C / C++ / MFC Discussion Boards

Amit Dey15-Jun-02 12:00

15-Jun-02 12:00

I think you will find open source filesystem drivers for NTFS,ext2(Linux) at sourceforge.net very illuminating.There are a couple of Beos ports to NTFS drivers too(I think).

sonork: 100:18407

You yard yellow years yieldingly; you yanked your yearning yoke.

-Jeremy Falcon

Re: Writing a filesystem driver?

Mike Nordell16-Jun-02 5:27

Mike Nordell

16-Jun-02 5:27

Then you're in for a treat. Not!

MS has the FSD SDK for $5k (or is it perhaps $10k+ nowadays?) with no support, sparse documentation and no promises they won't change everything behind your back even from service-pack to service-pack, making this one of the "great developer experiences" that MS loves to force down our throats with a rough 2-by-4 without any K-Y.

Good luck, I really think you need it.

I guess after I am successfull in mounting a drive (including all the code to read and write) I will need to make it a shell extension so that it shows up transparently in Explorer? Is that right?

No. Once you have the filesystem mounted it's just like any other filesystem (except that you can't use half of the features in BFS on this DOS-emulation platform). You perhaps could get access to the attributes using the NT-only syntax for alternate file streams "filename:stream_name", but since you'd be the only one using it (since it's incompatible with everything except NTFS almost no-one ever uses it) you'd (obviously) have to write your own programs to access that too.

The only place I can point you to is http://ntfsd.org

++luck;

Re: Writing a filesystem driver?

clintsinger16-Jun-02 13:21

clintsinger

16-Jun-02 13:21

Thanks for you information. I am too poor to afford anything so I guess the FSD SDK is out of the question.

One thing I would like clarification on is about the dos-emulation layer you mentioned. I am aware that there is an emulation layer but are you saying I could write a driver for dos and use it to access another filesystem from within win2k?

Thanks again,
Clint

Re: Writing a filesystem driver?

Mike Nordell17-Jun-02 17:49

Mike Nordell

17-Jun-02 17:49

One thing I would like clarification on is about the dos-emulation layer you mentioned.

I forgot to add the <sarcasm> tag. I was referring to the fact that Win32 is little more than MS-DOS with longer filenames when it comes to filesystem and filename handling.
I'm sorry if it got your hopes up.

Re: Writing a filesystem driver?

18-Jun-02 10:20

Doh... I should have known better. I didn't dismiss the idea because I did read bout something like that years ago where int 0x21 was extended for Win95 to support the long file names and such. I guess I just thought there might be something for mounting partitions.

If you have done anything like this could you please check out my last post. I have source code to read from the disk sectors but it fails on ReadFile.

Thanks again,
Clint

Heaps and DisableHeapLookAside

stereo15-Jun-02 5:58

stereo

15-Jun-02 5:58

I'm having the most difficult time determining the cause of a bug somewhere
in the heap manager (I'm guessing?).

I'm using Windows 2000 SP2, with MSVC 6.0 SP5, and ATL 3.0.

In my application, *sometimes*, when adding a treeview item, the memory
allocated internally by comctl32.dll for the item is later overwritten by a
call to RegQueryKeyValueEx(). As a result the treeview is corrupted, and
the application becomes unstable. The default process heap is being used in
both allocations, and both allocations are happening on the same thread.

Here is an assembly listing of the code before and after the call to
HeapAlloc() in comctl32:

_text:7179A8ED push 24h ; dwBytes
_text:7179A8EF push 8 ; dwFlags
_text:7179A8F1 push dword ptr [edi+94h] ; hHeap

Values at ESP before the call ... 130000 seems to be the default heap
00130000 00000008 00000024

_text:7179A8F7 call ds:HeapAlloc

eax after the call to HeapAlloc() is 172de0

Now, after the treeview item is added, the _ATL_DEBUG_INTERFACES macro
queries the registry for the value of an interface name ... and this code is
in advapi32.dll!RegQueryKeyValueEx:

_text:77DB2B3E push ecx
_text:77DB2B3F push edx
_text:77DB2B40 push dword ptr [eax+18h]

ESP before the call:
00130000 00000000 00000024

_text:77DB2B43 call ds:__imp__RtlAllocateHeap@12

eax after the call to __imp__RtlAllocateHeap@12() is 172de0!

