Moved sha3 from `obscuren`

2 files changed, 387 insertions, 0 deletions

diff --git a/crypto/sha3/keccakf.go b/crypto/sha3/keccakf.go
new file mode 100644
index 000000000..3baf13ba3
--- /dev/null
+++ b/crypto/sha3/keccakf.go
@@ -0,0 +1,171 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package sha3
+
+// This file implements the core Keccak permutation function necessary for computing SHA3.
+// This is implemented in a separate file to allow for replacement by an optimized implementation.
+// Nothing in this package is exported.
+// For the detailed specification, refer to the Keccak web site (http://keccak.noekeon.org/).
+
+// rc stores the round constants for use in the ι step.
+var rc = [...]uint64{
+        0x0000000000000001,
+        0x0000000000008082,
+        0x800000000000808A,
+        0x8000000080008000,
+        0x000000000000808B,
+        0x0000000080000001,
+        0x8000000080008081,
+        0x8000000000008009,
+        0x000000000000008A,
+        0x0000000000000088,
+        0x0000000080008009,
+        0x000000008000000A,
+        0x000000008000808B,
+        0x800000000000008B,
+        0x8000000000008089,
+        0x8000000000008003,
+        0x8000000000008002,
+        0x8000000000000080,
+        0x000000000000800A,
+        0x800000008000000A,
+        0x8000000080008081,
+        0x8000000000008080,
+        0x0000000080000001,
+        0x8000000080008008,
+}
+
+// ro_xx represent the rotation offsets for use in the χ step.
+// Defining them as const instead of in an array allows the compiler to insert constant shifts.
+const (
+        ro_00 = 0
+        ro_01 = 36
+        ro_02 = 3
+        ro_03 = 41
+        ro_04 = 18
+        ro_05 = 1
+        ro_06 = 44
+        ro_07 = 10
+        ro_08 = 45
+        ro_09 = 2
+        ro_10 = 62
+        ro_11 = 6
+        ro_12 = 43
+        ro_13 = 15
+        ro_14 = 61
+        ro_15 = 28
+        ro_16 = 55
+        ro_17 = 25
+        ro_18 = 21
+        ro_19 = 56
+        ro_20 = 27
+        ro_21 = 20
+        ro_22 = 39
+        ro_23 = 8
+        ro_24 = 14
+)
+
+// keccakF computes the complete Keccak-f function consisting of 24 rounds with a different
+// constant (rc) in each round. This implementation fully unrolls the round function to avoid
+// inner loops, as well as pre-calculating shift offsets.
+func (d *digest) keccakF() {
+        for _, roundConstant := range rc {
+                // θ step
+                d.c[0] = d.a[0] ^ d.a[5] ^ d.a[10] ^ d.a[15] ^ d.a[20]
+                d.c[1] = d.a[1] ^ d.a[6] ^ d.a[11] ^ d.a[16] ^ d.a[21]
+                d.c[2] = d.a[2] ^ d.a[7] ^ d.a[12] ^ d.a[17] ^ d.a[22]
+                d.c[3] = d.a[3] ^ d.a[8] ^ d.a[13] ^ d.a[18] ^ d.a[23]
+                d.c[4] = d.a[4] ^ d.a[9] ^ d.a[14] ^ d.a[19] ^ d.a[24]
+
+                d.d[0] = d.c[4] ^ (d.c[1]<<1 ^ d.c[1]>>63)
+                d.d[1] = d.c[0] ^ (d.c[2]<<1 ^ d.c[2]>>63)
+                d.d[2] = d.c[1] ^ (d.c[3]<<1 ^ d.c[3]>>63)
+                d.d[3] = d.c[2] ^ (d.c[4]<<1 ^ d.c[4]>>63)
+                d.d[4] = d.c[3] ^ (d.c[0]<<1 ^ d.c[0]>>63)
