            std::shared_ptr< llvm::IRBuilder<> > &ir_builder, llvm::Module *llvm_module,
+                            vmp2::v4::file_header *vmp2_file );
     };
 } // namespace vm

src/devirt_t.cpp

@@ -3,8 +3,8 @@
 
 namespace vm
 {
-    devirt_t::devirt_t( llvm::LLVMContext *llvm_ctx, llvm::Module *llvm_module )
+    devirt_t::devirt_t( llvm::LLVMContext *llvm_ctx, llvm::Module *llvm_module, vmp2::v4::file_header *vmp2_file )
-        : llvm_ctx( llvm_ctx ), llvm_module( llvm_module )
+        : llvm_ctx( llvm_ctx ), llvm_module( llvm_module ), vmp2_file( vmp2_file )
     {
         ir_builder = std::make_shared< llvm::IRBuilder<> >( *llvm_ctx );
     }
@@ -106,7 +106,8 @@ namespace vm
 
     llvm::Function *devirt_t::lift( std::uintptr_t rtn_begin, std::vector< vm::instrs::code_block_t > code_blocks )
     {
-        vmp_rtns.push_back( std::make_shared< vm::vmp_rtn_t >( rtn_begin, code_blocks, ir_builder, llvm_module ) );
+        vmp_rtns.push_back(
+            std::make_shared< vm::vmp_rtn_t >( rtn_begin, code_blocks, ir_builder, llvm_module, vmp2_file ) );
 
         auto &vmp_rtn = vmp_rtns.back();
         auto lifters = vm::lifters_t::get_instance();
@@ -154,8 +155,8 @@ namespace vm
         return ir_builder->CreateCall( popcount_intrinsic, { extended_bits } );
     }
 
-    llvm::Value *devirt_t::flags( llvm::Value *cf, llvm::Value *pf, llvm::Value *af, llvm::Value *zf,
+    llvm::Value *devirt_t::flags( llvm::Value *cf, llvm::Value *pf, llvm::Value *af, llvm::Value *zf, llvm::Value *sf,
-                                          llvm::Value *sf, llvm::Value *of )
+                                  llvm::Value *of )
     {
         auto shifted_pf = ir_builder->CreateShl( pf, 2, "shifted_pf", true, true );
         auto shifted_af = ir_builder->CreateShl( llvm::ConstantInt::get( llvm::IntegerType::get( *llvm_ctx, 64 ), 0 ),

src/devirt_utils.cpp

@@ -51,4 +51,18 @@ namespace devirt
             return true;
         }
     } // namespace util
+
+    bool append( std::vector< std::uint8_t > &obj, std::vector< std::uint8_t > &bin )
+    {
+        if ( obj.empty() || bin.empty() )
+            return false;
+
+        // TODO: append a new section to the pe file...
+        // put the entire obj file into this new section...
+        // fix up the vmenter functions to jmp to their respective devirtualized routines...
+        //
+        // move relocation directory to the end of the module and append new relocations...
+        // export all of the new devirtualized/llvm generated functions...
+        return true;
+    }
 } // namespace devirt

