crypto/hmac: speed up repeated operations with the same key

Speed up repeated HMAC operations with the same key by not recomputing
the first block of the inner and outer hashes in Reset and Sum, saving
two block computations each time.

This is a significant win for applications which hash many small
messages with the same key. In x/crypto/pbkdf2 for example, this
optimization cuts the number of block computations in half, speeding it
up by 25%-40% depending on the hash function.

The hash function needs to implement binary.Marshaler and
binary.Unmarshaler for this optimization to work, so that we can save
and restore its internal state. All hash functions in the standard
library are marshalable (CL 66710) but if the hash isn't marshalable, we
fall back on the old behaviour.

Marshaling the hashes does add a couple unavoidable new allocations, but
this only has to be done once, so the cost is amortized over repeated
uses. To minimize impact to applications which don't (or can't) reuse
hmac objects, marshaling is performed in Reset (rather than in New),
since calling Reset seems like a good indication that the caller intends
to reuse the hmac object later.

I had to add a boolean field to the hmac state to remember if we've
marshaled the hashes or not. This is paid for by removing the size and
blocksize fields, which were basically unused except for some
initialization work in New, and to fulfill the Size and Blocksize
methods. Size and Blocksize can just be forwarded to the underlying
hash, so there doesn't really seem to be any reason to waste space
caching their values.

crypto/hmac benchmarks:

name                    old time/op    new time/op     delta
HMAC_Reset/SHA1/1K-2      4.06µs ± 0%     3.77µs ± 0%   -7.29%         (p=0.000 n=8+10)
HMAC_Reset/SHA1/32-2      1.08µs ± 0%     0.78µs ± 1%  -27.67%        (p=0.000 n=10+10)
HMAC_Reset/SHA256/1K-2    10.3µs ± 0%      9.4µs ± 0%   -9.03%        (p=0.000 n=10+10)
HMAC_Reset/SHA256/32-2    2.32µs ± 0%     1.42µs ± 0%  -38.87%        (p=0.000 n=10+10)
HMAC_Reset/SHA512/1K-2    8.22µs ± 0%     7.04µs ± 0%  -14.32%          (p=0.000 n=9+9)
HMAC_Reset/SHA512/32-2    3.08µs ± 0%     1.89µs ± 0%  -38.54%         (p=0.000 n=10+9)
HMAC_New/SHA1/1K-2        4.86µs ± 1%     4.93µs ± 1%   +1.30%         (p=0.000 n=10+9)
HMAC_New/SHA1/32-2        1.91µs ± 1%     1.95µs ± 1%   +1.84%         (p=0.000 n=10+9)
HMAC_New/SHA256/1K-2      11.2µs ± 1%     11.2µs ± 0%     ~            (p=1.000 n=9+10)
HMAC_New/SHA256/32-2      3.22µs ± 2%     3.19µs ± 2%   -1.07%         (p=0.018 n=9+10)
HMAC_New/SHA512/1K-2      9.54µs ± 0%     9.66µs ± 1%   +1.31%         (p=0.000 n=9+10)
HMAC_New/SHA512/32-2      4.37µs ± 1%     4.46µs ± 1%   +1.97%         (p=0.000 n=10+9)

name                    old speed      new speed       delta
HMAC_Reset/SHA1/1K-2     252MB/s ± 0%    272MB/s ± 0%   +7.86%         (p=0.000 n=8+10)
HMAC_Reset/SHA1/32-2    29.7MB/s ± 0%   41.1MB/s ± 1%  +38.26%        (p=0.000 n=10+10)
HMAC_Reset/SHA256/1K-2  99.1MB/s ± 0%  108.9MB/s ± 0%   +9.93%        (p=0.000 n=10+10)
HMAC_Reset/SHA256/32-2  13.8MB/s ± 0%   22.6MB/s ± 0%  +63.57%        (p=0.000 n=10+10)
HMAC_Reset/SHA512/1K-2   125MB/s ± 0%    145MB/s ± 0%  +16.71%          (p=0.000 n=9+9)
HMAC_Reset/SHA512/32-2  10.4MB/s ± 0%   16.9MB/s ± 0%  +62.69%         (p=0.000 n=10+9)
HMAC_New/SHA1/1K-2       211MB/s ± 1%    208MB/s ± 1%   -1.29%         (p=0.000 n=10+9)
HMAC_New/SHA1/32-2      16.7MB/s ± 1%   16.4MB/s ± 1%   -1.81%         (p=0.000 n=10+9)
HMAC_New/SHA256/1K-2    91.3MB/s ± 1%   91.5MB/s ± 0%     ~            (p=0.950 n=9+10)
HMAC_New/SHA256/32-2    9.94MB/s ± 2%  10.04MB/s ± 2%   +1.09%         (p=0.021 n=9+10)
HMAC_New/SHA512/1K-2     107MB/s ± 0%    106MB/s ± 1%   -1.29%         (p=0.000 n=9+10)
HMAC_New/SHA512/32-2    7.32MB/s ± 1%   7.18MB/s ± 1%   -1.89%         (p=0.000 n=10+9)

