-
Notifications
You must be signed in to change notification settings - Fork 60
/
gcm_amd64.go
333 lines (277 loc) · 10.4 KB
/
gcm_amd64.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
package sm4
import (
"crypto/cipher"
"crypto/subtle"
"encoding/binary"
"errors"
smcipher "github.com/emmansun/gmsm/cipher"
)
// Assert that sm4CipherAsm implements the gcmAble interface.
var _ gcmAble = (*sm4CipherAsm)(nil)
// NewGCM returns the SM4 cipher wrapped in Galois Counter Mode. This is only
// called by crypto/cipher.NewGCM via the gcmAble interface.
func (c *sm4CipherAsm) NewGCM(nonceSize, tagSize int) (cipher.AEAD, error) {
var key [gcmBlockSize]byte
c.Encrypt(key[:], key[:])
g := &gcm{cipher: c, nonceSize: nonceSize, tagSize: tagSize}
// We precompute 16 multiples of |key|. However, when we do lookups
// into this table we'll be using bits from a field element and
// therefore the bits will be in the reverse order. So normally one
// would expect, say, 4*key to be in index 4 of the table but due to
// this bit ordering it will actually be in index 0010 (base 2) = 2.
x := gcmFieldElement{
binary.BigEndian.Uint64(key[:8]),
binary.BigEndian.Uint64(key[8:]),
}
g.productTable[reverseBits(1)] = x
for i := 2; i < 16; i += 2 {
g.productTable[reverseBits(i)] = gcmDouble(&g.productTable[reverseBits(i/2)])
g.productTable[reverseBits(i+1)] = gcmAdd(&g.productTable[reverseBits(i)], &x)
}
return g, nil
}
// gcmFieldElement represents a value in GF(2¹²⁸). In order to reflect the GCM
// standard and make binary.BigEndian suitable for marshaling these values, the
// bits are stored in big endian order. For example:
// the coefficient of x⁰ can be obtained by v.low >> 63.
// the coefficient of x⁶³ can be obtained by v.low & 1.
// the coefficient of x⁶⁴ can be obtained by v.high >> 63.
// the coefficient of x¹²⁷ can be obtained by v.high & 1.
type gcmFieldElement struct {
low, high uint64
}
// gcm represents a Galois Counter Mode with a specific key. See
// https://csrc.nist.gov/groups/ST/toolkit/BCM/documents/proposedmodes/gcm/gcm-revised-spec.pdf
type gcm struct {
cipher *sm4CipherAsm
nonceSize int
tagSize int
// productTable contains the first sixteen powers of the key, H.
// However, they are in bit reversed order. See NewGCMWithNonceSize.
productTable [16]gcmFieldElement
}
const (
gcmBlockSize = 16
gcmTagSize = 16
gcmMinimumTagSize = 12 // NIST SP 800-38D recommends tags with 12 or more bytes.
gcmStandardNonceSize = 12
)
func (g *gcm) NonceSize() int {
return g.nonceSize
}
func (g *gcm) Overhead() int {
return g.tagSize
}
func (g *gcm) Seal(dst, nonce, plaintext, data []byte) []byte {
if len(nonce) != g.nonceSize {
panic("crypto/cipher: incorrect nonce length given to GCM")
}
if uint64(len(plaintext)) > ((1<<32)-2)*uint64(g.cipher.BlockSize()) {
panic("crypto/cipher: message too large for GCM")
}
ret, out := smcipher.SliceForAppend(dst, len(plaintext)+g.tagSize)
if smcipher.InexactOverlap(out, plaintext) {
panic("crypto/cipher: invalid buffer overlap")
}
var counter, tagMask [gcmBlockSize]byte
g.deriveCounter(&counter, nonce)
g.cipher.Encrypt(tagMask[:], counter[:])
gcmInc32(&counter)
g.counterCrypt(out, plaintext, &counter)
var tag [gcmTagSize]byte
g.auth(tag[:], out[:len(plaintext)], data, &tagMask)
copy(out[len(plaintext):], tag[:])
return ret
}
var errOpen = errors.New("cipher: message authentication failed")
func (g *gcm) Open(dst, nonce, ciphertext, data []byte) ([]byte, error) {
if len(nonce) != g.nonceSize {
panic("crypto/cipher: incorrect nonce length given to GCM")
}
// Sanity check to prevent the authentication from always succeeding if an implementation
// leaves tagSize uninitialized, for example.