How on earth can HeapAlloc() be returning the same value twice?

Now the interesting part is that if I define the DisableHeapLookAside value
in the Image File Execution Options, everything works fine, and HeapAlloc()
never returns duplicate values. If I do a HeapValidate() on the treeview
hItem value, sometimes it fixes it. I've also instrumented all my code with
Boundschecker, and it reported no memory corruption issues.

After checking the KB, I found article Q195009, which describes the Heap
Manager change in NT4 SP4 ... and goes on to say that sometimes the cause of
this bug is when an application uses memory previously freed. So I set some
breakpoints on HeapFree() calls, but never saw the pointer being freed
between the two calls to HeapAlloc().

My real question is why is this happening to begin with, and what does
DisableHeapLookAside do to cause the problem to go away?

Thanks in advance,

Scott.

Re: Heaps and DisableHeapLookAside

Mike Nordell16-Jun-02 5:52

Mike Nordell

16-Jun-02 5:52

You also have to monitor HeapReAlloc in addition to HeapFree.

Before jumping to conclusions, have you tried this with a "checked build" and/or using e.g Purify or something similar to really rule out the possibility that your application is overwriting some heap memory it doesn't own? If you're longing for an adventure you might even replace the malloc/free implementation functions from the CRT with your own functions that use a private heap.

If it is beyond all dubt ruled out that the a memory overflow bug is introduced by your app, one can only conclude that it has to be a bug in the NT heap manager and you should notify MS of this ASAP.

If this is an MS bug I could guess that they don't have a (nasty enough) regression test for HeapReAlloc and the heap lookaside combination (which I personaly think is a fairly reasonable guess).

The following assumes the bug is not in your code:

My real question is why is this happening to begin with

Bugs in the heap manager implementation.

what does
DisableHeapLookAside do to cause the problem to go away?

Answering that question would probably involve explaining how their NT4sp4+ and NT5+ heap manager works in such a detailed way that one could write it oneself. That would probably violate some kind of "trade secrets". Smile | :)

Please check HeapReAlloc also and report back. This can become interesting (unless it's a bug in your code)!

Re: Heaps and DisableHeapLookAside

stereo16-Jun-02 6:12

stereo

16-Jun-02 6:12

Hi,

Thanks for the reply ... I was wondering if my question was even going to get one Smile | :)

I've done a full instrumentation of all the entire using Boundschecker, and no memory overwrites turned up. Maybe I should have mentioned in my original post that not a lot of code is being executed in the time it takes HeapAlloc() to spit out the duplicate pointer.

Basically, the structure of code is like:

Call one event sink
Insert the treeview item (first pointer)
Enumerate the treeview using TVM_GETNEXTITEM to ensure the item was added
Call next event sink
ATL event tracing (duplicate pointer)

I'd estimate that there were about ~30 calls to HeapAlloc in the time between the duplicate pointer returns.

I didn't check calls to HeapReAlloc, but I'll do that and post my results.

Fortunately, at work we have an MSDN subscription which includes 2 developer support calls, and I've been authorized to use one of them for this particular instance ... so hopefully I'll get an answer sometime soon.

Thanks for the help,

Scott

Re: Heaps and DisableHeapLookAside

Mike Nordell16-Jun-02 6:42

Mike Nordell

16-Jun-02 6:42

Just a though that struck me. If you can write some code before that "ATL event tracing (duplicate pointer)", and you can duplicate the bug even if you break into a debugger (which it seems you can), what about using a local variable and set it (using the debugger) to point to the memory you think is already allocated (but the "second" allocation returns) and just call IsBadReadPtr/IsBadWritePtr on it?