name                    old alloc/op   new alloc/op    delta
HMAC_Reset/SHA1/1K-2      0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA1/32-2      0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA256/1K-2    0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA256/32-2    0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA512/1K-2    0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA512/32-2    0.00B ±NaN%     0.00B ±NaN%     ~     (all samples are equal)
HMAC_New/SHA1/1K-2          448B ± 0%       448B ± 0%     ~     (all samples are equal)
HMAC_New/SHA1/32-2          448B ± 0%       448B ± 0%     ~     (all samples are equal)
HMAC_New/SHA256/1K-2        480B ± 0%       480B ± 0%     ~     (all samples are equal)
HMAC_New/SHA256/32-2        480B ± 0%       480B ± 0%     ~     (all samples are equal)
HMAC_New/SHA512/1K-2        800B ± 0%       800B ± 0%     ~     (all samples are equal)
HMAC_New/SHA512/32-2        800B ± 0%       800B ± 0%     ~     (all samples are equal)

name                    old allocs/op  new allocs/op   delta
HMAC_Reset/SHA1/1K-2       0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA1/32-2       0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA256/1K-2     0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA256/32-2     0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA512/1K-2     0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_Reset/SHA512/32-2     0.00 ±NaN%      0.00 ±NaN%     ~     (all samples are equal)
HMAC_New/SHA1/1K-2          5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)
HMAC_New/SHA1/32-2          5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)
HMAC_New/SHA256/1K-2        5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)
HMAC_New/SHA256/32-2        5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)
HMAC_New/SHA512/1K-2        5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)
HMAC_New/SHA512/32-2        5.00 ± 0%       5.00 ± 0%     ~     (all samples are equal)

x/crypto/pbkdf2 benchmarks:

name          old time/op    new time/op    delta
HMACSHA1-2      4.63ms ± 0%    3.40ms ± 0%  -26.58%   (p=0.000 n=10+9)
HMACSHA256-2    9.75ms ± 0%    5.98ms ± 0%  -38.62%   (p=0.000 n=9+10)

name          old alloc/op   new alloc/op   delta
HMACSHA1-2        516B ± 0%      708B ± 0%  +37.21%  (p=0.000 n=10+10)
HMACSHA256-2      549B ± 0%      772B ± 0%  +40.62%  (p=0.000 n=10+10)

name          old allocs/op  new allocs/op  delta
HMACSHA1-2        8.00 ± 0%     10.00 ± 0%  +25.00%  (p=0.000 n=10+10)
HMACSHA256-2      8.00 ± 0%     10.00 ± 0%  +25.00%  (p=0.000 n=10+10)

Fixes #19941

Change-Id: I7077a6f875be68d3da05f7b3664e18514861886f
Reviewed-on: https://go-review.googlesource.com/c/go/+/27458
Run-TryBot: Emmanuel Odeke <emm.odeke@gmail.com>
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Filippo Valsorda <filippo@golang.org>

src/crypto/hmac/hmac.go

@@ -34,18 +34,36 @@ import (
 // opad = 0x5c byte repeated for key length
 // hmac = H([key ^ opad] H([key ^ ipad] text))
 
+// Marshalable is the combination of encoding.BinaryMarshaler and
+// encoding.BinaryUnmarshaler. Their method definitions are repeated here to
+// avoid a dependency on the encoding package.
+type marshalable interface {
+	MarshalBinary() ([]byte, error)
+	UnmarshalBinary([]byte) error
+}
+
 type hmac struct {
-	size         int
-	blocksize    int
 	opad, ipad   []byte
 	outer, inner hash.Hash
+
+	// If marshaled is true, then opad and ipad do not contain a padded
+	// copy of the key, but rather the marshaled state of outer/inner after
+	// opad/ipad has been fed into it.
+	marshaled bool
 }
 
 func (h *hmac) Sum(in []byte) []byte {
 	origLen := len(in)
 	in = h.inner.Sum(in)
-	h.outer.Reset()
+
-	h.outer.Write(h.opad)
+	if h.marshaled {
+		if err := h.outer.(marshalable).UnmarshalBinary(h.opad); err != nil {
+			panic(err)
+		}
+	} else {
+		h.outer.Reset()
+		h.outer.Write(h.opad)
+	}
 	h.outer.Write(in[origLen:])
 	return h.outer.Sum(in[:origLen])
 }
@@ -54,13 +72,51 @@ func (h *hmac) Write(p []byte) (n int, err error) {
 	return h.inner.Write(p)
 }
 
