package org.bouncycastle.crypto.digests;
/**
* implementation of SHA-1 as outlined in "Handbook of Applied Cryptography", pages 346 - 349.
*
* It is interesting to ponder why the, apart from the extra IV, the other difference here from MD5
* is the "endienness" of the word processing!
*/
public class SHA1Digest
extends GeneralDigest
{
private static final int DIGEST_LENGTH = 20;
private int H1, H2, H3, H4, H5;
private int[] X = new int[80];
private int xOff;
/**
* Standard constructor
*/
public SHA1Digest()
{
reset();
}
/**
* Copy constructor. This will copy the state of the provided
* message digest.
*/
public SHA1Digest(SHA1Digest t)
{
super(t);
H1 = t.H1;
H2 = t.H2;
H3 = t.H3;
H4 = t.H4;
H5 = t.H5;
System.arraycopy(t.X, 0, X, 0, t.X.length);
xOff = t.xOff;
}
public String getAlgorithmName()
{
return "SHA-1";
}
public int getDigestSize()
{
return DIGEST_LENGTH;
}
protected void processWord(
byte[] in,
int inOff)
{
X[xOff++] = ((in[inOff] & 0xff) << 24) | ((in[inOff + 1] & 0xff) << 16)
| ((in[inOff + 2] & 0xff) << 8) | ((in[inOff + 3] & 0xff));
if (xOff == 16)
{
processBlock();
}
}
private void unpackWord(
int word,
byte[] out,
int outOff)
{
out[outOff] = (byte)(word >>> 24);
out[outOff + 1] = (byte)(word >>> 16);
out[outOff + 2] = (byte)(word >>> 8);
out[outOff + 3] = (byte)word;
}
protected void processLength(
long bitLength)
{
if (xOff > 14)
{
processBlock();
}
X[14] = (int)(bitLength >>> 32);
X[15] = (int)(bitLength & 0xffffffff);
}
public int doFinal(
byte[] out,
int outOff)
{
finish();
unpackWord(H1, out, outOff);
unpackWord(H2, out, outOff + 4);
unpackWord(H3, out, outOff + 8);
unpackWord(H4, out, outOff + 12);
unpackWord(H5, out, outOff + 16);
reset();
return DIGEST_LENGTH;
}
/**
* reset the chaining variables
*/
public void reset()
{
super.reset();
H1 = 0x67452301;
H2 = 0xefcdab89;
H3 = 0x98badcfe;
H4 = 0x10325476;
H5 = 0xc3d2e1f0;
xOff = 0;
for (int i = 0; i != X.length; i++)
{
X[i] = 0;
}
}
//
// Additive constants
//
private static final int Y1 = 0x5a827999;
private static final int Y2 = 0x6ed9eba1;
private static final int Y3 = 0x8f1bbcdc;
private static final int Y4 = 0xca62c1d6;
private int f(
int u,
int v,
int w)
{
return ((u & v) | ((~u) & w));
}
private int h(
int u,
int v,
int w)
{
return (u ^ v ^ w);
}
private int g(
int u,
int v,
int w)
{
return ((u & v) | (u & w) | (v & w));
}
private int rotateLeft(
int x,
int n)
{
return (x << n) | (x >>> (32 - n));
}
protected void processBlock()
{
//
// expand 16 word block into 80 word block.
//
for (int i = 16; i <= 79; i++)
{
X[i] = rotateLeft((X[i - 3] ^ X[i - 8] ^ X[i - 14] ^ X[i - 16]), 1);
}
//
// set up working variables.
//
int A = H1;
int B = H2;
int C = H3;
int D = H4;
int E = H5;
//
// round 1
//
for (int j = 0; j <= 19; j++)
{
int t = rotateLeft(A, 5) + f(B, C, D) + E + X[j] + Y1;
E = D;
D = C;
C = rotateLeft(B, 30);
B = A;
A = t;
}
//
// round 2
//
for (int j = 20; j <= 39; j++)
{
int t = rotateLeft(A, 5) + h(B, C, D) + E + X[j] + Y2;
E = D;
D = C;
C = rotateLeft(B, 30);
B = A;
A = t;
}
//
// round 3
//
for (int j = 40; j <= 59; j++)
{
int t = rotateLeft(A, 5) + g(B, C, D) + E + X[j] + Y3;
E = D;
D = C;
C = rotateLeft(B, 30);
B = A;
A = t;
}
//
// round 4
//
for (int j = 60; j <= 79; j++)
{
int t = rotateLeft(A, 5) + h(B, C, D) + E + X[j] + Y4;
E = D;
D = C;
C = rotateLeft(B, 30);
B = A;
A = t;
}
H1 += A;
H2 += B;
H3 += C;
H4 += D;
H5 += E;
//
// reset the offset and clean out the word buffer.
//
xOff = 0;
for (int i = 0; i != X.length; i++)
{
X[i] = 0;
}
}
}