go.mod: golang.org/x/crypto 5770296d904e90f15f38f77dfc2e43fdf5efc083

full diff: 0c34fe9e7d...5770296d90

includes a fix in golang.org/x/crypto/ssh for CVE-2021-43565

- golang/go#49932
- 5770296d90

Signed-off-by: Sebastiaan van Stijn <github@gone.nl>

vendor.conf

@@ -147,7 +147,7 @@ github.com/golang/protobuf                          84668698ea25b64748563aa20726
 github.com/cloudflare/cfssl                         5d63dbd981b5c408effbb58c442d54761ff94fbd # 1.3.2
 github.com/fernet/fernet-go                         9eac43b88a5efb8651d24de9b68e87567e029736
 github.com/google/certificate-transparency-go       37a384cd035e722ea46e55029093e26687138edf # v1.0.20
-golang.org/x/crypto                                 0c34fe9e7dc2486962ef9867e3edb3503537209f
+golang.org/x/crypto                                 5770296d904e90f15f38f77dfc2e43fdf5efc083
 golang.org/x/time                                   3af7569d3a1e776fc2a3c1cec133b43105ea9c2e
 github.com/hashicorp/go-memdb                       cb9a474f84cc5e41b273b20c6927680b2a8776ad
 github.com/hashicorp/go-immutable-radix             826af9ccf0feeee615d546d69b11f8e98da8c8f1 git://github.com/tonistiigi/go-immutable-radix.git

vendor/golang.org/x/crypto/chacha20/chacha_arm64.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build go1.11 && gc && !purego
 // +build go1.11,gc,!purego
 
 #include "textflag.h"

vendor/golang.org/x/crypto/chacha20/chacha_ppc64le.s

@@ -19,6 +19,7 @@
 // The differences in this and the original implementation are
 // due to the calling conventions and initialization of constants.
 
+//go:build gc && !purego
 // +build gc,!purego
 
 #include "textflag.h"

vendor/golang.org/x/crypto/chacha20/chacha_s390x.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build gc && !purego
 // +build gc,!purego
 
 #include "go_asm.h"

vendor/golang.org/x/crypto/cryptobyte/asn1.go

@@ -117,6 +117,19 @@ func (b *Builder) AddASN1GeneralizedTime(t time.Time) {
 	})
 }
 
+// AddASN1UTCTime appends a DER-encoded ASN.1 UTCTime.
+func (b *Builder) AddASN1UTCTime(t time.Time) {
+	b.AddASN1(asn1.UTCTime, func(c *Builder) {
+		// As utilized by the X.509 profile, UTCTime can only
+		// represent the years 1950 through 2049.
+		if t.Year() < 1950 || t.Year() >= 2050 {
+			b.err = fmt.Errorf("cryptobyte: cannot represent %v as a UTCTime", t)
+			return
+		}
+		c.AddBytes([]byte(t.Format(defaultUTCTimeFormatStr)))
+	})
+}
+
 // AddASN1BitString appends a DER-encoded ASN.1 BIT STRING. This does not
 // support BIT STRINGs that are not a whole number of bytes.
 func (b *Builder) AddASN1BitString(data []byte) {
@@ -466,6 +479,45 @@ func (s *String) ReadASN1GeneralizedTime(out *time.Time) bool {
 	return true
 }
 
+const defaultUTCTimeFormatStr = "060102150405Z0700"
+
+// ReadASN1UTCTime decodes an ASN.1 UTCTime into out and advances.
+// It reports whether the read was successful.
+func (s *String) ReadASN1UTCTime(out *time.Time) bool {
+	var bytes String
+	if !s.ReadASN1(&bytes, asn1.UTCTime) {
+		return false
+	}
+	t := string(bytes)
+
+	formatStr := defaultUTCTimeFormatStr
+	var err error
+	res, err := time.Parse(formatStr, t)
+	if err != nil {
+		// Fallback to minute precision if we can't parse second
+		// precision. If we are following X.509 or X.690 we shouldn't
+		// support this, but we do.
+		formatStr = "0601021504Z0700"
+		res, err = time.Parse(formatStr, t)
+	}
+	if err != nil {
+		return false
+	}
+
+	if serialized := res.Format(formatStr); serialized != t {
+		return false
+	}
+
+	if res.Year() >= 2050 {
+		// UTCTime interprets the low order digits 50-99 as 1950-99.
+		// This only applies to its use in the X.509 profile.
+		// See https://tools.ietf.org/html/rfc5280#section-4.1.2.5.1
+		res = res.AddDate(-100, 0, 0)
+	}
+	*out = res
+	return true
+}
+
 // ReadASN1BitString decodes an ASN.1 BIT STRING into out and advances.
 // It reports whether the read was successful.
 func (s *String) ReadASN1BitString(out *encoding_asn1.BitString) bool {

vendor/golang.org/x/crypto/curve25519/curve25519.go

@@ -10,6 +10,8 @@ package curve25519 // import "golang.org/x/crypto/curve25519"
 import (
 	"crypto/subtle"
 	"fmt"
+
+	"golang.org/x/crypto/curve25519/internal/field"
 )
 
