So you can imagine how annoyed I was to come into work twice this week to the blue Windows logon screen. Every time this happens it takes me twenty minutes to figure out what I was doing the day before, what I have to do today, and where I stopped with my work. And really what made this so much more painful was that it happened without any advanced warning.

And that’s what got me thinking: could I block restart requests? I researched the Windows shut down process online and then went to work on a prototype. From what I read, calling ExitWindowsEx sends WM_QUERYENDSESSION to all top-level windows. Applications that are not ready to shut down should return false. I figured the best strategy was to install a system-wide hook and filter the message.

Initially I attempted to capture WM_QUERYENDSESSION with the WH_GETMESSAGE hook and replace it with WM_NULL, but trial-and-error revealed that it’s sent through SendMessage and not posted to the window’s queue. This meant that I couldn’t filter out the message.

I switched to WH_CALLWNDPROC and was able to capture the message, but not actually modify it. Since my DLL is memory-mapped into the local process space, it seemed like the only way to filter the message was to create a new WindowProc function that handles WM_QUERYENDSESSION and always returns false. Then inside the hook procedure, I could intercept the message and call SetWindowLong to replace the window’s message procedure.

This demonstrates the basic concept:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22

LRESULT CALLBACK CallWndProc( int nCode, WPARAM wParam, LPARAM lParam )
{
  if( nCode == HC_ACTION ) {
      CWPSTRUCT *msg = (CWPSTRUCT*)lParam;
 
      /* hijack the window proc when we see a shut down message */
      if( msg->message == WM_QUERYENDSESSION )
          oldwndproc = SetWindowLong(msg->hwnd, GWL_WNDPROC, (DWORD)WindowProc);
  }
 
  return CallNextHookEx(g_callwndhk, nCode, wParam, lParam);
}
 
LRESULT CALLBACK WindowProc( HWND hWnd, UINT uiMessage, 
        WPARAM wParam, LPARAM lParam )
{
  /* intercept shut down messages */
  if( uiMessage == WM_QUERYENDSESSION )
      return 0;
 
  return DefWindowProc(hWnd, uiMessage, wParam, lParam);
}

When my little application starts, it calls SetProcessShutdownParameters with level 0×4FF to increase the chances of trapping the message first. I figured this was a good idea since I know its WindowProc function can be safely hijacked. Now when Windows sends WM_QUERYENDSESSION the response is always “NO!”. The added exclamation there is a call to AbortSystemShutdown, which is probably unnecessary but I do it just to be safe. Also, I added an alert message box to warn me when a reboot is triggered.

I’m sort of amazed this actually worked. Some day I’ll test it against InitiateSystemShutdown and ExitWindowsEx with EWX_FORCE to see how it holds up. Interestingly, Windows Vista/7 provides ShutdownBlockReasonCreate for seemingly outright blocking shut down attempts.

You can obtain the sources to this project here.

I use Spark on my workstation to connect to the company’s internal IM server. The application works alright, but the contacts window always appears in the taskbar. So I started to think about ways I could programmatically solve my problem.

I could set the WS_EX_TOOLWINDOW style on the window, but that would alter the window’s appearance. What I really wanted was a way to tell Windows to remove the tab. A quick search on Google turned up the answer: use COM to create an instance of ITaskbarList. The interface has the function ITaskbarList::DeleteTab() which takes a window handle. Perfect!

Now I just needed to get the window’s handle. FindWindow() would have worked, but that meant hard-coding my username into the program. I felt a more elegant solution was to enumerate all of the windows, and look for the one with the right class and title prefix.