src/lifters/vmexit.cpp

@@ -2,10 +2,60 @@
 
 namespace vm
 {
-    lifters_t::lifter_callback_t lifters_t::vmexit =
+    lifters_t::lifter_callback_t lifters_t::vmexit = [ & ]( vm::devirt_t *rtn,
-        [ & ]( vm::devirt_t *rtn, const vm::instrs::code_block_t &vm_code_block, const vm::instrs::virt_instr_t &vinstr,
+                                                            const vm::instrs::code_block_t &vm_code_block,
+                                                            const vm::instrs::virt_instr_t &vinstr,
                                                             llvm::IRBuilder<> *ir_builder ) {
+        std::stringstream rtn_name;
+        llvm::Function *exit_func = nullptr;
+
+        rtn_name << "vmexit_" << std::hex << vinstr.trace_data.vm_handler_rva + rtn->vmp2_file->image_base;
+        if ( !( exit_func = rtn->llvm_module->getFunction( rtn_name.str() ) ) )
+        {
+            auto vmexit_func_type = llvm::FunctionType::get(
+                ir_builder->getVoidTy(), llvm::PointerType::getInt8PtrTy( ir_builder->getContext() ) );
+
+            exit_func = llvm::Function::Create( vmexit_func_type, llvm::GlobalValue::LinkageTypes::ExternalLinkage,
+                                                rtn_name.str().c_str(), *rtn->llvm_module );
+
+            auto entry_block = llvm::BasicBlock::Create( ir_builder->getContext(), "", exit_func );
+            auto vmexit_handler_addr = reinterpret_cast< std::uintptr_t >( rtn->vmp2_file ) +
+                                       rtn->vmp2_file->module_offset + vinstr.trace_data.vm_handler_rva;
+
+            zydis_routine_t vmexit_handler;
+            vm::util::flatten( vmexit_handler, vmexit_handler_addr );
+            vm::util::deobfuscate( vmexit_handler );
+
+            std::string asm_str( "mov rsp, rcx; " );
+            ZydisFormatter formatter;
+            ZydisFormatterInit( &formatter, ZYDIS_FORMATTER_STYLE_INTEL );
+
+            for ( const auto &instr_data : vmexit_handler )
+            {
+                if ( instr_data.instr.mnemonic == ZYDIS_MNEMONIC_POP ||
+                     instr_data.instr.mnemonic == ZYDIS_MNEMONIC_POPFQ )
+                {
+                    char buffer[ 256 ];
+                    ZydisFormatterFormatInstruction( &formatter, &instr_data.instr, buffer, sizeof( buffer ), 0ull );
+                    asm_str.append( buffer ).append( "; " );
+                }
+            }
+
+            asm_str.append( "ret" );
+            auto ib = ir_builder->GetInsertBlock();
+            ir_builder->SetInsertPoint( entry_block );
+
+            auto exit_stub = llvm::InlineAsm::get( llvm::FunctionType::get( ir_builder->getVoidTy(), false ), asm_str,
+                                                   "", false, false, llvm::InlineAsm::AD_Intel );
+
+            ir_builder->CreateCall( exit_stub );
+            ir_builder->CreateRetVoid();
+            ir_builder->SetInsertPoint( ib );
+        }
+
         auto &vmp_rtn = rtn->vmp_rtns.back();
-            ir_builder->CreateRet( ir_builder->CreateLoad( vmp_rtn->stack ) );
+        auto stack_ptr = ir_builder->CreateLoad( vmp_rtn->stack );
+        ir_builder->CreateCall( exit_func, stack_ptr );
+        ir_builder->CreateRet( stack_ptr );
     };
 }

src/main.cpp

@@ -26,7 +26,7 @@ int main( int argc, const char *argv[] )
         return 0;
     }
 
-    std::vector< std::uint8_t > vmp2file;
+    std::vector< std::uint8_t > vmp2file, bin;
     const auto umtils = xtils::um_t::get_instance();
 
     if ( !umtils->open_binary_file( parser.get< std::string >( "vmp2file" ), vmp2file ) )
@@ -35,6 +35,12 @@ int main( int argc, const char *argv[] )
         return -1;
     }
 
+    if ( !umtils->open_binary_file( parser.get< std::string >( "bin" ), bin ) )
+    {
+        std::printf( "[!] failed to open original binary file...\n" );
+        return -1;
+    }
+
     const auto file_header = reinterpret_cast< vmp2::v4::file_header * >( vmp2file.data() );
     std::vector< std::pair< std::uint32_t, std::vector< vm::instrs::code_block_t > > > virt_rtns;
 