 // ScalarMult sets dst to the product scalar * point.
@@ -18,7 +20,55 @@ import (
 // zeroes, irrespective of the scalar. Instead, use the X25519 function, which
 // will return an error.
 func ScalarMult(dst, scalar, point *[32]byte) {
-	scalarMult(dst, scalar, point)
+	var e [32]byte
+
+	copy(e[:], scalar[:])
+	e[0] &= 248
+	e[31] &= 127
+	e[31] |= 64
+
+	var x1, x2, z2, x3, z3, tmp0, tmp1 field.Element
+	x1.SetBytes(point[:])
+	x2.One()
+	x3.Set(&x1)
+	z3.One()
+
+	swap := 0
+	for pos := 254; pos >= 0; pos-- {
+		b := e[pos/8] >> uint(pos&7)
+		b &= 1
+		swap ^= int(b)
+		x2.Swap(&x3, swap)
+		z2.Swap(&z3, swap)
+		swap = int(b)
+
+		tmp0.Subtract(&x3, &z3)
+		tmp1.Subtract(&x2, &z2)
+		x2.Add(&x2, &z2)
+		z2.Add(&x3, &z3)
+		z3.Multiply(&tmp0, &x2)
+		z2.Multiply(&z2, &tmp1)
+		tmp0.Square(&tmp1)
+		tmp1.Square(&x2)
+		x3.Add(&z3, &z2)
+		z2.Subtract(&z3, &z2)
+		x2.Multiply(&tmp1, &tmp0)
+		tmp1.Subtract(&tmp1, &tmp0)
+		z2.Square(&z2)
+
+		z3.Mult32(&tmp1, 121666)
+		x3.Square(&x3)
+		tmp0.Add(&tmp0, &z3)
+		z3.Multiply(&x1, &z2)
+		z2.Multiply(&tmp1, &tmp0)
+	}
+
+	x2.Swap(&x3, swap)
+	z2.Swap(&z3, swap)
+
+	z2.Invert(&z2)
+	x2.Multiply(&x2, &z2)
+	copy(dst[:], x2.Bytes())
 }
 