Since the tab would reappear every time I brought the contact list window to the foreground, I ended up wrapping my fix with a loop and a timer. Here’s the finished product:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69

/* NoSpark.cpp - Hides the Spark contacts window tab in the taskbar */
 
#include <windows.h>
#include <Shobjidl.h>
 
BOOL CALLBACK EnumWindowsProc(HWND, LPARAM);
 
int WINAPI WinMain( HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpszCmdLine, int nCmdShow )
{
	HANDLE htmr = CreateWaitableTimer(NULL, TRUE, L"CheckSpark");
	LARGE_INTEGER lidt;
	ITaskbarList* ptl;
 
	__int64 qwdt = -60 * 10000000; // 1 minute
	lidt.LowPart = (DWORD) (qwdt & 0xFFFFFFFF);
	lidt.HighPart = (LONG) (qwdt >> 32);
 
	while( TRUE ) {
		SetWaitableTimer(htmr, &lidt, 0, NULL, NULL, FALSE);
		WaitForSingleObject(htmr, INFINITE);
 
		HWND hsparkwnd = 0;
		EnumWindows(&EnumWindowsProc, (LPARAM) &hsparkwnd);
		if( hsparkwnd == 0 ) continue;
 
		CoInitialize(NULL);
		HRESULT ret = CoCreateInstance(
			CLSID_TaskbarList, 
			NULL, 
			CLSCTX_SERVER, 
			IID_ITaskbarList, 
			(LPVOID*) &ptl
		);
 
		if( ret == S_OK )
			ptl->DeleteTab(hsparkwnd);
 
		ptl->Release();
		CoUninitialize();
	}
 
	return 0;
}
 
BOOL CALLBACK EnumWindowsProc(HWND hwnd, LPARAM lParam)
{
	int tmpsz = 16;
	LPWSTR lptmp = (LPWSTR) malloc(sizeof(WCHAR) * tmpsz);
 
	RtlZeroMemory(lptmp, tmpsz);
	GetClassName(hwnd, lptmp, tmpsz);
 
	if( wcscmp(lptmp, L"SunAwtFrame") != 0 ) {
		free(lptmp);
		return TRUE;
	}
 
	RtlZeroMemory(lptmp, tmpsz);
	GetWindowText(hwnd, lptmp, tmpsz);
 
	if( wcscmp(lptmp, L"Spark -") > 0 ) {
		*((HWND*) lParam) = hwnd;
		free(lptmp);
		return FALSE;
	}
 
	free(lptmp);
	return TRUE;
}

To compile this program, create a new, empty Visual C++ project. Create a new cpp file and drop the code above inside. If you get compile errors about converting wchar_t* to const char* then change your character set from Multibyte to Unicode in the project properties.

Now you might be wondering why I don’t just change the screensaver time-out or turn off the requirement for a password. Well even though I’m a local administrator, there is a domain-wide GPO that prevents me from doing so. (Yes I know I can edit the registry, but that setting doesn’t survive a GP refresh.) I understand the reason for the policy, but five minutes seems a bit too short.

I wanted to fix this problem AND keep my job at the same time. Alice’s “dipping duck” inspired me to write a simple program to simulate mouse movement.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43

#include <windows.h>
 
int WINAPI WinMain( HINSTANCE hInstance, HINSTANCE hPrevInstance, LPSTR lpszCmdLine, int nCmdShow )
{
    HANDLE htmr = CreateWaitableTimer(NULL, TRUE, L"CheckIdle");
    LARGE_INTEGER lidt;
    LASTINPUTINFO lii;
 
    __int64 qwdt = -30 * 10000000; // 30 seconds
    lidt.LowPart = (DWORD) (qwdt & 0xFFFFFFFF);
    lidt.HighPart = (LONG) (qwdt >> 32);
 
    while( TRUE ) {
        SetWaitableTimer(htmr, &lidt, 0, NULL, NULL, FALSE);
        WaitForSingleObject(htmr, INFINITE);
 
        RtlZeroMemory(&lii, sizeof(LASTINPUTINFO));
        lii.cbSize = sizeof(LASTINPUTINFO);
        BOOL ret = GetLastInputInfo(&lii);
 
        int threshold = 3 * 60; // 3 minutes
        int idletime = ret ? (GetTickCount() - lii.dwTime) / 1000 : 0;
 
