Password SPY++

• Download demo project - 141 Kb
• Download source - 371 Kb

Figure 1: SuperPasswordSpy++ Peeking a Hotmail "Forget Your Password" Page

Figure 2: SuperPasswordSpy++ Peeking a Password Edit inside IIS 5.0 on WinXP

Introduction

If you need a tool to peek not only the password edit boxes on different programs, but also the **** password input field on a homepage residing inside Internet Explorer, here it is — SuperPasswordSpy++.

With Windows Hooks, it is not difficult to peek into a remote process' password edit, but how about the password input field inside a homepage, say, in an Internet Explorer window? The password input field, obviously, is not a window; you have to turn to the IHTMLDocument2 interface to enumerate the inside stuff and extract the password. This program will show you how to do it with both the password edit common control and the password input field inside IE. Above are the screenshot of the program running, peeking the password in both cases.

(This article is transferred from CodeGuru posted in Jan 2003.)

Architecture

After the user begins to drag the magnifying glass around on the screen, the program captures the mouse and keeps track of the mouse move message. Whenever the mouse moves onto a new window, it checks the window's class name and window style to decide whether it is a password edit or a web browser control. In the latter case, a hook DLL has to be injected into the web browser control immediately to decide whether it contains a password input field.

Please note: here we have two options to implement the hook:

• The first way is setting a WH_GETMESSAGE hook, which was done by Mr. Brian Friesen in his article PasswordSpy. This hook DLL will intercept the posted message and a synchronization object (mutex, event, etc.) will be applied inside it. Please refer to Figure 3:

  

  Figure 3: General Spy Program's Hook Architecture

  There are five steps involved:

  1. The Spy program injects a hook DLL into the target program.
  2. The Spy program posts a user message to the target program that will be intercepted by the DLL.
  3. The DLL intercepts the user message and reads the password edit content.
  4. The DLL sends the data back to the Spy program.
  5. The Spy program gets the data and un-hooks the DLL.

  Because the Spy program posts the user message, step 2 will not "block" and the Spy program does not know when step 4 will happen. Usually, this kind of a Spy program uses WM_COPYDATA to transfer byte data; it is a "Send" message. Okay, here is the tough thing: The user may move the magnifying glass here and there, and let's imagine the following case. The Spy program finds a password edit, injects the DLL into the target program, posts the user message, and suddenly the user moves the magnifying glass to another password edit, so the Spy program has to un-hook the DLL out of the old target program, hook into the new target program, and post the user message. Unfortunately, the WM_COPYDATA from the old target program has queued into the Spy program's message queue, and if you do not append the target window handle information to the WM_COPYDATA, you cannot tell the data of the current target program from the old target program...

• The second option is setting the WH_CALLWNDPROC hook. It's all the same as the first option, except in step 2 we send the message instead of posting it to the target program. The hook DLL will intercept the user message and does the password reading, and call SendMessage with WM_COPYDATA to pass the data to the Spy program. In one word, step 2 is blocked until steps 3 and 4 are finished; so, after step 2 is finished, we can do step 5 directly. In this way, the code will be much simpler than the first way. It is just like blocked socket and non-blocked socket programming.

Sure, the second way has its disadvantage. Consider the following case: there are two password edits on the same window and the user must leave one to enter another. In the first way, we can check the two edits are on the same thread, so we can hook the DLL only once. In the second way, we hook twice which means (ignorable here in this program) a performance penalty.

Implementation Description

1. How to get IHTMLDocument from the browser control window handle:

   (With ref. from MSDN KB Q249232 HOWTO: Get IHTMLDocument2 from a HWND)

   BOOL HWnd2HtmlDocument()
   {
      CoUninitialize();
      HINSTANCE hInst = ::LoadLibrary( _T("OLEACC.DLL") );
      if ( hInst == NULL ) return FALSE;
      LRESULT lRes;
      UINT nMsg = ::RegisterWindowMessage( _T("WM_HTML_GETOBJECT") );
      ::SendMessageTimeout( g_hTarget, nMsg,
             0L, 0L, SMTO_ABORTIFHUNG, 1000, (DWORD*)&lRes );
      LPFNOBJECTFROMLRESULT pfObjectFromLresult =
           (LPFNOBJECTFROMLRESULT)::GetProcAddress( hInst, "ObjectFromLresult");
      if ( pfObjectFromLresult == NULL )
      {
         ::FreeLibrary( hInst );
         CoUninitialize();
         return FALSE;
      }
      WCHAR strDoc[] = L"{626fc520-a41e-11cf-a731-00a0c9082637}";
      //IID_IHTMLDocument2 CLSID
      CLSID uuidDoc;
      HRESULT hrDoc = CLSIDFromString((LPOLESTR)strDoc,
         &uuidDoc //IID_IHTMLDocument2
      );
      if(!SUCCEEDED(hrDoc))
      {
         ::FreeLibrary( hInst );
         CoUninitialize();
         return FALSE;
      }
      HRESULT hr = (*pfObjectFromLresult)( lRes, uuidDoc,
         //IID_IHTMLDocument,
         0, (void**)&g_lpHTMLDocument2);
      if ( SUCCEEDED(hr) )
      {
         //OK, We Get Here Successfully
      }
      else
      {
         ::FreeLibrary( hInst );
         CoUninitialize();
         return FALSE;
      }
      ::FreeLibrary( hInst );
      CoUninitialize();
      return TRUE;
   }

