What's wrong with the "StretchKey" implementation?

I'm trying to build a Delphi version of this algorithm:

void PWSfileV3::StretchKey(const unsigned char *salt, unsigned long saltLen,
                           const StringX &passkey,
                           unsigned int N, unsigned char *Ptag)
{
  /*
  * P' is the "stretched key" of the user passphrase and the SALT, as defined
  * by the hash-function-based key stretching algorithm in
  * http://www.schneier.com/paper-low-entropy.pdf (Section 4.1), with SHA-256
  * as the hash function, and N iterations.
  */
  int passLen = 0;
  unsigned char *pstr = NULL;

  ConvertString(passkey, pstr, passLen);
  unsigned char *X = Ptag;
  SHA256 H0;
  H0.Update(pstr, passLen);
  H0.Update(salt, saltLen);
  H0.Final(X);

#ifdef UNICODE
  trashMemory(pstr, passLen);
  delete[] pstr;
#endif

  ASSERT(N >= MIN_HASH_ITERATIONS); // minimal value we're willing to use
  for (unsigned int i = 0; i < N; i++) {
    SHA256 H;
    // The 2nd param in next line was sizeof(X) in Beta-1
    // (bug #1451422). This change broke the ability to read beta-1
    // generated databases. If this is really needed, we should
    // hack the read functionality to try both variants (ugh).
    H.Update(X, SHA256::HASHLEN);
    H.Final(X);
  }
}

      

        
        
        
      

    

Update: (Feature missing)

void ConvertString(const StringX &text,
                   unsigned char *&txt,
                   int &txtlen)
{
  LPCTSTR txtstr = text.c_str(); 
  txtlen = text.length();

#ifndef UNICODE
  txt = (unsigned char *)txtstr; // don't delete[] (ugh)!!!
#else
#ifdef _WIN32
  txt = new unsigned char[3*txtlen]; // safe upper limit
  int len = WideCharToMultiByte(CP_ACP, 0, txtstr, txtlen,
    LPSTR(txt), 3*txtlen, NULL, NULL);
  ASSERT(len != 0);
#else
  mbstate_t mbs;
  memset(&mbs, '\0', sizeof(mbs));
  size_t len = wcsrtombs(NULL, &txtstr, 0, &mbs);
  txt = new unsigned char[len+1];
  len = wcsrtombs((char *)txt, &txtstr, len, &mbs);
  ASSERT(len != (size_t)-1);
#endif
  txtlen = len;
  txt[len] = '\0';
#endif /* UNICODE */
}

      

        
        
        
      

    

This is what I got (version D2009):

(Note: T256BitArray is defined as an array of [0..31] bytes)

procedure StretchKey(Const Salt:T256BitArray; Const Passkey:string; Const Iterations:LongWord; Var KeyResult:T256BitArray);
var
   pStr : RawByteString;
   wHash : THash_sha256;
   loop : integer;
begin
  pStr := AnsiString(PassKey);

  wHash := THash_SHA256.Create;
  try
     wHash.Init;
     wHash.Calc(pStr[1], Length(pStr));
     wHash.Calc(Salt, Length(Salt));
     wHash.Done;
     PStr := wHash.DigestStr;
  finally
     FreeAndNil(wHash);
  end;

  for loop := 0 to Iterations-1 do
  begin
     wHash := THash_sha256.Create;
     try
        wHash.Init;
        wHash.Calc(PStr[1], wHash.DigestSize);
        wHash.Done;
        PStr := wHash.DigestStr;
     finally
        FreeAndNil(wHash);
     end;
  end;

  move(pStr[1], KeyResult, sizeof(KeyResult));
end;

      

        
        
        
      

    

The original code snippet from the open source application Password Safe.

I am trying to open an existing Password Save (v3) database for reading.

It doesn't seem to matter what I do, I cannot get the algorithm to generate the hash I want.

In the above Delphi snippet, I am using the DEC v5.2 2009 component set. I also tried the DCPcrypt library. Funny how I get the same values ​​from both libraries, but nothing comes out consistent with the hash from the PWSv3 file.

The SHA256 components that I have used pass the hashes of the SHA256 test vector, so I guess this is what I did wrong when re-encoding the method.

Did I miss something?

SOLVED : everything is correct. The problem is with converting the access key string. I figured out that I need to use the WideCharToMultiByte function to get the correct code page conversion.