        BOOL scrnsvr = FALSE;
        SystemParametersInfo(SPI_GETSCREENSAVERRUNNING, 0, &scrnsvr, 0);
 
        if( idletime > threshold && !scrnsvr ) {
            MOUSEINPUT mi;
            RtlZeroMemory(&mi, sizeof(MOUSEINPUT));
            mi.dwFlags = MOUSEEVENTF_MOVE;
            mi.dx = 1;
            mi.dy = 1;
 
            INPUT in;
            in.type = INPUT_MOUSE;
            in.mi = mi;
 
            SendInput(1, &in, sizeof(in));
        }
    }
 
    return 0;
}

Every thirty seconds the program checks to see if the computer is idle. After three minutes of inactivity (and if the screensaver isn’t running), it moves the mouse cursor one pixel down and to the right. If I lock the workstation or manually activate the screensaver, the program won’t do anything. To compile this program, create a new empty C++ Win32 application project. Add a new cpp file, drop in the code above, and hit “Build Solution”.

Now the screensaver won’t be a nuisance when I’m trying to read. I just have to make sure to hit Win-L before I leave my desk!

My solution worked but it was pretty ugly. So I wondered if it would be better to implement a FUSE file system to do the same task. Unfortunately PHP didn’t have bindings for libfuse at the time. After several failed attempts at using SWIG to automagically generate an extension, I came to the harsh realization that if I wanted FUSE bindings I’d have to do it myself. I mean… how hard could it be?

It actually wasn’t too bad. I read a bunch of non-existent documentation on the Zend API and went to work on a prototype. It took me about a week, but I now have a working PHP extension. I’ve also since implemented TracFS to replace my ugly script.

I’m almost finished with writing the documentation for the PHP website. To build the extension, you have to check out the sources from CVS.

$ svn co http://svn.php.net/repository/pecl/fuse/trunk/ php_fuse
$ cd php_fuse && phpize
$ ./configure
$ make && sudo make install

You’ll need to install PHP and FUSE (or MacFUSE) before you can build the extension. Once everything is installed, you can test php_fuse with the scripts in pecl/fuse/examples. For now, the extension does not support writing files, only reading. Sometime in the near future I will implement the rest of the VFS callbacks.

I tried following the On OS X guide with limited success. Apparently, ffmpeg won’t compile on Leopard and that caused the entire libxine compilation to fail. I tried various combinations of configure flags, but none seemed to help. I looked at the source to try and fix it, but unfortunately I’m not terribly familiar with x86 assembly.

The problem code had to do with video rendering, so since Amarok doesn’t use it I just cut it out of the build process. Edit src/combined/ffmpeg/Makefile and find

xineplug_LTLIBRARIES = xineplug_decode_ff.la xineplug_decode_dvaudio.la

xineplug_LTLIBRARIES = xineplug_decode_dvaudio.la

After that, edit src/post/Makefile and find

SUBDIRS = planar goom visualizations mosaico deinterlace audio

SUBDIRS = goom visualizations mosaico audio

With libxine installed, I had a few more steps to build Amarok. I installed taglib

$ sudo fink install taglib

LDFLAGS="-Wl,-dylib_file,/System/Library/Frameworks/OpenGL.framework/Versions/ \
  A/Libraries/libGL.dylib:/System/Library/Frameworks/OpenGL.framework/Versions/ \
  A/Libraries/libGL.dylib"

./configure --prefix=/sw --with-qt-includes=/sw/include/qt \
  --with-qt-libraries=/sw/lib/qt3/lib

I now have Amarok successfully compiled and running on Leopard. This took me quite a bit of time to get right, so I’m hoping someone else will find it useful.

I had originally implemented the ListView using a control style to eliminate the annoying flicker every time the list was repainted. Once I started custom drawing the control, I saw odd artifacts in the ListView’s client area when I scrolled or moved the mouse. If I turned off double buffering, it worked fine.