   There is something I need to explain here: g_hTarget is the handle of the browser, and its class name is "Internet Explorer_Server". Usually, it is not a problem with MS Internet Explorer, which always starts to navigate somewhere when it starts. But, in some applications, when hosting a Web Browser with ActiveX, before the browser navigates, the "Internet Explorer_Server" window does not exist. Let me give you an example:

   

   Figure 4: Spy++'s Screenshot on a Dialog Hosting Two MS Web Browser ActiveXs

   It is a screenshot from Spy++. There are two browsers on this dialog (00060294); only browser 000202D6 navigates to some URL. You can make an experiment by putting a Web Browser ActiveX on a dialog, and checking this with Spy++.

2. Whether current homepage contains a password input field:

   DWORD CheckHtmlDocument()
   //Ret: 0 -- No Password Input; Else -- Password Input Number
   {
       MSHTML::IHTMLElementCollection *pForm;
       HRESULT hr = g_lpHTMLDocument2->get_all(&pForm);
       //g_lpHTMLDocument2 is a pointer of IHTMLDocument2
       if(FAILED(hr)) return 0;
       long len;
       pForm->get_length(&len); //How many elements on this form?
       DWORD dwRet = 0;
   
       for(int i = 0; i < len; i++)
       {
           LPDISPATCH lpItem = pForm->item(CComVariant(i), CComVariant(i));
           MSHTML::IHTMLInputElementPtr lpInput;
           HRESULT hr = lpItem->QueryInterface(&lpInput);
           //Is it a input field?
   
           if(FAILED(hr)) continue;
           _bstr_t type(_T("password"));
   
           if(lpInput->Gettype() == type) //Check Field Type
           {
                //_bstr_t x = lpInput->Getvalue();
                //If you want its string
                dwRet++;
           }
           lpItem->Release(); //Remember To Release this!
           lpItem = NULL;
       }
       pForm->Release();
       pForm = NULL;
       return dwRet;
   }

3. Extract password from current homepage's password input field:

   _bstr_t x = lpInput->Getvalue(); //And you go!
   
   LPCTSTR lpWhatEver = (LPCTSTR)x;
   
   //Do something with the password here

Special Note for SuperPasswordSpy++

Once again, SuperPasswordSpy++ is for WinNT/2K/XP only and IE 4.0+ is needed; it is a Unicode-based program.

Interested readers can learn how to track the window and hook remote processes by studying the source code of SuperPasswordSpy++. It is highly recommended that the readers study MS Platform SDK's Spy sample and PasswordSpy by Mr. Brian Friesen. The author dislikes "recreating the wheel" things, and borrows mouse tracking code from the Spy sample and a function "SmallestWindowFromPoint", plus the resource file from PasswordSpy. And, if the reader has questions about the hook technique, please read Mr. Brian Friesen's article; it is already detailed enough. In addition, readers can read chapters 19 and 20 of Programming Applications for MS Windows, 4^th Edition written by Jeffrey Richter, published by Microsoft Press, 1999, for more detailed information about shared sections in DLLs.

Well, besides, something unclear with SuperPasswordSpy++'s implementation does exist:

1. Criteria of identifying a password edit

   BOOL IsPasswordEdit(HWND hWnd)
   {
       TCHAR szClassName[64];
       int nRet = GetClassName(hWnd, szClassName, 64);
       if(nRet == 0) return FALSE;
       szClassName[nRet] = 0;
       if(::lstrcmp(szClassName, _T("Edit")) != 0 &&
            ::lstrcmp(szClassName, _T("TEdit")) != 0 &&
            ::lstrcmp(szClassName, _T("ThunderTextBox"))
            != 0 ) return FALSE;
       //Here, is it OK?
   