 // ScalarBaseMult sets dst to the product scalar * base where base is the

vendor/golang.org/x/crypto/curve25519/curve25519_amd64.go

@@ -1,241 +0,0 @@
-// Copyright 2012 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.
-
-//go:build amd64 && gc && !purego
-// +build amd64,gc,!purego
-
-package curve25519
-
-// These functions are implemented in the .s files. The names of the functions
-// in the rest of the file are also taken from the SUPERCOP sources to help
-// people following along.
-
-//go:noescape
-
-func cswap(inout *[5]uint64, v uint64)
-
-//go:noescape
-
-func ladderstep(inout *[5][5]uint64)
-
-//go:noescape
-
-func freeze(inout *[5]uint64)
-
-//go:noescape
-
-func mul(dest, a, b *[5]uint64)
-
-//go:noescape
-
-func square(out, in *[5]uint64)
-
-// mladder uses a Montgomery ladder to calculate (xr/zr) *= s.
-func mladder(xr, zr *[5]uint64, s *[32]byte) {
-	var work [5][5]uint64
-
-	work[0] = *xr
-	setint(&work[1], 1)
-	setint(&work[2], 0)
-	work[3] = *xr
-	setint(&work[4], 1)
-
-	j := uint(6)
-	var prevbit byte
-
-	for i := 31; i >= 0; i-- {
-		for j < 8 {
-			bit := ((*s)[i] >> j) & 1
-			swap := bit ^ prevbit
-			prevbit = bit
-			cswap(&work[1], uint64(swap))
-			ladderstep(&work)
-			j--
-		}
-		j = 7
-	}
-
-	*xr = work[1]
-	*zr = work[2]
-}
-
-func scalarMult(out, in, base *[32]byte) {
-	var e [32]byte
-	copy(e[:], (*in)[:])
-	e[0] &= 248
-	e[31] &= 127
-	e[31] |= 64
-
-	var t, z [5]uint64
-	unpack(&t, base)
-	mladder(&t, &z, &e)
-	invert(&z, &z)
-	mul(&t, &t, &z)
-	pack(out, &t)
-}
-
-func setint(r *[5]uint64, v uint64) {
-	r[0] = v
-	r[1] = 0
-	r[2] = 0
-	r[3] = 0
-	r[4] = 0
-}
-
-// unpack sets r = x where r consists of 5, 51-bit limbs in little-endian
-// order.
-func unpack(r *[5]uint64, x *[32]byte) {
-	r[0] = uint64(x[0]) |
-		uint64(x[1])<<8 |
-		uint64(x[2])<<16 |
-		uint64(x[3])<<24 |
-		uint64(x[4])<<32 |
-		uint64(x[5])<<40 |
-		uint64(x[6]&7)<<48
-
-	r[1] = uint64(x[6])>>3 |
-		uint64(x[7])<<5 |
-		uint64(x[8])<<13 |
-		uint64(x[9])<<21 |
-		uint64(x[10])<<29 |
-		uint64(x[11])<<37 |
-		uint64(x[12]&63)<<45
-
-	r[2] = uint64(x[12])>>6 |
-		uint64(x[13])<<2 |
-		uint64(x[14])<<10 |
-		uint64(x[15])<<18 |
-		uint64(x[16])<<26 |
-		uint64(x[17])<<34 |
-		uint64(x[18])<<42 |
-		uint64(x[19]&1)<<50
-
-	r[3] = uint64(x[19])>>1 |
-		uint64(x[20])<<7 |
-		uint64(x[21])<<15 |
-		uint64(x[22])<<23 |
-		uint64(x[23])<<31 |
-		uint64(x[24])<<39 |
-		uint64(x[25]&15)<<47
-
-	r[4] = uint64(x[25])>>4 |
-		uint64(x[26])<<4 |
-		uint64(x[27])<<12 |
-		uint64(x[28])<<20 |
-		uint64(x[29])<<28 |
-		uint64(x[30])<<36 |
-		uint64(x[31]&127)<<44
-}
-
-// pack sets out = x where out is the usual, little-endian form of the 5,
-// 51-bit limbs in x.
-func pack(out *[32]byte, x *[5]uint64) {
-	t := *x
-	freeze(&t)
-
-	out[0] = byte(t[0])
-	out[1] = byte(t[0] >> 8)
-	out[2] = byte(t[0] >> 16)
-	out[3] = byte(t[0] >> 24)
-	out[4] = byte(t[0] >> 32)
-	out[5] = byte(t[0] >> 40)
-	out[6] = byte(t[0] >> 48)
-
-	out[6] ^= byte(t[1]<<3) & 0xf8
-	out[7] = byte(t[1] >> 5)
-	out[8] = byte(t[1] >> 13)
-	out[9] = byte(t[1] >> 21)
-	out[10] = byte(t[1] >> 29)
-	out[11] = byte(t[1] >> 37)
-	out[12] = byte(t[1] >> 45)
-
-	out[12] ^= byte(t[2]<<6) & 0xc0
-	out[13] = byte(t[2] >> 2)
-	out[14] = byte(t[2] >> 10)
-	out[15] = byte(t[2] >> 18)
-	out[16] = byte(t[2] >> 26)
-	out[17] = byte(t[2] >> 34)
-	out[18] = byte(t[2] >> 42)
-	out[19] = byte(t[2] >> 50)
-
-	out[19] ^= byte(t[3]<<1) & 0xfe
-	out[20] = byte(t[3] >> 7)
-	out[21] = byte(t[3] >> 15)
-	out[22] = byte(t[3] >> 23)
-	out[23] = byte(t[3] >> 31)
-	out[24] = byte(t[3] >> 39)
-	out[25] = byte(t[3] >> 47)
-
-	out[25] ^= byte(t[4]<<4) & 0xf0
-	out[26] = byte(t[4] >> 4)
-	out[27] = byte(t[4] >> 12)
-	out[28] = byte(t[4] >> 20)
-	out[29] = byte(t[4] >> 28)
-	out[30] = byte(t[4] >> 36)
-	out[31] = byte(t[4] >> 44)
-}
-
-// invert calculates r = x^-1 mod p using Fermat's little theorem.
-func invert(r *[5]uint64, x *[5]uint64) {
-	var z2, z9, z11, z2_5_0, z2_10_0, z2_20_0, z2_50_0, z2_100_0, t [5]uint64
-
-	square(&z2, x)        /* 2 */
-	square(&t, &z2)       /* 4 */
-	square(&t, &t)        /* 8 */
-	mul(&z9, &t, x)       /* 9 */
-	mul(&z11, &z9, &z2)   /* 11 */
-	square(&t, &z11)      /* 22 */
-	mul(&z2_5_0, &t, &z9) /* 2^5 - 2^0 = 31 */
-
-	square(&t, &z2_5_0)      /* 2^6 - 2^1 */
-	for i := 1; i < 5; i++ { /* 2^20 - 2^10 */
-		square(&t, &t)
-	}
-	mul(&z2_10_0, &t, &z2_5_0) /* 2^10 - 2^0 */
-
-	square(&t, &z2_10_0)      /* 2^11 - 2^1 */
-	for i := 1; i < 10; i++ { /* 2^20 - 2^10 */
-		square(&t, &t)
-	}
-	mul(&z2_20_0, &t, &z2_10_0) /* 2^20 - 2^0 */
-
-	square(&t, &z2_20_0)      /* 2^21 - 2^1 */
-	for i := 1; i < 20; i++ { /* 2^40 - 2^20 */
-		square(&t, &t)
-	}
-	mul(&t, &t, &z2_20_0) /* 2^40 - 2^0 */
-
-	square(&t, &t)            /* 2^41 - 2^1 */
-	for i := 1; i < 10; i++ { /* 2^50 - 2^10 */
-		square(&t, &t)
-	}
-	mul(&z2_50_0, &t, &z2_10_0) /* 2^50 - 2^0 */
-
-	square(&t, &z2_50_0)      /* 2^51 - 2^1 */
-	for i := 1; i < 50; i++ { /* 2^100 - 2^50 */
-		square(&t, &t)
-	}
-	mul(&z2_100_0, &t, &z2_50_0) /* 2^100 - 2^0 */
-
-	square(&t, &z2_100_0)      /* 2^101 - 2^1 */
-	for i := 1; i < 100; i++ { /* 2^200 - 2^100 */
-		square(&t, &t)
-	}
-	mul(&t, &t, &z2_100_0) /* 2^200 - 2^0 */
-
-	square(&t, &t)            /* 2^201 - 2^1 */
-	for i := 1; i < 50; i++ { /* 2^250 - 2^50 */
-		square(&t, &t)
-	}
-	mul(&t, &t, &z2_50_0) /* 2^250 - 2^0 */
-
-	square(&t, &t) /* 2^251 - 2^1 */
-	square(&t, &t) /* 2^252 - 2^2 */
-	square(&t, &t) /* 2^253 - 2^3 */
-
-	square(&t, &t) /* 2^254 - 2^4 */
-
-	square(&t, &t)   /* 2^255 - 2^5 */
-	mul(r, &t, &z11) /* 2^255 - 21 */
-}