OK, checking HeapReAlloc isn't gonna hurt, but it never hurts to exhaust all options before that "support" call either. Smile | :)

Re: Heaps and DisableHeapLookAside

stereo16-Jun-02 6:50

stereo

16-Jun-02 6:50

Well,

I after adding the treeview item, I set a breakpoint on all HeapReAlloc() calls, and it was hit one time, on a call to realloc(), which is using the CRT heap ... aside from that, the GetProcessHeap() heap wasn't touched by any HeapReAlloc calls.

I tried your last suggestion ... I took the HTREEITEM returned by the TVM_INSERTITEM (which actually points to a 0x24 byte internal structure ... gotta love reverse engineering MS dll's) and called IsBadReadPtr(hItem, 0x24), and it returned 0 so ... hmmm ...

If you can think of anything else, let me know ... I feel as though I've tried every thing conceivable Frown | :(

Thanks,

Scott.

How do i put a new line in an Edit control? (non-MFC)

redeemer15-Jun-02 6:01

redeemer

15-Jun-02 6:01

If i do this:

char s[] = "hello\nhow\nare\nyou\doing"
SetDlgItemText(ghWnd, IDC_EDIT, s);

then a newline doesen't appear, only a "|". what am i doing wrong? and it doesen't work if i use \r instead of \n either.

Re: How do i put a new line in an Edit control? (non-MFC)

Michael Dunn15-Jun-02 5:59

Michael Dunn

15-Jun-02 5:59

See the VC forum FAQ

--Mike--
Just released - RightClick-Encrypt - Adds fast & easy file encryption to Explorer
Like the Google toolbar? Then check out UltraBar, with more features & customizable search engines!
My really out-of-date homepage
Sonork-100.19012 Acid_Helm

Re: How do i put a new line in an Edit control? (non-MFC)

Max Santos15-Jun-02 6:26

Max Santos

15-Jun-02 6:26

use "\r\n"

like this char s[] = "hello\r\nhow\r\nare\r\nyou\r\doing"

Casa.Sapo.pt

fork() and pipe() problem

ChiYung15-Jun-02 5:43

ChiYung

15-Jun-02 5:43

Hi,

I know it is a VC++ forum, but i really need help!!! Let's see my code first:

********************************************************************************
void operation(int[], char[]);
void waitprocess(int);
void main(void)
{
int i, j;
static char buffer[BUFSIZE+10], X_value[5], message[BUFSIZE];
int f_des[2];

if (pipe(f_des) == -1) //create the pipe
{
perror("Pipe");
exit(2);
}

if (fork()==0) //In the CHILD
{
// do nothing here because we will
// create 5 CHILDs by PARENT again later.
}
else //PARENT
{
for (j=1; j<=5; j++) //create 5 CHILD process
{
if (fork()==0)
{
static char temp[BUFSIZE];
sprintf(temp, "Thread %d", getpid()); //Get the thread ID
operation(f_des, temp); //Send the request
}
}

while (X>0) // keep doing (read from pipe) until X=0
{
close(f_des[1]);
if (read(f_des[0], message, BUFSIZE) != -1) //Read from pipe
{
sprintf(buffer, " - X = %d \n", X);
strcat(message, buffer);
write(1, message, sizeof(message)); //display output
waitprocess(getpid()); //force sleep awhile
X--;
}
else
{
printf("error");
}
}
}
}
********************************************************************************

I have to create 5 "threads" but using fork() because i have to use pipe() to communicate with "server". This program is doing that 5 "threads" are trying to decrement the global X until X=0. I use pipe as a mechanisam to protect this "critical section". I don't know whether i did it right or wrong. Please help me to check!!!

Also, I found that the output looks strange. The output is always in a pattern. For example, the output looks liked:

Thread 1001 - X = 20
Thread 1003 - X = 19
Thread 1002 - X = 18
Thread 1005 - X = 17
Thread 1004 - X = 16
Thread 1001 - X = 15
Thread 1003 - X = 14
Thread 1002 - X = 13
Thread 1005 - X = 12
Thread 1004 - X = 11
*
*
*

You can see the sequence is 1,3,2,5,4,1,3,2,5,4...etc. I think the output should be in random sequence. Why it's happened???