+
+                d.a[0] ^= d.d[0]
+                d.a[1] ^= d.d[1]
+                d.a[2] ^= d.d[2]
+                d.a[3] ^= d.d[3]
+                d.a[4] ^= d.d[4]
+                d.a[5] ^= d.d[0]
+                d.a[6] ^= d.d[1]
+                d.a[7] ^= d.d[2]
+                d.a[8] ^= d.d[3]
+                d.a[9] ^= d.d[4]
+                d.a[10] ^= d.d[0]
+                d.a[11] ^= d.d[1]
+                d.a[12] ^= d.d[2]
+                d.a[13] ^= d.d[3]
+                d.a[14] ^= d.d[4]
+                d.a[15] ^= d.d[0]
+                d.a[16] ^= d.d[1]
+                d.a[17] ^= d.d[2]
+                d.a[18] ^= d.d[3]
+                d.a[19] ^= d.d[4]
+                d.a[20] ^= d.d[0]
+                d.a[21] ^= d.d[1]
+                d.a[22] ^= d.d[2]
+                d.a[23] ^= d.d[3]
+                d.a[24] ^= d.d[4]
+
+                // ρ and π steps
+                d.b[0] = d.a[0]
+                d.b[1] = d.a[6]<<ro_06 ^ d.a[6]>>(64-ro_06)
+                d.b[2] = d.a[12]<<ro_12 ^ d.a[12]>>(64-ro_12)
+                d.b[3] = d.a[18]<<ro_18 ^ d.a[18]>>(64-ro_18)
+                d.b[4] = d.a[24]<<ro_24 ^ d.a[24]>>(64-ro_24)
+                d.b[5] = d.a[3]<<ro_15 ^ d.a[3]>>(64-ro_15)
+                d.b[6] = d.a[9]<<ro_21 ^ d.a[9]>>(64-ro_21)
+                d.b[7] = d.a[10]<<ro_02 ^ d.a[10]>>(64-ro_02)
+                d.b[8] = d.a[16]<<ro_08 ^ d.a[16]>>(64-ro_08)
+                d.b[9] = d.a[22]<<ro_14 ^ d.a[22]>>(64-ro_14)
+                d.b[10] = d.a[1]<<ro_05 ^ d.a[1]>>(64-ro_05)
+                d.b[11] = d.a[7]<<ro_11 ^ d.a[7]>>(64-ro_11)
+                d.b[12] = d.a[13]<<ro_17 ^ d.a[13]>>(64-ro_17)
+                d.b[13] = d.a[19]<<ro_23 ^ d.a[19]>>(64-ro_23)
+                d.b[14] = d.a[20]<<ro_04 ^ d.a[20]>>(64-ro_04)
+                d.b[15] = d.a[4]<<ro_20 ^ d.a[4]>>(64-ro_20)
+                d.b[16] = d.a[5]<<ro_01 ^ d.a[5]>>(64-ro_01)
+                d.b[17] = d.a[11]<<ro_07 ^ d.a[11]>>(64-ro_07)
+                d.b[18] = d.a[17]<<ro_13 ^ d.a[17]>>(64-ro_13)
+                d.b[19] = d.a[23]<<ro_19 ^ d.a[23]>>(64-ro_19)
+                d.b[20] = d.a[2]<<ro_10 ^ d.a[2]>>(64-ro_10)
+                d.b[21] = d.a[8]<<ro_16 ^ d.a[8]>>(64-ro_16)
+                d.b[22] = d.a[14]<<ro_22 ^ d.a[14]>>(64-ro_22)
+                d.b[23] = d.a[15]<<ro_03 ^ d.a[15]>>(64-ro_03)
+                d.b[24] = d.a[21]<<ro_09 ^ d.a[21]>>(64-ro_09)
+
+                // χ step
+                d.a[0] = d.b[0] ^ (^d.b[1] & d.b[2])
+                d.a[1] = d.b[1] ^ (^d.b[2] & d.b[3])
+                d.a[2] = d.b[2] ^ (^d.b[3] & d.b[4])
+                d.a[3] = d.b[3] ^ (^d.b[4] & d.b[0])
+                d.a[4] = d.b[4] ^ (^d.b[0] & d.b[1])
+                d.a[5] = d.b[5] ^ (^d.b[6] & d.b[7])
+                d.a[6] = d.b[6] ^ (^d.b[7] & d.b[8])
+                d.a[7] = d.b[7] ^ (^d.b[8] & d.b[9])
+                d.a[8] = d.b[8] ^ (^d.b[9] & d.b[5])
+                d.a[9] = d.b[9] ^ (^d.b[5] & d.b[6])
+                d.a[10] = d.b[10] ^ (^d.b[11] & d.b[12])