vendor/golang.org/x/crypto/curve25519/curve25519_generic.go

@@ -1,828 +0,0 @@
-// 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 curve25519
-
-import "encoding/binary"
-
-// This code is a port of the public domain, "ref10" implementation of
-// curve25519 from SUPERCOP 20130419 by D. J. Bernstein.
-
-// fieldElement represents an element of the field GF(2^255 - 19). An element
-// t, entries t[0]...t[9], represents the integer t[0]+2^26 t[1]+2^51 t[2]+2^77
-// t[3]+2^102 t[4]+...+2^230 t[9]. Bounds on each t[i] vary depending on
-// context.
-type fieldElement [10]int32
-
-func feZero(fe *fieldElement) {
-	for i := range fe {
-		fe[i] = 0
-	}
-}
-
-func feOne(fe *fieldElement) {
-	feZero(fe)
-	fe[0] = 1
-}
-
-func feAdd(dst, a, b *fieldElement) {
-	for i := range dst {
-		dst[i] = a[i] + b[i]
-	}
-}
-
-func feSub(dst, a, b *fieldElement) {
-	for i := range dst {
-		dst[i] = a[i] - b[i]
-	}
-}
-
-func feCopy(dst, src *fieldElement) {
-	for i := range dst {
-		dst[i] = src[i]
-	}
-}
-
-// feCSwap replaces (f,g) with (g,f) if b == 1; replaces (f,g) with (f,g) if b == 0.
-//
-// Preconditions: b in {0,1}.
-func feCSwap(f, g *fieldElement, b int32) {
-	b = -b
-	for i := range f {
-		t := b & (f[i] ^ g[i])
-		f[i] ^= t
-		g[i] ^= t
-	}
-}
-
-// load3 reads a 24-bit, little-endian value from in.
-func load3(in []byte) int64 {
-	var r int64
-	r = int64(in[0])
-	r |= int64(in[1]) << 8
-	r |= int64(in[2]) << 16
-	return r
-}
-
-// load4 reads a 32-bit, little-endian value from in.
-func load4(in []byte) int64 {
-	return int64(binary.LittleEndian.Uint32(in))
-}
-
-func feFromBytes(dst *fieldElement, src *[32]byte) {
-	h0 := load4(src[:])
-	h1 := load3(src[4:]) << 6
-	h2 := load3(src[7:]) << 5
-	h3 := load3(src[10:]) << 3
-	h4 := load3(src[13:]) << 2
-	h5 := load4(src[16:])
-	h6 := load3(src[20:]) << 7
-	h7 := load3(src[23:]) << 5
-	h8 := load3(src[26:]) << 4
-	h9 := (load3(src[29:]) & 0x7fffff) << 2
-
-	var carry [10]int64
-	carry[9] = (h9 + 1<<24) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	carry[1] = (h1 + 1<<24) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[3] = (h3 + 1<<24) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[5] = (h5 + 1<<24) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	carry[7] = (h7 + 1<<24) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-
-	carry[0] = (h0 + 1<<25) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[2] = (h2 + 1<<25) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[4] = (h4 + 1<<25) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[6] = (h6 + 1<<25) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	carry[8] = (h8 + 1<<25) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-
-	dst[0] = int32(h0)
-	dst[1] = int32(h1)
-	dst[2] = int32(h2)
-	dst[3] = int32(h3)
-	dst[4] = int32(h4)
-	dst[5] = int32(h5)
-	dst[6] = int32(h6)
-	dst[7] = int32(h7)
-	dst[8] = int32(h8)
-	dst[9] = int32(h9)
-}
-
-// feToBytes marshals h to s.
-// Preconditions:
-//   |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-//
-// Write p=2^255-19; q=floor(h/p).
-// Basic claim: q = floor(2^(-255)(h + 19 2^(-25)h9 + 2^(-1))).
-//
-// Proof:
-//   Have |h|<=p so |q|<=1 so |19^2 2^(-255) q|<1/4.
-//   Also have |h-2^230 h9|<2^230 so |19 2^(-255)(h-2^230 h9)|<1/4.
-//
-//   Write y=2^(-1)-19^2 2^(-255)q-19 2^(-255)(h-2^230 h9).
-//   Then 0<y<1.
-//
-//   Write r=h-pq.
-//   Have 0<=r<=p-1=2^255-20.
-//   Thus 0<=r+19(2^-255)r<r+19(2^-255)2^255<=2^255-1.
-//
-//   Write x=r+19(2^-255)r+y.
-//   Then 0<x<2^255 so floor(2^(-255)x) = 0 so floor(q+2^(-255)x) = q.
-//
-//   Have q+2^(-255)x = 2^(-255)(h + 19 2^(-25) h9 + 2^(-1))
-//   so floor(2^(-255)(h + 19 2^(-25) h9 + 2^(-1))) = q.
-func feToBytes(s *[32]byte, h *fieldElement) {
-	var carry [10]int32