Thanks!!!!! Confused | :confused:

Change my window width/height?

15-Jun-02 5:08

Hi,
how can I change my window dimension at runtime?

Is there something better as SetWindowPos() or how do I use it to change only the window size (position stays untouched)?

Thanks

Re: Change my window width/height?

Hans Ruck15-Jun-02 5:20

Hans Ruck

15-Jun-02 5:20

Try MoveWindow which is an easyer way of calling SetWindowPos.

rechi

Re: Change my window width/height?

Ravi Bhavnani15-Jun-02 5:30

Ravi Bhavnani

15-Jun-02 5:30

Use SWP_NOMOVE | SWP_NOZORDER in the "nFlags" argument of CWnd::SetWindowPos(). This will cause only the window's size to be changed.

/ravi

Let's put "civil" back into "civilization"
http://www.ravib.com
ravib@ravib.com

Re: Change my window width/height?

15-Jun-02 6:09

Okay, I use this code now:

RECT winRect = {0, 0, nWidth, nHeight};
AdjustWindowRectEx(&winRect, WS_OVERLAPPEDWINDOW, FALSE, WS_EX_APPWINDOW | WS_EX_WINDOWEDGE);
SetWindowPos(m_data.hWnd, HWND_TOP, 0, 0,
    winRect.right-winRect.left,
    winRect.bottom-winRect.top,
    SWP_NOMOVE|SWP_NOZORDER|SWP_NOCOPYBITS);

I still have a problem: When the window is maximized and I change the dimensions with my code, the maximize button is not changed. Instead Windows thinks my window is still maximized.

Any ideas? Thanks for help again! Smile | :)

Re: Change my window width/height? (solved)

15-Jun-02 11:00

Performing a ShowWindow(m_data.hWnd, SW_RESTORE) before resizing did solve the problem.

Re: Resize and maximize button is wrong?

15-Jun-02 6:15

PS: This should be the subject for last post

IStream question

Hans Ruck15-Jun-02 4:27

Hans Ruck

15-Jun-02 4:27

Does SetSize method of IStream preserve the information existing in the stream object if the new size is smaller then the previous one?

rechi

Re: IStream question

[James Pullicino]15-Jun-02 8:00

[James Pullicino]

15-Jun-02 8:00

It should truncate it I think. It might depend on the IStream implementation though. Best is to make a test program. Create an IStream working on global memory (use StreamFromHGlobal() or something like that) and write to it, then setsize then read it back.

James

Drinking In The Sun
Forgot Password?

Spy++

vikramlinux15-Jun-02 1:04

vikramlinux

15-Jun-02 1:04

HI I am developing a utility like spy++. Here till now I am able to highlight the mose controls & windows .But for the painted controls like static text in visual basic applications the things are not working.I am using
WindowFromPoint() API.
can anybody help me out ? Frown | :(

VIKS

Re: Spy++

Paul Selormey15-Jun-02 1:15

Paul Selormey

15-Jun-02 1:15

See the source code section of the following page - there is an implementation there.

http://freespace.virgin.net/james.brown7/

Best regards,
Paul.

Paul Selormey, Bsc (Elect Eng), MSc (Mobile Communication) is currently Windows open source developer in Japan.

Re: Spy++

Rama Krishna Vavilala15-Jun-02 1:08

Rama Krishna Vavilala

15-Jun-02 1:08

vikrams wrote:
.But for the painted controls like static text in visual basic applications the things are not working.

This is nothing strange nor a bug is your app. Visual Basic uses Windowless Controls so you don't get any window.

Use Ctrl+Left/Right to switch messages, Ctrl+Up/Down to switch threads, Ctrl+Shift+Left/Right to switch pages.