+                d.a[11] = d.b[11] ^ (^d.b[12] & d.b[13])
+                d.a[12] = d.b[12] ^ (^d.b[13] & d.b[14])
+                d.a[13] = d.b[13] ^ (^d.b[14] & d.b[10])
+                d.a[14] = d.b[14] ^ (^d.b[10] & d.b[11])
+                d.a[15] = d.b[15] ^ (^d.b[16] & d.b[17])
+                d.a[16] = d.b[16] ^ (^d.b[17] & d.b[18])
+                d.a[17] = d.b[17] ^ (^d.b[18] & d.b[19])
+                d.a[18] = d.b[18] ^ (^d.b[19] & d.b[15])
+                d.a[19] = d.b[19] ^ (^d.b[15] & d.b[16])
+                d.a[20] = d.b[20] ^ (^d.b[21] & d.b[22])
+                d.a[21] = d.b[21] ^ (^d.b[22] & d.b[23])
+                d.a[22] = d.b[22] ^ (^d.b[23] & d.b[24])
+                d.a[23] = d.b[23] ^ (^d.b[24] & d.b[20])
+                d.a[24] = d.b[24] ^ (^d.b[20] & d.b[21])
+
+                // ι step
+                d.a[0] ^= roundConstant
+        }
+}
diff --git a/crypto/sha3/sha3.go b/crypto/sha3/sha3.go
new file mode 100644
index 000000000..22df0ef11
--- /dev/null
+++ b/crypto/sha3/sha3.go
@@ -0,0 +1,216 @@
+// Copyright 2013 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+// Package sha3 implements the SHA3 hash algorithm (formerly called Keccak) chosen by NIST in 2012.
+// This file provides a SHA3 implementation which implements the standard hash.Hash interface.
+// Writing input data, including padding, and reading output data are computed in this file.
+// Note that the current implementation can compute the hash of an integral number of bytes only.
+// This is a consequence of the hash interface in which a buffer of bytes is passed in.
+// The internals of the Keccak-f function are computed in keccakf.go.
+// For the detailed specification, refer to the Keccak web site (http://keccak.noekeon.org/).
+package sha3
+
+import (
+	"encoding/binary"
+	"hash"
+)
+
+// laneSize is the size in bytes of each "lane" of the internal state of SHA3 (5 * 5 * 8).
+// Note that changing this size would requires using a type other than uint64 to store each lane.
+const laneSize = 8
+
+// sliceSize represents the dimensions of the internal state, a square matrix of
+// sliceSize ** 2 lanes. This is the size of both the "rows" and "columns" dimensions in the
+// terminology of the SHA3 specification.
+const sliceSize = 5
+
+// numLanes represents the total number of lanes in the state.
+const numLanes = sliceSize * sliceSize
+
+// stateSize is the size in bytes of the internal state of SHA3 (5 * 5 * WSize).
+const stateSize = laneSize * numLanes
+
+// digest represents the partial evaluation of a checksum.
+// Note that capacity, and not outputSize, is the critical security parameter, as SHA3 can output
+// an arbitrary number of bytes for any given capacity. The Keccak proposal recommends that
+// capacity = 2*outputSize to ensure that finding a collision of size outputSize requires
+// O(2^{outputSize/2}) computations (the birthday lower bound). Future standards may modify the
+// capacity/outputSize ratio to allow for more output with lower cryptographic security.