-
-	q := (19*h[9] + (1 << 24)) >> 25
-	q = (h[0] + q) >> 26
-	q = (h[1] + q) >> 25
-	q = (h[2] + q) >> 26
-	q = (h[3] + q) >> 25
-	q = (h[4] + q) >> 26
-	q = (h[5] + q) >> 25
-	q = (h[6] + q) >> 26
-	q = (h[7] + q) >> 25
-	q = (h[8] + q) >> 26
-	q = (h[9] + q) >> 25
-
-	// Goal: Output h-(2^255-19)q, which is between 0 and 2^255-20.
-	h[0] += 19 * q
-	// Goal: Output h-2^255 q, which is between 0 and 2^255-20.
-
-	carry[0] = h[0] >> 26
-	h[1] += carry[0]
-	h[0] -= carry[0] << 26
-	carry[1] = h[1] >> 25
-	h[2] += carry[1]
-	h[1] -= carry[1] << 25
-	carry[2] = h[2] >> 26
-	h[3] += carry[2]
-	h[2] -= carry[2] << 26
-	carry[3] = h[3] >> 25
-	h[4] += carry[3]
-	h[3] -= carry[3] << 25
-	carry[4] = h[4] >> 26
-	h[5] += carry[4]
-	h[4] -= carry[4] << 26
-	carry[5] = h[5] >> 25
-	h[6] += carry[5]
-	h[5] -= carry[5] << 25
-	carry[6] = h[6] >> 26
-	h[7] += carry[6]
-	h[6] -= carry[6] << 26
-	carry[7] = h[7] >> 25
-	h[8] += carry[7]
-	h[7] -= carry[7] << 25
-	carry[8] = h[8] >> 26
-	h[9] += carry[8]
-	h[8] -= carry[8] << 26
-	carry[9] = h[9] >> 25
-	h[9] -= carry[9] << 25
-	// h10 = carry9
-
-	// Goal: Output h[0]+...+2^255 h10-2^255 q, which is between 0 and 2^255-20.
-	// Have h[0]+...+2^230 h[9] between 0 and 2^255-1;
-	// evidently 2^255 h10-2^255 q = 0.
-	// Goal: Output h[0]+...+2^230 h[9].
-
-	s[0] = byte(h[0] >> 0)
-	s[1] = byte(h[0] >> 8)
-	s[2] = byte(h[0] >> 16)
-	s[3] = byte((h[0] >> 24) | (h[1] << 2))
-	s[4] = byte(h[1] >> 6)
-	s[5] = byte(h[1] >> 14)
-	s[6] = byte((h[1] >> 22) | (h[2] << 3))
-	s[7] = byte(h[2] >> 5)
-	s[8] = byte(h[2] >> 13)
-	s[9] = byte((h[2] >> 21) | (h[3] << 5))
-	s[10] = byte(h[3] >> 3)
-	s[11] = byte(h[3] >> 11)
-	s[12] = byte((h[3] >> 19) | (h[4] << 6))
-	s[13] = byte(h[4] >> 2)
-	s[14] = byte(h[4] >> 10)
-	s[15] = byte(h[4] >> 18)
-	s[16] = byte(h[5] >> 0)
-	s[17] = byte(h[5] >> 8)
-	s[18] = byte(h[5] >> 16)
-	s[19] = byte((h[5] >> 24) | (h[6] << 1))
-	s[20] = byte(h[6] >> 7)
-	s[21] = byte(h[6] >> 15)
-	s[22] = byte((h[6] >> 23) | (h[7] << 3))
-	s[23] = byte(h[7] >> 5)
-	s[24] = byte(h[7] >> 13)
-	s[25] = byte((h[7] >> 21) | (h[8] << 4))
-	s[26] = byte(h[8] >> 4)
-	s[27] = byte(h[8] >> 12)
-	s[28] = byte((h[8] >> 20) | (h[9] << 6))
-	s[29] = byte(h[9] >> 2)
-	s[30] = byte(h[9] >> 10)
-	s[31] = byte(h[9] >> 18)
-}
-
-// feMul calculates h = f * g
-// Can overlap h with f or g.
-//
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//    |g| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-//
-// Notes on implementation strategy:
-//
-// Using schoolbook multiplication.
-// Karatsuba would save a little in some cost models.
-//
-// Most multiplications by 2 and 19 are 32-bit precomputations;
-// cheaper than 64-bit postcomputations.
-//
-// There is one remaining multiplication by 19 in the carry chain;
-// one *19 precomputation can be merged into this,
-// but the resulting data flow is considerably less clean.
-//
-// There are 12 carries below.
-// 10 of them are 2-way parallelizable and vectorizable.
-// Can get away with 11 carries, but then data flow is much deeper.
-//
-// With tighter constraints on inputs can squeeze carries into int32.
-func feMul(h, f, g *fieldElement) {
-	f0 := f[0]
-	f1 := f[1]
-	f2 := f[2]
-	f3 := f[3]
-	f4 := f[4]
-	f5 := f[5]
-	f6 := f[6]
-	f7 := f[7]
-	f8 := f[8]
-	f9 := f[9]
-	g0 := g[0]
-	g1 := g[1]
-	g2 := g[2]
-	g3 := g[3]
-	g4 := g[4]
-	g5 := g[5]
-	g6 := g[6]
-	g7 := g[7]
-	g8 := g[8]
-	g9 := g[9]
-	g1_19 := 19 * g1 // 1.4*2^29
-	g2_19 := 19 * g2 // 1.4*2^30; still ok
-	g3_19 := 19 * g3
-	g4_19 := 19 * g4
-	g5_19 := 19 * g5
-	g6_19 := 19 * g6
-	g7_19 := 19 * g7
-	g8_19 := 19 * g8
-	g9_19 := 19 * g9
-	f1_2 := 2 * f1
-	f3_2 := 2 * f3
-	f5_2 := 2 * f5
-	f7_2 := 2 * f7
-	f9_2 := 2 * f9
-	f0g0 := int64(f0) * int64(g0)
