Range Check Exception when using KDF with empty password

[danielmarschall]

In my app, I noticed a "Range Check Exception" when the user didn't enter a password (which I clearly forgot to check), and therefore the, empty password was tried to be processed through KDF. The reason for the Range Check Exception is that Data[0] is accessed, which does not exist if Data is empty.

class function TDECHashAuthentication.KDFx(const Data, Seed: TBytes;
                                           MaskSize: Integer;
                                           Index: UInt32 = 1): TBytes;
begin
  if (length(Seed) > 0) then
    Result := KDFx(Data[0], Length(Data), Seed[0], Length(Seed), MaskSize, Index)
  else
    Result := KDFx(Data[0], Length(Data), NullStr, Length(Seed), MaskSize, Index)
end;

(This is just an example. Not only KDFx is affected. Other methods are affected too.)

It is dangerous that a method can fail this way, especially because "range check" can be disabled in the compiler settings for performance reasons.

In my opinion we should do the following: Find out (from official sources / reference implementation) what KDF123x/PBKDF/MGF1 is supposed to do if the input is 0 bytes. Is it supposed to fail, or is it supposed to return a key "X"?

• If it shall output a key "X", then we should implement it this way
• If it is supposed to fail, then we must throw an Exception instead of risking a memory-corruption if range checks are disabled

(edit: typo. of course I meant "instead of risking")

[MHumm]

I found one first source, which is linked in this stack overflow discussion (std6.pdf). See p. 21 and following in the PDF. https://crypto.stackexchange.com/questions/67280/formal-description-of-kdf1-and-kdf2

But: it doesn't specify what to do if the input length is 0. Quote: A key derivation function is a function KDF(x, l) that takes as input an octet string x and an integer l >= 0, and outputs an octet string of length l. The string x is of arbitrary length, although an implementation may define a (very large) maximum length for x and maximum size for l, and fail if these bounds are exceeded.

[MHumm]

Further ressearch shows, that the real specification ight be in IEEE P1363, which is a withdrawn standard from IEEE for which the specification is behind a pay wall.

[MHumm]

Is there anybody with access to IEEE P1363?

[MHumm]

Somebody provided me some documents (after I requested the IEEE P1363 spect. in another forum). I need to find the time to read them. I hope they contain information about that case.