if g.tagSize < gcmMinimumTagSize {
panic("crypto/cipher: incorrect GCM tag size")
}
if len(ciphertext) < g.tagSize {
return nil, errOpen
}
if uint64(len(ciphertext)) > ((1<<32)-2)*uint64(g.cipher.BlockSize())+uint64(g.tagSize) {
return nil, errOpen
}
tag := ciphertext[len(ciphertext)-g.tagSize:]
ciphertext = ciphertext[:len(ciphertext)-g.tagSize]
var counter, tagMask [gcmBlockSize]byte
g.deriveCounter(&counter, nonce)
g.cipher.Encrypt(tagMask[:], counter[:])
gcmInc32(&counter)
var expectedTag [gcmTagSize]byte
g.auth(expectedTag[:], ciphertext, data, &tagMask)
ret, out := smcipher.SliceForAppend(dst, len(ciphertext))
if smcipher.InexactOverlap(out, ciphertext) {
panic("crypto/cipher: invalid buffer overlap")
}
if subtle.ConstantTimeCompare(expectedTag[:g.tagSize], tag) != 1 {
// The AESNI code decrypts and authenticates concurrently, and
// so overwrites dst in the event of a tag mismatch. That
// behavior is mimicked here in order to be consistent across
// platforms.
for i := range out {
out[i] = 0
}
return nil, errOpen
}
g.counterCrypt(out, ciphertext, &counter)
return ret, nil
}
// reverseBits reverses the order of the bits of 4-bit number in i.
func reverseBits(i int) int {
i = ((i << 2) & 0xc) | ((i >> 2) & 0x3)
i = ((i << 1) & 0xa) | ((i >> 1) & 0x5)
return i
}
// gcmAdd adds two elements of GF(2¹²⁸) and returns the sum.
func gcmAdd(x, y *gcmFieldElement) gcmFieldElement {
// Addition in a characteristic 2 field is just XOR.
return gcmFieldElement{x.low ^ y.low, x.high ^ y.high}
}
// gcmDouble returns the result of doubling an element of GF(2¹²⁸).
func gcmDouble(x *gcmFieldElement) (double gcmFieldElement) {
msbSet := x.high&1 == 1
// Because of the bit-ordering, doubling is actually a right shift.
double.high = x.high >> 1
double.high |= x.low << 63
double.low = x.low >> 1
// If the most-significant bit was set before shifting then it,
// conceptually, becomes a term of x^128. This is greater than the
// irreducible polynomial so the result has to be reduced. The
// irreducible polynomial is 1+x+x^2+x^7+x^128. We can subtract that to
// eliminate the term at x^128 which also means subtracting the other
// four terms. In characteristic 2 fields, subtraction == addition ==
// XOR.
if msbSet {
double.low ^= 0xe100000000000000
}
return
}
var gcmReductionTable = []uint16{
0x0000, 0x1c20, 0x3840, 0x2460, 0x7080, 0x6ca0, 0x48c0, 0x54e0,
0xe100, 0xfd20, 0xd940, 0xc560, 0x9180, 0x8da0, 0xa9c0, 0xb5e0,
}
// mul sets y to y*H, where H is the GCM key, fixed during NewGCMWithNonceSize.
func (g *gcm) mul(y *gcmFieldElement) {
var z gcmFieldElement
for i := 0; i < 2; i++ {
word := y.high
if i == 1 {
word = y.low
}
// Multiplication works by multiplying z by 16 and adding in
// one of the precomputed multiples of H.
for j := 0; j < 64; j += 4 {
msw := z.high & 0xf
z.high >>= 4
z.high |= z.low << 60
z.low >>= 4
z.low ^= uint64(gcmReductionTable[msw]) << 48
// the values in |table| are ordered for
// little-endian bit positions. See the comment
// in NewGCMWithNonceSize.
t := &g.productTable[word&0xf]
z.low ^= t.low
z.high ^= t.high
word >>= 4
}
}
*y = z
}
// updateBlocks extends y with more polynomial terms from blocks, based on
// Horner's rule. There must be a multiple of gcmBlockSize bytes in blocks.