I studied the custom draw process looking for my mistake, but I couldn’t find any. I even used the .NET Reflector to look at Microsoft’s implementation. The message I’m filtering has a handle to the control’s back-buffer. So it seemed like I should just be able to draw directly into that. But no matter what I did, it produced the same artifacts.

After about three days of banging my head, I managed to make it work by drawing into my own buffer and then using BitBlt() to copy my image to the control’s buffer. I also had to set the double buffer style with a window message rather than using the built-in property, although admittedly, I’m not sure why.

I like visual aids, so here’s a code snippet from my project. I changed names and removed a bunch of unnecessary lines for easier reading.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116

#include "stdafx.h"
#include "MyListViewControl.h"
 
using namespace MyListViewControl;
using namespace System::Drawing;
using namespace System::Windows::Forms;
 
MyListViewControl::MyListViewControl() { }
 
bool MyListViewControl::CustomDraw( Message% m )
{
	System::Drawing::Rectangle^ mprc;
	System::String^ cchQueueIndex;
	ListViewItem^ mpItem;
	LPNMCUSTOMDRAW lpnmcd;
	Graphics^ mpgfx;
	HDC hmemdc;
	RECT sOrderRect;
	RECT sStopRect;
	Brush^ mpbgbr;
	Brush^ mpfgbr;
	int nOrderLen;
	int nOrderOffset;
	int nRightEdge;
	int nRectWidth;
	int nItem;
 
	// Grab the custom draw info
	lpnmcd = (LPNMCUSTOMDRAW) m.LParam.ToPointer();
 
	// Figure out which drawing stage we're in
	switch( lpnmcd->dwDrawStage )
	{	
 
	// Control Pre-paint
	case( CDDS_PREPAINT ):
 
		// Request to be notified during item draw
		__super::WndProc( m );
		m.Result = (System::IntPtr) CDRF_NOTIFYITEMDRAW;
		return true;
 
 
	// List Item Pre-paint
	case( CDDS_ITEMPREPAINT ):
 
		// Request to be notified after item draw
		__super::WndProc( m );
		m.Result = (System::IntPtr) CDRF_NOTIFYPOSTPAINT;
		return true;
 
 
	// List Item Post-paint
	case( CDDS_ITEMPOSTPAINT ):
 
		__super::WndProc( m );
 
		// Get the current item
		nItem = static_cast<int>( lpnmcd->dwItemSpec );
		mpItem = (ListViewItem^) this->Items[nItem];		
 
		// Get the device context and item bounds
		mpgfx = this->CreateGraphics();
		mprc = this->GetItemRect( nItem, ItemBoundsPortion::Label );
 
		// *** Your custom drawing code here ***
 
		// Copy the bubble to the screen DC
		hmemdc = reinterpret_cast<HDC>( static_cast<void*>(mpgfx->GetHdc()) );
		BitBlt( 
			lpnmcd->hdc, 
			sOrderRect.left, 
			mprc->Y, 
			nRectWidth, 
			mprc->Height, 
			hmemdc, 
			sOrderRect.left, 
			mprc->Y, 
			SRCCOPY );
		mpgfx->ReleaseHdc( (System::IntPtr) hmemdc );
 
		// Let Windows redraw the border
		m.Result = (System::IntPtr) CDRF_DODEFAULT;
		return true;
	}
 
	return false;
}
 
void MyListViewControl::WndProc( Message% m )
{
	LPNMHDR lpnmhdr;
 
	switch( m.Msg )
	{
 
	// Reflected Notifications
	case( OCM_NOTIFY ):
 
		// Get the notification
		lpnmhdr = (LPNMHDR) m.LParam.ToPointer();
		if( lpnmhdr == NULL ) break;
 
		// We want custom draw messages
		if( lpnmhdr->code == NM_CUSTOMDRAW  
			&& lpnmhdr->hwndFrom == this->Handle.ToPointer() )
		{
			if( this->CustomDraw(m) ) return;
		}
 
		break;
 
