You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
320 lines
8.9 KiB
320 lines
8.9 KiB
4 weeks ago
|
/**
|
||
|
* Secure Hash Algorithm with 160-bit digest (SHA-1) implementation.
|
||
|
*
|
||
|
* @author Dave Longley
|
||
|
*
|
||
|
* Copyright (c) 2010-2015 Digital Bazaar, Inc.
|
||
|
*/
|
||
|
var forge = require('./forge');
|
||
|
require('./md');
|
||
|
require('./util');
|
||
|
|
||
|
var sha1 = module.exports = forge.sha1 = forge.sha1 || {};
|
||
|
forge.md.sha1 = forge.md.algorithms.sha1 = sha1;
|
||
|
|
||
|
/**
|
||
|
* Creates a SHA-1 message digest object.
|
||
|
*
|
||
|
* @return a message digest object.
|
||
|
*/
|
||
|
sha1.create = function() {
|
||
|
// do initialization as necessary
|
||
|
if(!_initialized) {
|
||
|
_init();
|
||
|
}
|
||
|
|
||
|
// SHA-1 state contains five 32-bit integers
|
||
|
var _state = null;
|
||
|
|
||
|
// input buffer
|
||
|
var _input = forge.util.createBuffer();
|
||
|
|
||
|
// used for word storage
|
||
|
var _w = new Array(80);
|
||
|
|
||
|
// message digest object
|
||
|
var md = {
|
||
|
algorithm: 'sha1',
|
||
|
blockLength: 64,
|
||
|
digestLength: 20,
|
||
|
// 56-bit length of message so far (does not including padding)
|
||
|
messageLength: 0,
|
||
|
// true message length
|
||
|
fullMessageLength: null,
|
||
|
// size of message length in bytes
|
||
|
messageLengthSize: 8
|
||
|
};
|
||
|
|
||
|
/**
|
||
|
* Starts the digest.
|
||
|
*
|
||
|
* @return this digest object.
|
||
|
*/
|
||
|
md.start = function() {
|
||
|
// up to 56-bit message length for convenience
|
||
|
md.messageLength = 0;
|
||
|
|
||
|
// full message length (set md.messageLength64 for backwards-compatibility)
|
||
|
md.fullMessageLength = md.messageLength64 = [];
|
||
|
var int32s = md.messageLengthSize / 4;
|
||
|
for(var i = 0; i < int32s; ++i) {
|
||
|
md.fullMessageLength.push(0);
|
||
|
}
|
||
|
_input = forge.util.createBuffer();
|
||
|
_state = {
|
||
|
h0: 0x67452301,
|
||
|
h1: 0xEFCDAB89,
|
||
|
h2: 0x98BADCFE,
|
||
|
h3: 0x10325476,
|
||
|
h4: 0xC3D2E1F0
|
||
|
};
|
||
|
return md;
|
||
|
};
|
||
|
// start digest automatically for first time
|
||
|
md.start();
|
||
|
|
||
|
/**
|
||
|
* Updates the digest with the given message input. The given input can
|
||
|
* treated as raw input (no encoding will be applied) or an encoding of
|
||
|
* 'utf8' maybe given to encode the input using UTF-8.
|
||
|
*
|
||
|
* @param msg the message input to update with.
|
||
|
* @param encoding the encoding to use (default: 'raw', other: 'utf8').
|
||
|
*
|
||
|
* @return this digest object.
|
||
|
*/
|
||
|
md.update = function(msg, encoding) {
|
||
|
if(encoding === 'utf8') {
|
||
|
msg = forge.util.encodeUtf8(msg);
|
||
|
}
|
||
|
|
||
|
// update message length
|
||
|
var len = msg.length;
|
||
|
md.messageLength += len;
|
||
|
len = [(len / 0x100000000) >>> 0, len >>> 0];
|
||
|
for(var i = md.fullMessageLength.length - 1; i >= 0; --i) {
|
||
|
md.fullMessageLength[i] += len[1];
|
||
|
len[1] = len[0] + ((md.fullMessageLength[i] / 0x100000000) >>> 0);
|
||
|
md.fullMessageLength[i] = md.fullMessageLength[i] >>> 0;
|
||
|
len[0] = ((len[1] / 0x100000000) >>> 0);
|
||
|
}
|
||
|
|
||
|
// add bytes to input buffer
|
||
|
_input.putBytes(msg);
|
||
|
|
||
|
// process bytes
|
||
|
_update(_state, _w, _input);
|
||
|
|
||
|
// compact input buffer every 2K or if empty
|
||
|
if(_input.read > 2048 || _input.length() === 0) {
|
||
|
_input.compact();
|
||
|
}
|
||
|
|
||
|
return md;
|
||
|
};
|
||
|
|
||
|
/**
|
||
|
* Produces the digest.