-	f0g1 := int64(f0) * int64(g1)
-	f0g2 := int64(f0) * int64(g2)
-	f0g3 := int64(f0) * int64(g3)
-	f0g4 := int64(f0) * int64(g4)
-	f0g5 := int64(f0) * int64(g5)
-	f0g6 := int64(f0) * int64(g6)
-	f0g7 := int64(f0) * int64(g7)
-	f0g8 := int64(f0) * int64(g8)
-	f0g9 := int64(f0) * int64(g9)
-	f1g0 := int64(f1) * int64(g0)
-	f1g1_2 := int64(f1_2) * int64(g1)
-	f1g2 := int64(f1) * int64(g2)
-	f1g3_2 := int64(f1_2) * int64(g3)
-	f1g4 := int64(f1) * int64(g4)
-	f1g5_2 := int64(f1_2) * int64(g5)
-	f1g6 := int64(f1) * int64(g6)
-	f1g7_2 := int64(f1_2) * int64(g7)
-	f1g8 := int64(f1) * int64(g8)
-	f1g9_38 := int64(f1_2) * int64(g9_19)
-	f2g0 := int64(f2) * int64(g0)
-	f2g1 := int64(f2) * int64(g1)
-	f2g2 := int64(f2) * int64(g2)
-	f2g3 := int64(f2) * int64(g3)
-	f2g4 := int64(f2) * int64(g4)
-	f2g5 := int64(f2) * int64(g5)
-	f2g6 := int64(f2) * int64(g6)
-	f2g7 := int64(f2) * int64(g7)
-	f2g8_19 := int64(f2) * int64(g8_19)
-	f2g9_19 := int64(f2) * int64(g9_19)
-	f3g0 := int64(f3) * int64(g0)
-	f3g1_2 := int64(f3_2) * int64(g1)
-	f3g2 := int64(f3) * int64(g2)
-	f3g3_2 := int64(f3_2) * int64(g3)
-	f3g4 := int64(f3) * int64(g4)
-	f3g5_2 := int64(f3_2) * int64(g5)
-	f3g6 := int64(f3) * int64(g6)
-	f3g7_38 := int64(f3_2) * int64(g7_19)
-	f3g8_19 := int64(f3) * int64(g8_19)
-	f3g9_38 := int64(f3_2) * int64(g9_19)
-	f4g0 := int64(f4) * int64(g0)
-	f4g1 := int64(f4) * int64(g1)
-	f4g2 := int64(f4) * int64(g2)
-	f4g3 := int64(f4) * int64(g3)
-	f4g4 := int64(f4) * int64(g4)
-	f4g5 := int64(f4) * int64(g5)
-	f4g6_19 := int64(f4) * int64(g6_19)
-	f4g7_19 := int64(f4) * int64(g7_19)
-	f4g8_19 := int64(f4) * int64(g8_19)
-	f4g9_19 := int64(f4) * int64(g9_19)
-	f5g0 := int64(f5) * int64(g0)
-	f5g1_2 := int64(f5_2) * int64(g1)
-	f5g2 := int64(f5) * int64(g2)
-	f5g3_2 := int64(f5_2) * int64(g3)
-	f5g4 := int64(f5) * int64(g4)
-	f5g5_38 := int64(f5_2) * int64(g5_19)
-	f5g6_19 := int64(f5) * int64(g6_19)
-	f5g7_38 := int64(f5_2) * int64(g7_19)
-	f5g8_19 := int64(f5) * int64(g8_19)
-	f5g9_38 := int64(f5_2) * int64(g9_19)
-	f6g0 := int64(f6) * int64(g0)
-	f6g1 := int64(f6) * int64(g1)
-	f6g2 := int64(f6) * int64(g2)
-	f6g3 := int64(f6) * int64(g3)
-	f6g4_19 := int64(f6) * int64(g4_19)
-	f6g5_19 := int64(f6) * int64(g5_19)
-	f6g6_19 := int64(f6) * int64(g6_19)
-	f6g7_19 := int64(f6) * int64(g7_19)
-	f6g8_19 := int64(f6) * int64(g8_19)
-	f6g9_19 := int64(f6) * int64(g9_19)
-	f7g0 := int64(f7) * int64(g0)
-	f7g1_2 := int64(f7_2) * int64(g1)
-	f7g2 := int64(f7) * int64(g2)
-	f7g3_38 := int64(f7_2) * int64(g3_19)
-	f7g4_19 := int64(f7) * int64(g4_19)
-	f7g5_38 := int64(f7_2) * int64(g5_19)
-	f7g6_19 := int64(f7) * int64(g6_19)
-	f7g7_38 := int64(f7_2) * int64(g7_19)
-	f7g8_19 := int64(f7) * int64(g8_19)
-	f7g9_38 := int64(f7_2) * int64(g9_19)
-	f8g0 := int64(f8) * int64(g0)
-	f8g1 := int64(f8) * int64(g1)
-	f8g2_19 := int64(f8) * int64(g2_19)
-	f8g3_19 := int64(f8) * int64(g3_19)
-	f8g4_19 := int64(f8) * int64(g4_19)
-	f8g5_19 := int64(f8) * int64(g5_19)
-	f8g6_19 := int64(f8) * int64(g6_19)
-	f8g7_19 := int64(f8) * int64(g7_19)
-	f8g8_19 := int64(f8) * int64(g8_19)
-	f8g9_19 := int64(f8) * int64(g9_19)
-	f9g0 := int64(f9) * int64(g0)
-	f9g1_38 := int64(f9_2) * int64(g1_19)
-	f9g2_19 := int64(f9) * int64(g2_19)
-	f9g3_38 := int64(f9_2) * int64(g3_19)
-	f9g4_19 := int64(f9) * int64(g4_19)
-	f9g5_38 := int64(f9_2) * int64(g5_19)
-	f9g6_19 := int64(f9) * int64(g6_19)
-	f9g7_38 := int64(f9_2) * int64(g7_19)
-	f9g8_19 := int64(f9) * int64(g8_19)
-	f9g9_38 := int64(f9_2) * int64(g9_19)
-	h0 := f0g0 + f1g9_38 + f2g8_19 + f3g7_38 + f4g6_19 + f5g5_38 + f6g4_19 + f7g3_38 + f8g2_19 + f9g1_38
-	h1 := f0g1 + f1g0 + f2g9_19 + f3g8_19 + f4g7_19 + f5g6_19 + f6g5_19 + f7g4_19 + f8g3_19 + f9g2_19
-	h2 := f0g2 + f1g1_2 + f2g0 + f3g9_38 + f4g8_19 + f5g7_38 + f6g6_19 + f7g5_38 + f8g4_19 + f9g3_38