	}
 
	__super::WndProc( m );
}

And here’s the header file with the double buffering message call:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42

#pragma once
 
namespace MyListView
{
	public ref class MyListView : 
		public System::Windows::Forms::ListView
	{
 
	private:
		bool CustomDraw( System::Windows::Forms::Message% m );
 
	public:
		virtual property bool DoubleBuffered
		{
			bool get() override
			{
				LRESULT lStyles = ::SendMessage( 
					(HWND) this->Handle.ToPointer(), 
					LVM_GETEXTENDEDLISTVIEWSTYLE, 0, 0 );
				return (lStyles & LVS_EX_DOUBLEBUFFER) != 0;
			}
 
			void set( bool fValue ) override
			{
				// I tried using the DoubleBuffered property, but I
				// get artifacts. So, let's do this the MFC way...
				::SendMessage( 
					(HWND) this->Handle.ToPointer(), 
					LVM_SETEXTENDEDLISTVIEWSTYLE, 
					LVS_EX_DOUBLEBUFFER, 
					LVS_EX_DOUBLEBUFFER );
			}
		}
 
	protected:
		virtual void WndProc( System::Windows::Forms::Message% m ) override;
 
	public:
		MyListViewControl();
 
	};
}

I spend days searching online and couldn’t find anyone who talked about custom drawing a .NET ListView with double buffering. I hope somebody will find this useful.

I came home from work Monday to find that my computer had restarted after seventy-two days uptime. I had been in the middle of a project the night before so I already wasn’t happy when I sat down. I tried to launch the Event Viewer to see what had caused the reboot, but got this error:

“Windows cannot access the specified device, path, or file. You may not have appropriate permissions to access the item.”

I downloaded the NT Offline Password Cracker and unlocked the local administrator account. This let me at least login with admin privileges. I tried again to launch the Event Viewer, but this time it silently failed. Strangely, starting the Microsoft Management Console and then adding the Event Viewer snap-in worked.

The log showed that several services were failing to start on boot due to insufficient privileges. I opened “My Computer” to look at the file ACLs and the label for my boot disk said “Access Denied”. Clicking on the icon yielded the same error in a message box.

Upon reboot, most services were able to start again. The System Event Notification Service wouldn’t start because it couldn’t write to the event logs. So I granted “Local Service” full permissions on the logs. Everything seemed to be fixed.

I let Windows install its 3,278,904,732,984 updates and then finally Service Pack 1. After the SP1 installation, I tried to login but was greeted with this little gem:

“The Software Licensing Service reported that the license store contains inconsistent data.”

I wasn’t exactly sure what to do since my copy of Windows was genuine. So, I thought, what if I forced Windows to re-activate? I remembered reading several months prior, that it was possible to “re-arm” Windows and set the evaluation clock back to zero. That fixed the licensing error and I didn’t even have to re-activate.

Everything seemed to finally be working again. Except Windows Update wanted to download and install SP1 again. So, I let it do that and it broke the licensing mechanism again! After fixing that again, Windows Update still thought SP1 wasn’t installed. I checked the Windows version and sure enough, I was using build 6000! wtf?

In the end, I had to flatten my machine to completely fix everything. I now have Windows installed and up-to-date. I’m annoyed that I had go to such lengths to re-gain access to my computer. But as a developer, I understand that sometimes software fails. File system corruption isn’t fun, but it was mostly fixable.

The part that I find really bothersome is that Windows wouldn’t let me use MY computer with MY licensed installation because something was corrupted. I understand the desire for ISVs to protect their products, but licensing mechanisms can introduce defects and cause undesirable side-effects. Like any operating system, Linux has issues. But there’s one issue it will never have: failure due to licensing software.

I’m happy to see the “reduced functionality mode” has been eliminated in Vista SP1. But, in following with a good article on the subject, I think Microsoft should spend less time on license enforcement and more time on software quality.