-func (h *hmac) Size() int { return h.size }
+func (h *hmac) Size() int      { return h.outer.Size() }
-
+func (h *hmac) BlockSize() int { return h.inner.BlockSize() }
-func (h *hmac) BlockSize() int { return h.blocksize }
 
 func (h *hmac) Reset() {
+	if h.marshaled {
+		if err := h.inner.(marshalable).UnmarshalBinary(h.ipad); err != nil {
+			panic(err)
+		}
+		return
+	}
+
 	h.inner.Reset()
 	h.inner.Write(h.ipad)
+
+	// If the underlying hash is marshalable, we can save some time by
+	// saving a copy of the hash state now, and restoring it on future
+	// calls to Reset and Sum instead of writing ipad/opad every time.
+	//
+	// If either hash is unmarshalable for whatever reason,
+	// it's safe to bail out here.
+	marshalableInner, innerOK := h.inner.(marshalable)
+	if !innerOK {
+		return
+	}
+	marshalableOuter, outerOK := h.outer.(marshalable)
+	if !outerOK {
+		return
+	}
+
+	imarshal, err := marshalableInner.MarshalBinary()
+	if err != nil {
+		return
+	}
+
+	h.outer.Reset()
+	h.outer.Write(h.opad)
+	omarshal, err := marshalableOuter.MarshalBinary()
+	if err != nil {
+		return
+	}
+
+	// Marshaling succeeded; save the marshaled state for later
+	h.ipad = imarshal
+	h.opad = omarshal
+	h.marshaled = true
 }
 
 // New returns a new HMAC hash using the given hash.Hash type and key.
@@ -71,11 +127,10 @@ func New(h func() hash.Hash, key []byte) hash.Hash {
 	hm := new(hmac)
 	hm.outer = h()
 	hm.inner = h()
-	hm.size = hm.inner.Size()
+	blocksize := hm.inner.BlockSize()
-	hm.blocksize = hm.inner.BlockSize()
+	hm.ipad = make([]byte, blocksize)
-	hm.ipad = make([]byte, hm.blocksize)
+	hm.opad = make([]byte, blocksize)
-	hm.opad = make([]byte, hm.blocksize)
+	if len(key) > blocksize {
-	if len(key) > hm.blocksize {
 		// If key is too big, hash it.
 		hm.outer.Write(key)
 		key = hm.outer.Sum(nil)
@@ -89,6 +144,7 @@ func New(h func() hash.Hash, key []byte) hash.Hash {
 		hm.opad[i] ^= 0x5c
 	}
 	hm.inner.Write(hm.ipad)
+
 	return hm
 }
 

src/crypto/hmac/hmac_test.go

@@ -529,7 +529,7 @@ func TestHMAC(t *testing.T) {
 		if b := h.BlockSize(); b != tt.blocksize {
 			t.Errorf("BlockSize: got %v, want %v", b, tt.blocksize)
 		}
-		for j := 0; j < 2; j++ {
+		for j := 0; j < 4; j++ {
 			n, err := h.Write(tt.in)
 			if n != len(tt.in) || err != nil {
 				t.Errorf("test %d.%d: Write(%d) = %d, %v", i, j, len(tt.in), n, err)
@@ -546,10 +546,21 @@ func TestHMAC(t *testing.T) {
 
 			// Second iteration: make sure reset works.
 			h.Reset()
+
+			// Third and fourth iteration: make sure hmac works on
+			// hashes without MarshalBinary/UnmarshalBinary
+			if j == 1 {
+				h = New(func() hash.Hash { return justHash{tt.hash()} }, tt.key)
+			}
 		}
 	}
 }
 
+// justHash implements just the hash.Hash methods and nothing else
+type justHash struct {
+	hash.Hash
+}
+
 func TestEqual(t *testing.T) {
 	a := []byte("test")
 	b := []byte("test1")