+type digest struct {
+	a          [numLanes]uint64  // main state of the hash
+	b          [numLanes]uint64  // intermediate states
+	c          [sliceSize]uint64 // intermediate states
+	d          [sliceSize]uint64 // intermediate states
+	outputSize int               // desired output size in bytes
+	capacity   int               // number of bytes to leave untouched during squeeze/absorb
+	absorbed   int               // number of bytes absorbed thus far
+}
+
+// minInt returns the lesser of two integer arguments, to simplify the absorption routine.
+func minInt(v1, v2 int) int {
+	if v1 <= v2 {
+		return v1
+	}
+	return v2
+}
+
+// rate returns the number of bytes of the internal state which can be absorbed or squeezed
+// in between calls to the permutation function.
+func (d *digest) rate() int {
+	return stateSize - d.capacity
+}
+
+// Reset clears the internal state by zeroing bytes in the state buffer.
+// This can be skipped for a newly-created hash state; the default zero-allocated state is correct.
+func (d *digest) Reset() {
+	d.absorbed = 0
+	for i := range d.a {
+		d.a[i] = 0
+	}
+}
+
+// BlockSize, required by the hash.Hash interface, does not have a standard intepretation
+// for a sponge-based construction like SHA3. We return the data rate: the number of bytes which
+// can be absorbed per invocation of the permutation function. For Merkle-Damgård based hashes
+// (ie SHA1, SHA2, MD5) the output size of the internal compression function is returned.
+// We consider this to be roughly equivalent because it represents the number of bytes of output
+// produced per cryptographic operation.
+func (d *digest) BlockSize() int { return d.rate() }
+
+// Size returns the output size of the hash function in bytes.
+func (d *digest) Size() int {
+	return d.outputSize
+}
+
+// unalignedAbsorb is a helper function for Write, which absorbs data that isn't aligned with an
+// 8-byte lane. This requires shifting the individual bytes into position in a uint64.
+func (d *digest) unalignedAbsorb(p []byte) {
+	var t uint64
+	for i := len(p) - 1; i >= 0; i-- {
+		t <<= 8
+		t |= uint64(p[i])
+	}
+	offset := (d.absorbed) % d.rate()
+	t <<= 8 * uint(offset%laneSize)
+	d.a[offset/laneSize] ^= t
+	d.absorbed += len(p)
+}
+
+// Write "absorbs" bytes into the state of the SHA3 hash, updating as needed when the sponge
+// "fills up" with rate() bytes. Since lanes are stored internally as type uint64, this requires
+// converting the incoming bytes into uint64s using a little endian interpretation. This
+// implementation is optimized for large, aligned writes of multiples of 8 bytes (laneSize).
+// Non-aligned or uneven numbers of bytes require shifting and are slower.
+func (d *digest) Write(p []byte) (int, error) {
+	// An initial offset is needed if the we aren't absorbing to the first lane initially.
+	offset := d.absorbed % d.rate()
+	toWrite := len(p)
+
+	// The first lane may need to absorb unaligned and/or incomplete data.
+	if (offset%laneSize != 0 || len(p) < 8) && len(p) > 0 {
+		toAbsorb := minInt(laneSize-(offset%laneSize), len(p))
+		d.unalignedAbsorb(p[:toAbsorb])
+		p = p[toAbsorb:]
+		offset = (d.absorbed) % d.rate()
+
+		// For every rate() bytes absorbed, the state must be permuted via the F Function.
+		if (d.absorbed)%d.rate() == 0 {
+			d.keccakF()
+		}
+	}
+
+	// This loop should absorb the bulk of the data into full, aligned lanes.
+	// It will call the update function as necessary.
+	for len(p) > 7 {
+		firstLane := offset / laneSize
+		lastLane := minInt(d.rate()/laneSize, firstLane+len(p)/laneSize)
+
+		// This inner loop absorbs input bytes into the state in groups of 8, converted to uint64s.