@@ -48,7 +54,7 @@ int main( int argc, const char *argv[] )
     llvm::Module llvm_module( "VMProtect 2 Devirtualization", llvm_ctx );
 
     std::vector< std::uint8_t > compiled_obj;
-    vm::devirt_t vmp_devirt( &llvm_ctx, &llvm_module );
+    vm::devirt_t vmp_devirt( &llvm_ctx, &llvm_module, file_header );
 
     for ( auto &[ vm_enter_offset, vmp2_code_blocks ] : virt_rtns )
     {
@@ -71,7 +77,9 @@ int main( int argc, const char *argv[] )
 
     vmp_devirt.compile( compiled_obj );
     std::printf( "> compiled all routines... compiled obj size = %d\n", compiled_obj.size() );
-
     std::ofstream( "devirt.o", std::ios::binary )
         .write( reinterpret_cast< const char * >( compiled_obj.data() ), compiled_obj.size() );
+
+    devirt::append( compiled_obj, bin );
+    std::ofstream( "devirt.exe", std::ios::binary ).write( reinterpret_cast< const char * >( bin.data() ), bin.size() );
 }

src/vmp_rtn_t.cpp

@@ -3,9 +3,10 @@
 namespace vm
 {
     vmp_rtn_t::vmp_rtn_t( std::uintptr_t rtn_begin, std::vector< vm::instrs::code_block_t > vmp2_code_blocks,
-                          std::shared_ptr< llvm::IRBuilder<> > &ir_builder, llvm::Module *llvm_module )
+                          std::shared_ptr< llvm::IRBuilder<> > &ir_builder, llvm::Module *llvm_module,
+                          vmp2::v4::file_header *vmp2_file )
         : ir_builder( ir_builder ), vmp2_code_blocks( vmp2_code_blocks ), rtn_begin( rtn_begin ),
-          llvm_module( llvm_module )
+          llvm_module( llvm_module ), vmp2_file( vmp2_file )
     {
         create_routine();
         // create virtual registers in the first code block...
@@ -45,5 +46,53 @@ namespace vm
                 { vmp2_code_block.vip_begin,
                   llvm::BasicBlock::Create( ir_builder->getContext(), blk_name.str().c_str(), llvm_fptr ) } );
         }
+
+        zydis_routine_t vm_enter;
+        auto vm_enter_addr = reinterpret_cast< std::uintptr_t >( vmp2_file ) + vmp2_file->module_offset +
+                             ( rtn_begin - vmp2_file->image_base );
+
+        vm::util::flatten( vm_enter, vm_enter_addr );
+        vm::util::deobfuscate( vm_enter );
+
+        std::string asm_str;
+        ZydisFormatter formatter;
+        ZydisFormatterInit( &formatter, ZYDIS_FORMATTER_STYLE_INTEL );
+
+        for ( const auto &instr_data : vm_enter )
+        {
+            if ( instr_data.instr.mnemonic == ZYDIS_MNEMONIC_PUSH )
+            {
+                if ( instr_data.instr.operands[ 0 ].type == ZYDIS_OPERAND_TYPE_MEMORY )
+                {
+                    asm_str.append( "mov rax, 0x1000000000000000; push rax; mov rax, 0x1000000000000000; push rax; " );
+                    break;
+                }
+
+                char buffer[ 256 ];
+                ZydisFormatterFormatInstruction( &formatter, &instr_data.instr, buffer, sizeof( buffer ), 0ull );
+                asm_str.append( buffer ).append( "; " );
+            }
+            else if ( instr_data.instr.mnemonic == ZYDIS_MNEMONIC_PUSHFQ )
+            {
+                asm_str.append( "pushfq; " );
+            }
+        }
+
+        rtn_name.str( "" );
+        asm_str.append( "mov rcx, rsp; sub rsp, 0xA00; " );
+
+        rtn_name << "vmenter_" << std::hex << rtn_begin;
+        auto entry_func = llvm::Function::Create( llvm::FunctionType::get( ir_builder->getVoidTy(), false ),
+                                                  llvm::GlobalValue::LinkageTypes::ExternalLinkage,
+                                                  rtn_name.str().c_str(), *llvm_module );
+
+        auto entry_block = llvm::BasicBlock::Create( ir_builder->getContext(), "", entry_func );
+        ir_builder->SetInsertPoint( entry_block );
+
+        auto entry_stub = llvm::InlineAsm::get( llvm::FunctionType::get( ir_builder->getVoidTy(), false ), asm_str, "",
+                                                false, false, llvm::InlineAsm::AD_Intel );
+
+        ir_builder->CreateCall( entry_stub );
+        ir_builder->CreateRetVoid();
     }
 } // namespace vm