Also, make sure New inlines so it does not cause an allocation. With this, we have a zero allocation poly1305 flow and a zero allocation generic chacha20poly1305 composition! \o/ While at it, remove some redundant code, and prepare to drop some complexity once the last assembly implementation of sum is dropped. Benchstat with "-tags purego" on amd64 name old time/op new time/op delta Open-64-8 461ns ± 2% 415ns ± 1% -9.93% (p=0.000 n=10+8) Seal-64-8 450ns ± 3% 412ns ± 3% -8.41% (p=0.000 n=10+10) Open-64-X-8 603ns ± 2% 544ns ± 2% -9.84% (p=0.000 n=10+10) Seal-64-X-8 580ns ± 3% 553ns ± 1% -4.56% (p=0.000 n=9+8) Open-1350-8 3.98µs ± 2% 3.65µs ± 2% -8.28% (p=0.000 n=9+10) Seal-1350-8 3.95µs ± 2% 3.64µs ± 1% -7.93% (p=0.000 n=9+10) Open-1350-X-8 4.06µs ± 1% 3.68µs ± 3% -9.31% (p=0.000 n=9+10) Seal-1350-X-8 4.08µs ± 4% 3.64µs ± 1% -10.71% (p=0.000 n=9+10) Open-8192-8 21.7µs ± 3% 18.9µs ± 2% -13.10% (p=0.000 n=9+10) Seal-8192-8 21.5µs ± 3% 18.8µs ± 1% -12.51% (p=0.000 n=9+9) Open-8192-X-8 21.4µs ± 1% 19.1µs ± 2% -10.88% (p=0.000 n=10+10) Seal-8192-X-8 21.3µs ± 2% 19.0µs ± 3% -10.92% (p=0.000 n=10+10) name old speed new speed delta Open-64-8 139MB/s ± 2% 154MB/s ± 2% +11.05% (p=0.000 n=10+8) Seal-64-8 142MB/s ± 3% 155MB/s ± 3% +9.11% (p=0.000 n=10+10) Open-64-X-8 106MB/s ± 2% 118MB/s ± 2% +10.93% (p=0.000 n=10+10) Seal-64-X-8 110MB/s ± 3% 116MB/s ± 1% +4.75% (p=0.000 n=9+8) Open-1350-8 339MB/s ± 2% 370MB/s ± 2% +9.04% (p=0.000 n=9+10) Seal-1350-8 342MB/s ± 2% 371MB/s ± 1% +8.60% (p=0.000 n=9+10) Open-1350-X-8 333MB/s ± 1% 367MB/s ± 3% +10.30% (p=0.000 n=9+10) Seal-1350-X-8 331MB/s ± 4% 371MB/s ± 2% +11.96% (p=0.000 n=9+10) Open-8192-8 377MB/s ± 3% 434MB/s ± 2% +15.07% (p=0.000 n=9+10) Seal-8192-8 381MB/s ± 3% 436MB/s ± 1% +14.29% (p=0.000 n=9+9) Open-8192-X-8 383MB/s ± 1% 429MB/s ± 2% +12.21% (p=0.000 n=10+10) Seal-8192-X-8 385MB/s ± 2% 432MB/s ± 3% +12.26% (p=0.000 n=10+10) name old alloc/op new alloc/op delta Open-64-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Seal-64-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Open-64-X-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Seal-64-X-8 96.0B ± 0% 0.0B -100.00% (p=0.000 n=10+10) Open-1350-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-1350-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-1350-X-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-1350-X-8 1.41kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-8192-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-8192-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Open-8192-X-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) Seal-8192-X-8 9.47kB ± 0% 0.00kB -100.00% (p=0.000 n=10+10) name old allocs/op new allocs/op delta