CODEWITHSUNDEEP
c++mfcheap-corruption
shaunakde
shaunakde

Reputation: 3108

Heap corruption in MFC program

I am having a problem with a heap corruption issue in a program. In the program I am reading a block of data and performing FFT and IFFT on it. I am doing it for 2 images, master and slave. The EXACT same code works fine for the master but shows a heap corruption for the slave file when I try to delete the slave buffer.

fcomplex is defined as:

typedef struct {float real, imag;}fcomplex;

A snippet of relevant parts of the code is attached: Full code: http://sharetext.org/7xXe

The error does not occur if I do not call the fft and ifft functions for the slave image. (Everything works fine for the master though)

To debug the error I installed Application verifier but I was not able to decode the log file. Its here: http://sharetext.org/Y2ji (XML file copy pasted)

The error visual studio give is: Heap corruption Detected: after normal block (#194456) at 0x062C0040

CCoarseFun::fcomplex * slave_bfr;
CCoarseFun::fcomplex * slave_col;


slave_bfr = Pcoarse.init_1Dcmplx(SIZE*s_cols); 
slave_col = Pcoarse.init_1Dcmplx(SIZE);

Pcoarse.cfft1d_(&SIZE,slave_col,&FFTdir); // This function causes a problem
Pcoarse.complex_mult_col(filter, slave_col, SIZE, slave_col)
Pcoarse.cfft1d_(&SIZE,slave_col,&FFTdir); // As does this one

// delete memory related to slave
delete [] slave_bfr;    // Heap corruption here
delete [] slave_col;

What is baffleing me is that the code is pretty simple and it works 100% for only the master files. Why is it crashing for the slave?

Can some one guide me to a solution or maybe a tutorial on how to use the Application verifier as well?

Thanks, Shaunak

EDIT: Using Win7 x64 - VS2010

EDIT 2: Definition for init_1Dcmplx

CCoarseFun::fcomplex* CCoarseFun::init_1Dcmplx(int n)
{
  fcomplex *a;
  a=new fcomplex[n];
  for(int i=0;i<n;i++)
  {
    a[i].real=float(0.0);
    a[i].imag=float(0.0);
  }
  return a;
}

EDIT3: COde for cfft1D_ : http://sharetext.org/hzIg

EDIT4: Code for mem.delfloat()

void CMemAlloc::del_float(float *a)
{
  if (a!=NULL)
  {
    delete[] a;
    a=NULL;
  }
  else
  {
    return;
  }
}

Upvotes: 0

Views: 1748

Answers (3)

Werner Henze
Werner Henze

Reputation: 16726

"Heap corruption Detected: after normal block (#194456) at 0x062C0040" says that you allocated a block, got the block #194456 with address 0x062C0040 and you wrote more memory bytes than you allocated. So you have a classical buffer overflow. You should consider replacing pointers with STL containers. In your case I'ld prefer using std::vector over a raw float array allocated with new float[]. The STL containers can help you detect writing beyond boundaries immediately at the wrong access, not only after deleting the memory block.

Upvotes: 0

Lubo Antonov
Lubo Antonov

Reputation: 2318

The mem_float() function is not correct. It looks like it is setting the pointer to NULL after the delete, but it is only working on a copy of the pointer, so the caller's copy is still pointing to deleted memory block.

You can just do

delete [] cf;
cf = NULL;

You have a couple of lines that look like this:

four1(cf-1,nn,isign);

I think this is accessing memory before the beginning of the array.

Beyond this, the indexing inside four1() is insanely complicated - you will have to step through it with the debugger to check the edge cases.

Upvotes: 1

the_mandrill
the_mandrill

Reputation: 30842

Firstly it's important to understand what the heap corruption error is telling you. When you run a Debug build in Visual Studio it uses the Debug version of the run-time library which has a debug heap. Whenever you allocate some memory with new there are some extra guard bytes either side of it. When you delete it the debug heap checks that those guard bytes are intact, and if not then it issues this warning.

Assuming that Pcoarse.init_1Dcmplx() allocates memory on the heap then the two calls to it will most likely allocate memory one after the other:

XXXXXX - guard bytes
slave_bfr
XYXYXY - these are the guard bytes that are probably being corrupted
slave_cols
XXXXXX

You perform the operation Pcoarse.cfft1d_() on slave_cols but then you get the heap corruption error when deleting slave_bfr. This suggests that cfft1d_() is overwriting memory before the start of slave_cols so that it's corrupting the guard bytes of slave_bfr.

So I would look through your code for cfft1d_() for places where you might have negative array indexes as this could cause the memory in slave_bfr to be stomped on.

Also check some of the SO answers with useful tips on how to make the most of the debug heap, in particular this one about how to enable more intensive memory checks.

Upvotes: 0

Related Questions