       DWORD dw = ::GetWindowLong(hWnd,GWL_STYLE);
       dw &= ES_PASSWORD;
       if(dw == ES_PASSWORD)
           return TRUE;
       return FALSE;
   }

   Above is the implementation of SuperPasswordSpy++. But, in a tool sample called Eureka (note this tool is for Win9X/Me), the author identifies a password edit if its class name is "TEDIT", "IRIS.PASSWORD", or "EDIT". I played with Delphi and Borland C++ years ago, and I know TxxxClass is Borland's naming convention. But I don't have Delphi handy, so if anyone can help check this, please comment below. As to the "ThunderTextBox", it is created by Visual Basic. Thanks in advance. And I will add multiple name matching if necessary. Well, of course, it is impossible to cover all cases, for the class name can be changed by developers arbitrarily (for example, who knows what is the password edit name in Visual Studio 2003?). If it is such a case, please modify the source of my SuperPasswordSpy++ yourself.

2. When multiple frame sets exist inside an HTML document

   The current (Dec. 2002) version of SuperPasswordSpy++ assumes there is only one frame inside an HTML document if it contains a password field. It is fine with most homepages with passwords (for example, MSN Hotmail). But, to make it perfect, I will add multiple frame set support in the next version of SuperPasswordSpy++. Please check this at my site later.

More Words of ******

1. Windows Logon Password: (Win2K only)

   It sounds a little crazy, but please look at the following figure:

   

   Figure 5: SuperPasswordSpy++'s Peeking "Change Your Password" on WinLogon Screen of Win2K Server

   I have only the Japanese Win2K server at hand, so there is some Japanese here, but I think you can see, that the window is obtained from "Ctrl+Alt-Del" and then pressing the "Change Password" button. PasswordSpy++ has been launched under the SYSTEM context on the Winlogon Desktop of this Win2K, and it works. I mean, it can read the password from the "Change Password" dialog. Oh, by the way, you will never know the current password because it is never shown.

   To conduct this experiment, you have to use some tool to launch a program on the Windows Winlogon Desktop. You can go here to get my "GUI RunAs" program. You have to choose Winlogon desktop, make the user name edit box blank to use SYSTEM identity, and remember you must have Administrator rights to do so. Follow the article's instructions and you may need to log off the current session once (only once) to enable some privileges if you do not already have them. Some readers sent me e-mail saying there is already a "Runas" command line tool in Windows OS. I know, but with this MS-brand RunAs command-line tool, you can not choose the Desktop to launch the program while my tool works. If you have problems with the tool (for example, the launched program's GUI is blocked), please launch the tool itself as SYSTEM first, and then launch the program you use.

   The last word: to get the screenshot, press "Print Screen" when you are on the Winlogon desktop, go to the default desktop, and paste it into MS Paint. You can also use the RunAs tool to launch MS Paint onto the Winlogon screen too, and do your job there without switching back and forth. And, you may find this RunAs GUI tool is a good way to launch a task manager as SYSTEM and kill some stubborn processes (including NT Service).

2. Anti-peeking edit and crack anti-peeking edit ( -* - = +)

   I noticed someone asked how to do anti-peeking. Well, it does not cost too much to implement anti-peeking. Following is a code example using MFC. First, derive an edit class from CEdit. My first idea was to override its PreTranslate, this way:

   BOOL CAntiPeekEdit::PreTranslateMessage(MSG* pMsg)
   //This does NOT work!
   {
       if(pMsg->message == WM_GETTEXT)
       {
           //Only Report Text When Passing a Fixed Length Buffer;
           if(pMsg->wParam == 1024) //The Number Only You Know
           {
           }
           else
           {
               ::lstrcpy((LPTSTR)(pMsg->lParam), _T("Nothing"));
               return TRUE;
           }
       }
       return CEdit::PreTranslateMessage(pMsg);
   }

   Unfortunately, the inside if clause will never be called. Why? MFC team guys know. Well, I have to turn to the virtual function WindowsProc; this time it works:

   LRESULT CAntiPeekEdit::WindowProc( UINT message,
      WPARAM wParam,
      LPARAM lParam)
   //This works!
   {
      if(message == WM_GETTEXT)
      {
          if(wParam == 1024) //The Number Only You Know
          {
              return CEdit::WindowProc(message, wParam, lParam);
          }
          else
          {
              ::lstrcpy((LPTSTR)(lParam), _T("Nothing"));
              //Insert Dummy Text Here To the Peeker
              return 7;
          }
      }
      return CEdit::WindowProc(message, wParam, lParam);
   }

   In your own program, when you want to get the password text, you must do the following:

   TCHAR sz[1024];
   ::SendMessage(hPasswordEdit, WM_GETTEXT, 1024, (LPARAM)sz);

   If other routines call to retrieve the password, WM_GETTEXTLENGTH will tell them the correct length, but when the correct length buffer is asked by our AntiPeekEdit, we know it is called from some un-secure source, so we can just send back junk. Well, you can also abandon WM_GETTEXT completely and use a WM_USER + 123 message to get the text out.