My goal was to get CLAWS running the way RIT uses it, and then write patches to the main codebase. If certain people in high places liked the changes I made, they could take the patches and apply them upstream. Even if that doesn’t happen, I could always fork the project and continue development on my own. For political reasons, I’d have to wait to do this until after I graduate.

The PAWS Project is aimed at taking what is now a very RIT-centric software system and transforming it into something the general public can use. Much to the chagrin of some un-named information security officials, CLAWS is open source and so I can (at very least) develop from the r2977 snapshot.

In the coming few months I plan to have my documentation finished for building and installing CLAWS. I should have a lighter schedule this summer, so I’m hoping to get most of my development work done then.

The CLAWS central server manages communications between the various clients and back-end systems. A self-help tool allows students to activate an account and edit identity and mail preferences. The Help Desk client provides account management functions for staff that streamlines account creation and maintenance across the multiple systems.

Right now, CLAWS is used in production by both students and Help Desk staff. We are presently working to integrate IPEdit functionality into CLAWS. Visit the project homepage for more information.

The domain member box is running Gentoo Linux. So you may need to adjust the steps to fit your flavor. Make sure the USE flags kerberos, ldap, samba, ssl, and winbind are set. Start by installing an NTP client.

# emerge ntp

The purpose of NTP is to keep your computer’s clock in-sync with the domain controller. Edit your /etc/ntp.conf file to use your domain controller as the time server. Then sync your clock, start the NTP client, and install Samba:

# ntpdate ad01.rit.edu
# /etc/init.d/ntpd start
# rc-update add ntpd default
# emerge samba

Now you can configure kerberos. Open your /etc/krb5.conf file and make it look like this:

[libdefaults]
  ticket_lifetime = 600
  default_realm = RIT.EDU
  clockskew = 120

[realms]
  RIT.EDU = {
    kdc = ad01.rit.edu
    default_domain = RIT.EDU
  }

[domain_realm]
  .rit.edu = RIT.EDU
  rit.edu = RIT.EDU

Test your kerberos setup by requesting a ticket from your domain controller.

# kinit administrator
# klist
# kdestroy

Edit your Samba configuration.

[global]
  workgroup = RIT
  realm = RIT.EDU
  server string =

  log file = /var/log/samba/log.%m
  max log size = 50

  hosts allow = 127.0.0.1 129.21.0.0/16   
  hosts deny = 0.0.0.0/0
  security = ADS
  allow trusted domains = yes
  password server = ad01.rit.edu
  encrypt passwords = yes
  min protocol = NT1

  winbind enum users = yes
  winbind enum groups = yes
  winbind cache time = 600
  winbind use default domain = yes
  template homedir = /home/%U
  obey pam restrictions = yes
  template shell = /bin/bash

  socket options = TCP_NODELAY SO_RCVBUF=8192 SO_SNDBUF=8192
  domain master = no
  local master = no

  idmap uid = 10000-99999
  idmap gid = 10000-99999

Add winbind to the daemon_list variable in /etc/conf.d/samba.

daemon_list="smbd nmbd winbind"

Join the domain and start Samba.

# net ads join -U administrator
# /etc/init.d/samba start

Add winbind to the passwd and group lines in /etc/nsswitch.conf.

passwd:      compat winbind
shadow:      compat
group:       compat winbind

Test user resolution and add Samba to the startup:

# getent passwd administrator
# rc-update add samba default

Add the bolded lines to your /etc/pam.d/system-auth file.

auth required pam_env.so
auth sufficient	pam_winbind.so
auth sufficient	pam_unix.so likeauth nullok use_first_pass
auth required pam_deny.so

account	required pam_access.so
account	sufficient pam_winbind.so
account	required pam_unix.so

password required pam_cracklib.so difok=2 minlen=8 dcredit=2 ocredit=2 retry=3
password sufficient pam_unix.so nullok md5 shadow use_authtok
password required pam_deny.so

session	required pam_limits.so
session	required pam_unix.so
session required pam_mkhomedir.so

That’s it! Now all you need to do is start samba and your box is an AD domain member.