+		for lane := firstLane; lane < lastLane; lane++ {
+			d.a[lane] ^= binary.LittleEndian.Uint64(p[:laneSize])
+			p = p[laneSize:]
+		}
+		d.absorbed += (lastLane - firstLane) * laneSize
+		// For every rate() bytes absorbed, the state must be permuted via the F Function.
+		if (d.absorbed)%d.rate() == 0 {
+			d.keccakF()
+		}
+
+		offset = 0
+	}
+
+	// If there are insufficient bytes to fill the final lane, an unaligned absorption.
+	// This should always start at a correct lane boundary though, or else it would be caught
+	// by the uneven opening lane case above.
+	if len(p) > 0 {
+		d.unalignedAbsorb(p)
+	}
+
+	return toWrite, nil
+}
+
+// pad computes the SHA3 padding scheme based on the number of bytes absorbed.
+// The padding is a 1 bit, followed by an arbitrary number of 0s and then a final 1 bit, such that
+// the input bits plus padding bits are a multiple of rate(). Adding the padding simply requires
+// xoring an opening and closing bit into the appropriate lanes.
+func (d *digest) pad() {
+	offset := d.absorbed % d.rate()
+	// The opening pad bit must be shifted into position based on the number of bytes absorbed
+	padOpenLane := offset / laneSize
+	d.a[padOpenLane] ^= 0x0000000000000001 << uint(8*(offset%laneSize))
+	// The closing padding bit is always in the last position
+	padCloseLane := (d.rate() / laneSize) - 1
+	d.a[padCloseLane] ^= 0x8000000000000000
+}
+
+// finalize prepares the hash to output data by padding and one final permutation of the state.
+func (d *digest) finalize() {
+	d.pad()
+	d.keccakF()
+}
+
+// squeeze outputs an arbitrary number of bytes from the hash state.
+// Squeezing can require multiple calls to the F function (one per rate() bytes squeezed),
+// although this is not the case for standard SHA3 parameters. This implementation only supports
+// squeezing a single time, subsequent squeezes may lose alignment. Future implementations
+// may wish to support multiple squeeze calls, for example to support use as a PRNG.
+func (d *digest) squeeze(in []byte, toSqueeze int) []byte {
+	// Because we read in blocks of laneSize, we need enough room to read
+	// an integral number of lanes
+	needed := toSqueeze + (laneSize-toSqueeze%laneSize)%laneSize
+	if cap(in)-len(in) < needed {
+		newIn := make([]byte, len(in), len(in)+needed)
+		copy(newIn, in)
+		in = newIn
+	}
+	out := in[len(in) : len(in)+needed]
+
+	for len(out) > 0 {
+		for i := 0; i < d.rate() && len(out) > 0; i += laneSize {
+			binary.LittleEndian.PutUint64(out[:], d.a[i/laneSize])
+			out = out[laneSize:]
+		}
+		if len(out) > 0 {
+			d.keccakF()
+		}
+	}
+	return in[:len(in)+toSqueeze] // Re-slice in case we wrote extra data.
+}
+
+// Sum applies padding to the hash state and then squeezes out the desired nubmer of output bytes.
+func (d *digest) Sum(in []byte) []byte {
+	// Make a copy of the original hash so that caller can keep writing and summing.
+	dup := *d
+	dup.finalize()
+	return dup.squeeze(in, dup.outputSize)
+}
+
+// The NewKeccakX constructors enable initializing a hash in any of the four recommend sizes
+// from the Keccak specification, all of which set capacity=2*outputSize. Note that the final
+// NIST standard for SHA3 may specify different input/output lengths.
+// The output size is indicated in bits but converted into bytes internally.
+func NewKeccak224() hash.Hash { return &digest{outputSize: 224 / 8, capacity: 2 * 224 / 8} }
+func NewKeccak256() hash.Hash { return &digest{outputSize: 256 / 8, capacity: 2 * 256 / 8} }
+func NewKeccak384() hash.Hash { return &digest{outputSize: 384 / 8, capacity: 2 * 384 / 8} }
+func NewKeccak512() hash.Hash { return &digest{outputSize: 512 / 8, capacity: 2 * 512 / 8} }