-	h3 := f0g3 + f1g2 + f2g1 + f3g0 + f4g9_19 + f5g8_19 + f6g7_19 + f7g6_19 + f8g5_19 + f9g4_19
-	h4 := f0g4 + f1g3_2 + f2g2 + f3g1_2 + f4g0 + f5g9_38 + f6g8_19 + f7g7_38 + f8g6_19 + f9g5_38
-	h5 := f0g5 + f1g4 + f2g3 + f3g2 + f4g1 + f5g0 + f6g9_19 + f7g8_19 + f8g7_19 + f9g6_19
-	h6 := f0g6 + f1g5_2 + f2g4 + f3g3_2 + f4g2 + f5g1_2 + f6g0 + f7g9_38 + f8g8_19 + f9g7_38
-	h7 := f0g7 + f1g6 + f2g5 + f3g4 + f4g3 + f5g2 + f6g1 + f7g0 + f8g9_19 + f9g8_19
-	h8 := f0g8 + f1g7_2 + f2g6 + f3g5_2 + f4g4 + f5g3_2 + f6g2 + f7g1_2 + f8g0 + f9g9_38
-	h9 := f0g9 + f1g8 + f2g7 + f3g6 + f4g5 + f5g4 + f6g3 + f7g2 + f8g1 + f9g0
-	var carry [10]int64
-
-	// |h0| <= (1.1*1.1*2^52*(1+19+19+19+19)+1.1*1.1*2^50*(38+38+38+38+38))
-	//   i.e. |h0| <= 1.2*2^59; narrower ranges for h2, h4, h6, h8
-	// |h1| <= (1.1*1.1*2^51*(1+1+19+19+19+19+19+19+19+19))
-	//   i.e. |h1| <= 1.5*2^58; narrower ranges for h3, h5, h7, h9
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	// |h0| <= 2^25
-	// |h4| <= 2^25
-	// |h1| <= 1.51*2^58
-	// |h5| <= 1.51*2^58
-
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	// |h1| <= 2^24; from now on fits into int32
-	// |h5| <= 2^24; from now on fits into int32
-	// |h2| <= 1.21*2^59
-	// |h6| <= 1.21*2^59
-
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	// |h2| <= 2^25; from now on fits into int32 unchanged
-	// |h6| <= 2^25; from now on fits into int32 unchanged
-	// |h3| <= 1.51*2^58
-	// |h7| <= 1.51*2^58
-
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-	// |h3| <= 2^24; from now on fits into int32 unchanged
-	// |h7| <= 2^24; from now on fits into int32 unchanged
-	// |h4| <= 1.52*2^33
-	// |h8| <= 1.52*2^33
-
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-	// |h4| <= 2^25; from now on fits into int32 unchanged
-	// |h8| <= 2^25; from now on fits into int32 unchanged
-	// |h5| <= 1.01*2^24
-	// |h9| <= 1.51*2^58
-
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	// |h9| <= 2^24; from now on fits into int32 unchanged
-	// |h0| <= 1.8*2^37
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	// |h0| <= 2^25; from now on fits into int32 unchanged
-	// |h1| <= 1.01*2^24
-
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
-}
-
-// feSquare calculates h = f*f. Can overlap h with f.
-//
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-func feSquare(h, f *fieldElement) {
-	f0 := f[0]
-	f1 := f[1]
-	f2 := f[2]
-	f3 := f[3]
-	f4 := f[4]
-	f5 := f[5]
-	f6 := f[6]
-	f7 := f[7]
-	f8 := f[8]
-	f9 := f[9]
-	f0_2 := 2 * f0
-	f1_2 := 2 * f1
-	f2_2 := 2 * f2
-	f3_2 := 2 * f3
-	f4_2 := 2 * f4
-	f5_2 := 2 * f5
-	f6_2 := 2 * f6
-	f7_2 := 2 * f7
-	f5_38 := 38 * f5 // 1.31*2^30
-	f6_19 := 19 * f6 // 1.31*2^30
-	f7_38 := 38 * f7 // 1.31*2^30
-	f8_19 := 19 * f8 // 1.31*2^30
-	f9_38 := 38 * f9 // 1.31*2^30
-	f0f0 := int64(f0) * int64(f0)
-	f0f1_2 := int64(f0_2) * int64(f1)
-	f0f2_2 := int64(f0_2) * int64(f2)
-	f0f3_2 := int64(f0_2) * int64(f3)
-	f0f4_2 := int64(f0_2) * int64(f4)
-	f0f5_2 := int64(f0_2) * int64(f5)
-	f0f6_2 := int64(f0_2) * int64(f6)
-	f0f7_2 := int64(f0_2) * int64(f7)
-	f0f8_2 := int64(f0_2) * int64(f8)
-	f0f9_2 := int64(f0_2) * int64(f9)
-	f1f1_2 := int64(f1_2) * int64(f1)
-	f1f2_2 := int64(f1_2) * int64(f2)
-	f1f3_4 := int64(f1_2) * int64(f3_2)
-	f1f4_2 := int64(f1_2) * int64(f4)
-	f1f5_4 := int64(f1_2) * int64(f5_2)
-	f1f6_2 := int64(f1_2) * int64(f6)
-	f1f7_4 := int64(f1_2) * int64(f7_2)
-	f1f8_2 := int64(f1_2) * int64(f8)
-	f1f9_76 := int64(f1_2) * int64(f9_38)
-	f2f2 := int64(f2) * int64(f2)
-	f2f3_2 := int64(f2_2) * int64(f3)
-	f2f4_2 := int64(f2_2) * int64(f4)
-	f2f5_2 := int64(f2_2) * int64(f5)
-	f2f6_2 := int64(f2_2) * int64(f6)
-	f2f7_2 := int64(f2_2) * int64(f7)
-	f2f8_38 := int64(f2_2) * int64(f8_19)
