ansicon/injdll32.c

199 lines
6.0 KiB
C
Raw Normal View History

2010-11-08 15:31:01 +10:00
/*
Inject code into the target process to load our DLL. The target thread
should be suspended on entry; it remains suspended on exit.
Initially I used the "stack" method of injection. However, this fails
when DEP is active, since that doesn't allow code to execute in the stack.
To overcome this I used the "CreateRemoteThread" method. However, this
would fail with Wselect, a program to assist batch files. Wselect runs,
but it has no output. As it turns out, removing the suspended flag would
make Wselect work, but it caused problems with everything else. So now I
allocate a section of memory and change the context to run from there. At
first I had an event to signal when the library was loaded, then the memory
was released. However, that wouldn't work with -p and CMD.EXE (4NT v8
worked fine). Since it's possible the DLL might start a process suspended,
I've decided to simply keep the memory.
*/
2010-12-12 21:58:35 +10:00
#include "ansicon.h"
2010-11-08 15:31:01 +10:00
#ifdef _WIN64
#ifndef WOW64_CONTEXT_ALL
2010-11-08 15:31:01 +10:00
#include "wow64.h"
TWow64GetThreadContext Wow64GetThreadContext;
TWow64SetThreadContext Wow64SetThreadContext;
#define IMPORT_WOW64
2010-11-15 21:51:38 +10:00
#endif
2010-11-08 15:31:01 +10:00
#define CONTEXT WOW64_CONTEXT
#undef CONTEXT_CONTROL
#define CONTEXT_CONTROL WOW64_CONTEXT_CONTROL
#define GetThreadContext Wow64GetThreadContext
#define SetThreadContext Wow64SetThreadContext
#define MakeVA( cast, offset ) (cast)((DWORD_PTR)base + (DWORD)(offset))
extern DWORD LLW32;
LPVOID base;
int export_cmp( const void* a, const void* b )
{
return strcmp( (LPCSTR)a, MakeVA( LPCSTR, *(const PDWORD)b ) );
}
/*
Get the address of the 32-bit LoadLibraryW function from 64-bit code. This
was originally done via executing a helper program (ANSI-LLW.exe), but for
some reason, virus scanners really didn't like it (even when I rewrote it in
assembly so it was 1024 bytes and literally two instructions, virustotal
still had three scanners complaining). Now I do it the "hard" way - load the
32-bit kernel32.dll directly and search the exports. Even worse, it seems
Wow64 loads kernel32.dll at a different base address each boot (at least, I
hope it only changes each boot). Fortunately, loading the DLL as an image in
64-bit code seems to use the 32-bit address.
*/
BOOL get_LLW32( void )
{
HMODULE kernel32;
TCHAR buf[MAX_PATH];
UINT len;
PIMAGE_DOS_HEADER pDosHeader;
PIMAGE_NT_HEADERS32 pNTHeader;
PIMAGE_EXPORT_DIRECTORY pExportDir;
PDWORD fun_table, name_table;
PWORD ord_table;
PDWORD pLLW;
len = GetSystemWow64Directory( buf, MAX_PATH );
wcscpy( buf + len, L"\\kernel32.dll" );
// MinGW-w64 has a typo, calling it LINRARY.
kernel32 = LoadLibraryEx( buf, NULL, 0x20/*LOAD_LIBRARY_AS_IMAGE_RESOURCE*/ );
if (kernel32 == NULL)
{
DEBUGSTR( 1, L"Unable to load 32-bit kernel32.dll!" );
return FALSE;
}
// The handle uses low bits as flags, so strip 'em off.
base = (LPVOID)((DWORD_PTR)kernel32 & 0xFFFF0000);
// Tests to make sure we're looking at a module image (the 'MZ' header)
pDosHeader = (PIMAGE_DOS_HEADER)base;
if (pDosHeader->e_magic != IMAGE_DOS_SIGNATURE)
{
DEBUGSTR( 1, L"Image has no DOS header!" );
return FALSE;
}
// The MZ header has a pointer to the PE header
pNTHeader = MakeVA( PIMAGE_NT_HEADERS32, pDosHeader->e_lfanew );
// One more test to make sure we're looking at a "PE" image
if (pNTHeader->Signature != IMAGE_NT_SIGNATURE)
{
DEBUGSTR( 1, L"Image has no NT header!" );
return FALSE;
}
// We now have a valid pointer to the module's PE header.