Open-64-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-64-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-64-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-64-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-1350-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-1350-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-1350-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-1350-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-8192-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-8192-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Open-8192-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Seal-8192-X-8 1.00 ± 0% 0.00 -100.00% (p=0.000 n=10+10) Change-Id: I2c30ddc960a889b49c8ee8ff8073ffc4e75f43af Reviewed-on: https://go-review.googlesource.com/c/crypto/+/206977 Run-TryBot: Filippo Valsorda <filippo@golang.org> TryBot-Result: Gobot Gobot <gobot@golang.org> Reviewed-by: Katie Hockman <katie@golang.org>
82 lines
2.1 KiB
Go
82 lines
2.1 KiB
Go
// Copyright 2016 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 chacha20poly1305
|
|
|
|
import (
|
|
"encoding/binary"
|
|
|
|
"golang.org/x/crypto/chacha20"
|
|
"golang.org/x/crypto/internal/subtle"
|
|
"golang.org/x/crypto/poly1305"
|
|
)
|
|
|
|
func writeWithPadding(p *poly1305.MAC, b []byte) {
|
|
p.Write(b)
|
|
if rem := len(b) % 16; rem != 0 {
|
|
var buf [16]byte
|
|
padLen := 16 - rem
|
|
p.Write(buf[:padLen])
|
|
}
|
|
}
|
|
|
|
func writeUint64(p *poly1305.MAC, n int) {
|
|
var buf [8]byte
|
|
binary.LittleEndian.PutUint64(buf[:], uint64(n))
|
|
p.Write(buf[:])
|
|
}
|
|
|
|
func (c *chacha20poly1305) sealGeneric(dst, nonce, plaintext, additionalData []byte) []byte {
|
|
ret, out := sliceForAppend(dst, len(plaintext)+poly1305.TagSize)
|
|
ciphertext, tag := out[:len(plaintext)], out[len(plaintext):]
|
|
if subtle.InexactOverlap(out, plaintext) {
|
|
panic("chacha20poly1305: invalid buffer overlap")
|
|
}
|
|
|
|
var polyKey [32]byte
|
|
s, _ := chacha20.NewUnauthenticatedCipher(c.key[:], nonce)
|
|
s.XORKeyStream(polyKey[:], polyKey[:])
|
|
s.SetCounter(1) // set the counter to 1, skipping 32 bytes
|
|
s.XORKeyStream(ciphertext, plaintext)
|
|
|
|
p := poly1305.New(&polyKey)
|
|
writeWithPadding(p, additionalData)
|
|
writeWithPadding(p, ciphertext)
|
|
writeUint64(p, len(additionalData))
|
|
writeUint64(p, len(plaintext))
|
|
p.Sum(tag[:0])
|
|
|
|
return ret
|
|
}
|
|
|
|
func (c *chacha20poly1305) openGeneric(dst, nonce, ciphertext, additionalData []byte) ([]byte, error) {
|
|
tag := ciphertext[len(ciphertext)-16:]
|
|
ciphertext = ciphertext[:len(ciphertext)-16]
|
|
|
|
var polyKey [32]byte
|
|
s, _ := chacha20.NewUnauthenticatedCipher(c.key[:], nonce)
|
|
s.XORKeyStream(polyKey[:], polyKey[:])
|
|
s.SetCounter(1) // set the counter to 1, skipping 32 bytes
|
|
|
|
p := poly1305.New(&polyKey)
|
|
writeWithPadding(p, additionalData)
|
|
writeWithPadding(p, ciphertext)
|
|
writeUint64(p, len(additionalData))
|
|
writeUint64(p, len(ciphertext))
|
|
|
|
ret, out := sliceForAppend(dst, len(ciphertext))
|
|
if subtle.InexactOverlap(out, ciphertext) {
|
|
panic("chacha20poly1305: invalid buffer overlap")
|
|
}
|
|
if !p.Verify(tag) {
|
|
for i := range out {
|
|
out[i] = 0
|
|
}
|
|
return nil, errOpen
|
|
}
|
|
|
|
s.XORKeyStream(out, ciphertext)
|
|
return ret, nil
|
|
}
|