|
||
|
*
|
||
|
* @return a byte buffer containing the digest value.
|
||
|
*/
|
||
|
md.digest = function() {
|
||
|
/* Note: Here we copy the remaining bytes in the input buffer and
|
||
|
add the appropriate SHA-1 padding. Then we do the final update
|
||
|
on a copy of the state so that if the user wants to get
|
||
|
intermediate digests they can do so. */
|
||
|
|
||
|
/* Determine the number of bytes that must be added to the message
|
||
|
to ensure its length is congruent to 448 mod 512. In other words,
|
||
|
the data to be digested must be a multiple of 512 bits (or 128 bytes).
|
||
|
This data includes the message, some padding, and the length of the
|
||
|
message. Since the length of the message will be encoded as 8 bytes (64
|
||
|
bits), that means that the last segment of the data must have 56 bytes
|
||
|
(448 bits) of message and padding. Therefore, the length of the message
|
||
|
plus the padding must be congruent to 448 mod 512 because
|
||
|
512 - 128 = 448.
|
||
|
|
||
|
In order to fill up the message length it must be filled with
|
||
|
padding that begins with 1 bit followed by all 0 bits. Padding
|
||
|
must *always* be present, so if the message length is already
|
||
|
congruent to 448 mod 512, then 512 padding bits must be added. */
|
||
|
|
||
|
var finalBlock = forge.util.createBuffer();
|
||
|
finalBlock.putBytes(_input.bytes());
|
||
|
|
||
|
// compute remaining size to be digested (include message length size)
|
||
|
var remaining = (
|
||
|
md.fullMessageLength[md.fullMessageLength.length - 1] +
|
||
|
md.messageLengthSize);
|
||
|
|
||
|
// add padding for overflow blockSize - overflow
|
||
|
// _padding starts with 1 byte with first bit is set (byte value 128), then
|
||
|
// there may be up to (blockSize - 1) other pad bytes
|
||
|
var overflow = remaining & (md.blockLength - 1);
|
||
|
finalBlock.putBytes(_padding.substr(0, md.blockLength - overflow));
|
||
|
|
||
|
// serialize message length in bits in big-endian order; since length
|
||
|
// is stored in bytes we multiply by 8 and add carry from next int
|
||
|
var next, carry;
|
||
|
var bits = md.fullMessageLength[0] * 8;
|
||
|
for(var i = 0; i < md.fullMessageLength.length - 1; ++i) {
|
||
|
next = md.fullMessageLength[i + 1] * 8;
|
||
|
carry = (next / 0x100000000) >>> 0;
|
||
|
bits += carry;
|
||
|
finalBlock.putInt32(bits >>> 0);
|
||
|
bits = next >>> 0;
|
||
|
}
|
||
|
finalBlock.putInt32(bits);
|
||
|
|
||
|
var s2 = {
|
||
|
h0: _state.h0,
|
||
|
h1: _state.h1,
|
||
|
h2: _state.h2,
|
||
|
h3: _state.h3,
|
||
|
h4: _state.h4
|
||
|
};
|
||
|
_update(s2, _w, finalBlock);
|
||
|
var rval = forge.util.createBuffer();
|
||
|
rval.putInt32(s2.h0);
|
||
|
rval.putInt32(s2.h1);
|
||
|
rval.putInt32(s2.h2);
|
||
|
rval.putInt32(s2.h3);
|
||
|
rval.putInt32(s2.h4);
|
||
|
return rval;
|
||
|
};
|
||
|
|
||
|
return md;
|
||
|
};
|
||
|
|
||
|
// sha-1 padding bytes not initialized yet
|
||
|
var _padding = null;
|
||
|
var _initialized = false;
|
||
|
|
||
|
/**
|
||
|
* Initializes the constant tables.