// Get a pointer to its exports section.
pExportDir = MakeVA( PIMAGE_EXPORT_DIRECTORY,
pNTHeader->OptionalHeader.
DataDirectory[IMAGE_DIRECTORY_ENTRY_EXPORT].
VirtualAddress );
fun_table = MakeVA( PDWORD, pExportDir->AddressOfFunctions );
name_table = MakeVA( PDWORD, pExportDir->AddressOfNames );
ord_table = MakeVA( PWORD, pExportDir->AddressOfNameOrdinals );
2010-11-08 15:31:01 +10:00
pLLW = bsearch( "LoadLibraryW", name_table, pExportDir->NumberOfNames,
sizeof(DWORD), export_cmp );
if (pLLW == NULL)
{
DEBUGSTR( 1, L"Could not find LoadLibraryW!" );
return FALSE;
}
LLW32 = MakeVA( DWORD, fun_table[ord_table[pLLW - name_table]] );
FreeLibrary( kernel32 );
return TRUE;
}
#else
DWORD LLW32;
#endif
2010-11-08 15:31:01 +10:00
2010-11-15 21:51:38 +10:00
void InjectDLL32( LPPROCESS_INFORMATION ppi, LPCTSTR dll )
2010-11-08 15:31:01 +10:00
{
CONTEXT context;
DWORD len;
LPVOID mem;
DWORD mem32;
#define CODESIZE 20
Exclude modules from being hooked; hook only selected GUI programs. Added environment variable ANSICON_EXC to specify modules that should not be hooked. This should work around the nvd3d9wrap.dll issue. Since it helps to know what the modules are, logging is now always available, controlled by -l or ANSICON_LOG. A side-effect caused debugstr.c to move to util.c. GUI programs are once again not hooked, unless run by "ansicon" directly or in the ANSICON_GUI environment variable. Since not hooking still leaves ANSICON in the environment, created ANSICON_VER as a dynamic-only variable, which can also serve as a version check. Due to an email requesting a reverse video option, realised I always take the current attributes as default. This means if you turned on reverse and ran a program, it would take the reverse as its default. Created ANSICON_DEF variable to explicitly set the default attribute, using the current if it doesn't exist. The reverse video option is done via a "negative" attribute (e.g. "-m-f0" is reversed black on white, meaning you'll get white on black, with foreground sequences changing the background). (The difference from "\e[7m" is that it won't be reset on "\e[m".) A child program will inherit the parent's modes (but not shift); the parent will read the child's modes on exit (but not unload). The exception is "ansicon", which will always start with the default modes and leave the parent unchanged. Improved the AutoRun entry, only running "ansicon" if ANSICON_VER doesn't exist. The "ansicon" command is always first. Stopped -u implying -p; return the program's exit code; don't restore the original color when just using -p; output error messages to stderr.
2011-12-14 20:53:51 +10:00
BYTE code[CODESIZE+TSIZE(MAX_PATH)];
2010-11-15 21:51:38 +10:00
union
{
PBYTE pB;
PDWORD pL;
} ip;
2010-11-08 15:31:01 +10:00
Exclude modules from being hooked; hook only selected GUI programs. Added environment variable ANSICON_EXC to specify modules that should not be hooked. This should work around the nvd3d9wrap.dll issue. Since it helps to know what the modules are, logging is now always available, controlled by -l or ANSICON_LOG. A side-effect caused debugstr.c to move to util.c. GUI programs are once again not hooked, unless run by "ansicon" directly or in the ANSICON_GUI environment variable. Since not hooking still leaves ANSICON in the environment, created ANSICON_VER as a dynamic-only variable, which can also serve as a version check. Due to an email requesting a reverse video option, realised I always take the current attributes as default. This means if you turned on reverse and ran a program, it would take the reverse as its default. Created ANSICON_DEF variable to explicitly set the default attribute, using the current if it doesn't exist. The reverse video option is done via a "negative" attribute (e.g. "-m-f0" is reversed black on white, meaning you'll get white on black, with foreground sequences changing the background). (The difference from "\e[7m" is that it won't be reset on "\e[m".) A child program will inherit the parent's modes (but not shift); the parent will read the child's modes on exit (but not unload). The exception is "ansicon", which will always start with the default modes and leave the parent unchanged. Improved the AutoRun entry, only running "ansicon" if ANSICON_VER doesn't exist. The "ansicon" command is always first. Stopped -u implying -p; return the program's exit code; don't restore the original color when just using -p; output error messages to stderr.