func (g *gcm) updateBlocks(y *gcmFieldElement, blocks []byte) {
for len(blocks) > 0 {
y.low ^= binary.BigEndian.Uint64(blocks)
y.high ^= binary.BigEndian.Uint64(blocks[8:])
g.mul(y)
blocks = blocks[gcmBlockSize:]
}
}
// update extends y with more polynomial terms from data. If data is not a
// multiple of gcmBlockSize bytes long then the remainder is zero padded.
func (g *gcm) update(y *gcmFieldElement, data []byte) {
fullBlocks := (len(data) >> 4) << 4
g.updateBlocks(y, data[:fullBlocks])
if len(data) != fullBlocks {
var partialBlock [gcmBlockSize]byte
copy(partialBlock[:], data[fullBlocks:])
g.updateBlocks(y, partialBlock[:])
}
}
// gcmInc32 treats the final four bytes of counterBlock as a big-endian value
// and increments it.
func gcmInc32(counterBlock *[16]byte) {
ctr := counterBlock[len(counterBlock)-4:]
binary.BigEndian.PutUint32(ctr, binary.BigEndian.Uint32(ctr)+1)
}
// counterCrypt crypts in to out using g.cipher in counter mode.
func (g *gcm) counterCrypt(out, in []byte, counter *[gcmBlockSize]byte) {
var mask [FourBlocksSize]byte
var couters [FourBlocksSize]byte
for len(in) >= FourBlocksSize {
copy(couters[:], counter[:])
gcmInc32(counter)
copy(couters[gcmBlockSize:], counter[:])
gcmInc32(counter)
copy(couters[2*gcmBlockSize:], counter[:])
gcmInc32(counter)
copy(couters[3*gcmBlockSize:], counter[:])
encryptBlocksAsm(&g.cipher.enc[0], &mask[0], &couters[0])
gcmInc32(counter)
smcipher.XorWords(out, in, mask[:])
out = out[FourBlocksSize:]
in = in[FourBlocksSize:]
}
if len(in) > 0 {
blocks := (len(in) + gcmBlockSize - 1) / gcmBlockSize
for i := 0; i < blocks; i++ {
copy(couters[i*gcmBlockSize:], counter[:])
gcmInc32(counter)
}
encryptBlocksAsm(&g.cipher.enc[0], &mask[0], &couters[0])
smcipher.XorBytes(out, in, mask[:blocks*gcmBlockSize])
}
}
// deriveCounter computes the initial GCM counter state from the given nonce.
// See NIST SP 800-38D, section 7.1. This assumes that counter is filled with
// zeros on entry.
func (g *gcm) deriveCounter(counter *[gcmBlockSize]byte, nonce []byte) {
// GCM has two modes of operation with respect to the initial counter
// state: a "fast path" for 96-bit (12-byte) nonces, and a "slow path"
// for nonces of other lengths. For a 96-bit nonce, the nonce, along
// with a four-byte big-endian counter starting at one, is used
// directly as the starting counter. For other nonce sizes, the counter
// is computed by passing it through the GHASH function.
if len(nonce) == gcmStandardNonceSize {
copy(counter[:], nonce)
counter[gcmBlockSize-1] = 1
} else {
var y gcmFieldElement
g.update(&y, nonce)
y.high ^= uint64(len(nonce)) * 8
g.mul(&y)
binary.BigEndian.PutUint64(counter[:8], y.low)
binary.BigEndian.PutUint64(counter[8:], y.high)
}
}
// auth calculates GHASH(ciphertext, additionalData), masks the result with
// tagMask and writes the result to out.
func (g *gcm) auth(out, ciphertext, additionalData []byte, tagMask *[gcmTagSize]byte) {
var y gcmFieldElement
g.update(&y, additionalData)
g.update(&y, ciphertext)
y.low ^= uint64(len(additionalData)) * 8
y.high ^= uint64(len(ciphertext)) * 8
g.mul(&y)
binary.BigEndian.PutUint64(out, y.low)
binary.BigEndian.PutUint64(out[8:], y.high)
smcipher.XorWords(out, out, tagMask[:])
}