   Let's go back to how to counterstrike this kind of AntiPeekEdit. Well, we know the reason why we use hook DLL and query password from inside the remote process is because the Win2K password-style edit will not accept WM_GETTEXT from outside the process boundary. And, the above strategy replaces the standard Edit class Windows procedure with a user-defined one. So, how about replacing this user-defined procedure back to the standard Edit class Windows procedure when our lovely SuperPasswordSpy++ peeks it:

   HWND hParent = ::GetParent(g_hTarget);
   //g_hTarget is the password edit handle we are interested
   HWND hwndEdit = CreateWindow(
       _T("EDIT"), // predefined class
       NULL, // no window title
       WS_CHILD | WS_VISIBLE | WS_VSCROLL |
       ES_LEFT | ES_MULTILINE | ES_AUTOVSCROLL,
       0, 0, 0, 0, // set size in WM_SIZE message
       hParent, // parent window
       (HMENU)123,
       // edit control ID -- Note: Must be Unique Among Sibling Windows
       (HINSTANCE) GetWindowLong(g_hTarget, GWL_HINSTANCE),
       NULL); // pointer not needed
   //Get the standard Windows Procedure of Edit class
   LONG_PTR lpNewEdit = GetWindowLongPtr(hwndEdit, GWLP_WNDPROC);
   LONG_PTR lp = ::SetWindowLongPtr(g_hTarget, 
                   GWLP_WNDPROC,(LONG_PTR)lpNewEdit);
   
   //Password Fetch Here --- I only use this call in
   //SuperPasswordSpy++
   SendMessage(g_hTarget, WM_GETTEXT, sizeof(szBuffer) /
   sizeof(TCHAR), (LPARAM)szBuffer);
   
   //Reset the original Windows Procedure
   ::SetWindowLongPtr(g_hTarget, GWLP_WNDPROC, (LONG_PTR)lp);

   Please note the Control ID parameter when creating the fake edit; it must be unique among its siblings. Well, I use 123 as a placeholder here. You can write additional code to enumerate the sibling windows and get a unique ID, and remember to destroy the fake edit in the end.

   But, it is really an overkill in most cases, so I did not include the above code in SuperPasswordSpy++ to keep performance and stability high. But, once you really meet such an anti-peek password edit, uncomment the additional code in the SuperPasswordSpy++ source code, and keep an eye on keeping the fake edit control ID unique.

   If someone is mad asking if we have a way to anti- anti- anti-peeking, well, maybe you can add a global variable flag; before you fetch the password set this flag, and after reading, reset it.... Then why not use some algorithm to encode the text in WM_GETTEXT handler? Faint...

3. More peeking Spy tools...

   If you need to peek into MSN Messenger/Windows Messenger inside information and chat contents, please refer to my previous article MessengerSpy++. The following figure shows we can get a 100% fresh RTF text and the emotional icons from MSN Messenger (the right-side window). Note: it is for Win2K/XP and supports MSN Messenger 4.6, 4.7, and 5.0. And what's more, it can send text/icons to MSN Messenger and let Messenger send it to the other person.

   

   Figure 5: Mate Tool MessengerSpy++'s Peeking a MSN Messenger Chat on WinXP

4. WinXP "Change User Password" Control Panel applet

   Unfortunately, even it is inside "Internet Explorer_Server" (res://D:\WINDOWS\system32\nusrmgr.cpl/nusrmgr.hta), the password input field is inside an ActiveX with CLSID:A5064426-D541-11D4-9523-00B0D022CA64 in my English WinXP Professional. So, it is completely impossible to read the password from the ActiveX black box.

Acknowledgements

Thanks to the following article/code contributor on CodeGuru: Mr. Giancarlo Iovino for his HyperLink control (I fixed a few lines of code so it can go under Unicode now), and Mr. Brian Friesen for his article and tool PasswordSpy.

History

• 1.1 - 2003

  Support for IE multiple frame set password peeking.

• 1.0 - Dec 7, 2002

  First version.