2011-12-14 20:53:51 +10:00
len = TSIZE(lstrlen( dll ) + 1);
if (len > TSIZE(MAX_PATH))
2010-11-08 15:31:01 +10:00
return;
if (LLW32 == 0)
2010-11-08 15:31:01 +10:00
{
#ifdef _WIN64
if (!get_LLW32())
return;
#else
LLW32 = (DWORD)GetProcAddress( GetModuleHandleA( "kernel32.dll" ),
"LoadLibraryW" );
#endif
}
#ifdef IMPORT_WOW64
if (Wow64GetThreadContext == 0)
{
2010-11-08 15:31:01 +10:00
#define GETPROC( proc ) proc = (T##proc)GetProcAddress( hKernel, #proc )
HMODULE hKernel = GetModuleHandleA( "kernel32.dll" );
2010-11-08 15:31:01 +10:00
GETPROC( Wow64GetThreadContext );
GETPROC( Wow64SetThreadContext );
// Assume if one is defined, so is the other.
if (Wow64GetThreadContext == 0)
2010-12-12 21:58:35 +10:00
{
Exclude modules from being hooked; hook only selected GUI programs. Added environment variable ANSICON_EXC to specify modules that should not be hooked. This should work around the nvd3d9wrap.dll issue. Since it helps to know what the modules are, logging is now always available, controlled by -l or ANSICON_LOG. A side-effect caused debugstr.c to move to util.c. GUI programs are once again not hooked, unless run by "ansicon" directly or in the ANSICON_GUI environment variable. Since not hooking still leaves ANSICON in the environment, created ANSICON_VER as a dynamic-only variable, which can also serve as a version check. Due to an email requesting a reverse video option, realised I always take the current attributes as default. This means if you turned on reverse and ran a program, it would take the reverse as its default. Created ANSICON_DEF variable to explicitly set the default attribute, using the current if it doesn't exist. The reverse video option is done via a "negative" attribute (e.g. "-m-f0" is reversed black on white, meaning you'll get white on black, with foreground sequences changing the background). (The difference from "\e[7m" is that it won't be reset on "\e[m".) A child program will inherit the parent's modes (but not shift); the parent will read the child's modes on exit (but not unload). The exception is "ansicon", which will always start with the default modes and leave the parent unchanged. Improved the AutoRun entry, only running "ansicon" if ANSICON_VER doesn't exist. The "ansicon" command is always first. Stopped -u implying -p; return the program's exit code; don't restore the original color when just using -p; output error messages to stderr.
2011-12-14 20:53:51 +10:00
DEBUGSTR( 1, L"Failed to get pointer to Wow64GetThreadContext.\n" );
2010-11-08 15:31:01 +10:00
return;
2010-12-12 21:58:35 +10:00
}
2010-11-08 15:31:01 +10:00
}
#endif
2010-11-08 15:31:01 +10:00
CopyMemory( code + CODESIZE, dll, len );
len += CODESIZE;
context.ContextFlags = CONTEXT_CONTROL;
GetThreadContext( ppi->hThread, &context );
mem = VirtualAllocEx( ppi->hProcess, NULL, len, MEM_COMMIT,
PAGE_EXECUTE_READWRITE );
mem32 = (DWORD)(DWORD_PTR)mem;
ip.pB = code;
*ip.pB++ = 0x68; // push eip
*ip.pL++ = context.Eip;
*ip.pB++ = 0x9c; // pushf
*ip.pB++ = 0x60; // pusha
2010-11-13 14:49:27 +10:00
*ip.pB++ = 0x68; // push L"path\to\ANSI32.dll"
2010-11-08 15:31:01 +10:00
*ip.pL++ = mem32 + CODESIZE;
2010-11-13 14:49:27 +10:00
*ip.pB++ = 0xe8; // call LoadLibraryW
*ip.pL++ = LLW32 - (mem32 + (DWORD)(ip.pB+4 - code));
2010-11-08 15:31:01 +10:00
*ip.pB++ = 0x61; // popa
*ip.pB++ = 0x9d; // popf
*ip.pB++ = 0xc3; // ret
WriteProcessMemory( ppi->hProcess, mem, code, len, NULL );
FlushInstructionCache( ppi->hProcess, mem, len );
context.Eip = mem32;
SetThreadContext( ppi->hThread, &context );
}