|
||
|
*/
|
||
|
function _init() {
|
||
|
// create padding
|
||
|
_padding = String.fromCharCode(128);
|
||
|
_padding += forge.util.fillString(String.fromCharCode(0x00), 64);
|
||
|
|
||
|
// now initialized
|
||
|
_initialized = true;
|
||
|
}
|
||
|
|
||
|
/**
|
||
|
* Updates a SHA-1 state with the given byte buffer.
|
||
|
*
|
||
|
* @param s the SHA-1 state to update.
|
||
|
* @param w the array to use to store words.
|
||
|
* @param bytes the byte buffer to update with.
|
||
|
*/
|
||
|
function _update(s, w, bytes) {
|
||
|
// consume 512 bit (64 byte) chunks
|
||
|
var t, a, b, c, d, e, f, i;
|
||
|
var len = bytes.length();
|
||
|
while(len >= 64) {
|
||
|
// the w array will be populated with sixteen 32-bit big-endian words
|
||
|
// and then extended into 80 32-bit words according to SHA-1 algorithm
|
||
|
// and for 32-79 using Max Locktyukhin's optimization
|
||
|
|
||
|
// initialize hash value for this chunk
|
||
|
a = s.h0;
|
||
|
b = s.h1;
|
||
|
c = s.h2;
|
||
|
d = s.h3;
|
||
|
e = s.h4;
|
||
|
|
||
|
// round 1
|
||
|
for(i = 0; i < 16; ++i) {
|
||
|
t = bytes.getInt32();
|
||
|
w[i] = t;
|
||
|
f = d ^ (b & (c ^ d));
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0x5A827999 + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
for(; i < 20; ++i) {
|
||
|
t = (w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16]);
|
||
|
t = (t << 1) | (t >>> 31);
|
||
|
w[i] = t;
|
||
|
f = d ^ (b & (c ^ d));
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0x5A827999 + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
// round 2
|
||
|
for(; i < 32; ++i) {
|
||
|
t = (w[i - 3] ^ w[i - 8] ^ w[i - 14] ^ w[i - 16]);
|
||
|
t = (t << 1) | (t >>> 31);
|
||
|
w[i] = t;
|
||
|
f = b ^ c ^ d;
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0x6ED9EBA1 + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
for(; i < 40; ++i) {
|
||
|
t = (w[i - 6] ^ w[i - 16] ^ w[i - 28] ^ w[i - 32]);
|
||
|
t = (t << 2) | (t >>> 30);
|
||
|
w[i] = t;
|
||
|
f = b ^ c ^ d;
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0x6ED9EBA1 + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
// round 3
|
||
|
for(; i < 60; ++i) {
|
||
|
t = (w[i - 6] ^ w[i - 16] ^ w[i - 28] ^ w[i - 32]);
|
||
|
t = (t << 2) | (t >>> 30);
|
||
|
w[i] = t;
|
||
|
f = (b & c) | (d & (b ^ c));
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0x8F1BBCDC + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
// round 4
|
||
|
for(; i < 80; ++i) {
|
||
|
t = (w[i - 6] ^ w[i - 16] ^ w[i - 28] ^ w[i - 32]);
|
||
|
t = (t << 2) | (t >>> 30);
|
||
|
w[i] = t;
|
||
|
f = b ^ c ^ d;
|
||
|
t = ((a << 5) | (a >>> 27)) + f + e + 0xCA62C1D6 + t;
|
||
|
e = d;
|
||
|
d = c;
|
||
|
// `>>> 0` necessary to avoid iOS/Safari 10 optimization bug
|
||
|
c = ((b << 30) | (b >>> 2)) >>> 0;
|
||
|
b = a;
|
||
|
a = t;
|
||
|
}
|
||
|
|
||
|
// update hash state
|
||
|
s.h0 = (s.h0 + a) | 0;
|
||
|
s.h1 = (s.h1 + b) | 0;
|
||
|
s.h2 = (s.h2 + c) | 0;
|
||
|
s.h3 = (s.h3 + d) | 0;
|
||
|
s.h4 = (s.h4 + e) | 0;
|
||
|
|
||
|
len -= 64;
|
||
|
}
|
||
|
}
|