-	f2f9_38 := int64(f2) * int64(f9_38)
-	f3f3_2 := int64(f3_2) * int64(f3)
-	f3f4_2 := int64(f3_2) * int64(f4)
-	f3f5_4 := int64(f3_2) * int64(f5_2)
-	f3f6_2 := int64(f3_2) * int64(f6)
-	f3f7_76 := int64(f3_2) * int64(f7_38)
-	f3f8_38 := int64(f3_2) * int64(f8_19)
-	f3f9_76 := int64(f3_2) * int64(f9_38)
-	f4f4 := int64(f4) * int64(f4)
-	f4f5_2 := int64(f4_2) * int64(f5)
-	f4f6_38 := int64(f4_2) * int64(f6_19)
-	f4f7_38 := int64(f4) * int64(f7_38)
-	f4f8_38 := int64(f4_2) * int64(f8_19)
-	f4f9_38 := int64(f4) * int64(f9_38)
-	f5f5_38 := int64(f5) * int64(f5_38)
-	f5f6_38 := int64(f5_2) * int64(f6_19)
-	f5f7_76 := int64(f5_2) * int64(f7_38)
-	f5f8_38 := int64(f5_2) * int64(f8_19)
-	f5f9_76 := int64(f5_2) * int64(f9_38)
-	f6f6_19 := int64(f6) * int64(f6_19)
-	f6f7_38 := int64(f6) * int64(f7_38)
-	f6f8_38 := int64(f6_2) * int64(f8_19)
-	f6f9_38 := int64(f6) * int64(f9_38)
-	f7f7_38 := int64(f7) * int64(f7_38)
-	f7f8_38 := int64(f7_2) * int64(f8_19)
-	f7f9_76 := int64(f7_2) * int64(f9_38)
-	f8f8_19 := int64(f8) * int64(f8_19)
-	f8f9_38 := int64(f8) * int64(f9_38)
-	f9f9_38 := int64(f9) * int64(f9_38)
-	h0 := f0f0 + f1f9_76 + f2f8_38 + f3f7_76 + f4f6_38 + f5f5_38
-	h1 := f0f1_2 + f2f9_38 + f3f8_38 + f4f7_38 + f5f6_38
-	h2 := f0f2_2 + f1f1_2 + f3f9_76 + f4f8_38 + f5f7_76 + f6f6_19
-	h3 := f0f3_2 + f1f2_2 + f4f9_38 + f5f8_38 + f6f7_38
-	h4 := f0f4_2 + f1f3_4 + f2f2 + f5f9_76 + f6f8_38 + f7f7_38
-	h5 := f0f5_2 + f1f4_2 + f2f3_2 + f6f9_38 + f7f8_38
-	h6 := f0f6_2 + f1f5_4 + f2f4_2 + f3f3_2 + f7f9_76 + f8f8_19
-	h7 := f0f7_2 + f1f6_2 + f2f5_2 + f3f4_2 + f8f9_38
-	h8 := f0f8_2 + f1f7_4 + f2f6_2 + f3f5_4 + f4f4 + f9f9_38
-	h9 := f0f9_2 + f1f8_2 + f2f7_2 + f3f6_2 + f4f5_2
-	var carry [10]int64
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
-}
-
-// feMul121666 calculates h = f * 121666. Can overlap h with f.
-//
-// Preconditions:
-//    |f| bounded by 1.1*2^26,1.1*2^25,1.1*2^26,1.1*2^25,etc.
-//
-// Postconditions:
-//    |h| bounded by 1.1*2^25,1.1*2^24,1.1*2^25,1.1*2^24,etc.
-func feMul121666(h, f *fieldElement) {
-	h0 := int64(f[0]) * 121666
-	h1 := int64(f[1]) * 121666
-	h2 := int64(f[2]) * 121666
-	h3 := int64(f[3]) * 121666
-	h4 := int64(f[4]) * 121666
-	h5 := int64(f[5]) * 121666
-	h6 := int64(f[6]) * 121666
-	h7 := int64(f[7]) * 121666
-	h8 := int64(f[8]) * 121666
-	h9 := int64(f[9]) * 121666
-	var carry [10]int64
-
-	carry[9] = (h9 + (1 << 24)) >> 25
-	h0 += carry[9] * 19
-	h9 -= carry[9] << 25
-	carry[1] = (h1 + (1 << 24)) >> 25
-	h2 += carry[1]
-	h1 -= carry[1] << 25
-	carry[3] = (h3 + (1 << 24)) >> 25
-	h4 += carry[3]
-	h3 -= carry[3] << 25
-	carry[5] = (h5 + (1 << 24)) >> 25
-	h6 += carry[5]
-	h5 -= carry[5] << 25
-	carry[7] = (h7 + (1 << 24)) >> 25
-	h8 += carry[7]
-	h7 -= carry[7] << 25
-
-	carry[0] = (h0 + (1 << 25)) >> 26
-	h1 += carry[0]
-	h0 -= carry[0] << 26
-	carry[2] = (h2 + (1 << 25)) >> 26
-	h3 += carry[2]
-	h2 -= carry[2] << 26
-	carry[4] = (h4 + (1 << 25)) >> 26
-	h5 += carry[4]
-	h4 -= carry[4] << 26
-	carry[6] = (h6 + (1 << 25)) >> 26
-	h7 += carry[6]
-	h6 -= carry[6] << 26
-	carry[8] = (h8 + (1 << 25)) >> 26
-	h9 += carry[8]
-	h8 -= carry[8] << 26
-
-	h[0] = int32(h0)
-	h[1] = int32(h1)
-	h[2] = int32(h2)
-	h[3] = int32(h3)
-	h[4] = int32(h4)
-	h[5] = int32(h5)
-	h[6] = int32(h6)
-	h[7] = int32(h7)
-	h[8] = int32(h8)
-	h[9] = int32(h9)
-}
-
-// feInvert sets out = z^-1.
-func feInvert(out, z *fieldElement) {
-	var t0, t1, t2, t3 fieldElement
-	var i int
-
-	feSquare(&t0, z)
-	for i = 1; i < 1; i++ {
-		feSquare(&t0, &t0)
-	}
-	feSquare(&t1, &t0)
-	for i = 1; i < 2; i++ {
-		feSquare(&t1, &t1)
-	}
-	feMul(&t1, z, &t1)
-	feMul(&t0, &t0, &t1)
-	feSquare(&t2, &t0)
-	for i = 1; i < 1; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t1, &t2)
-	feSquare(&t2, &t1)
-	for i = 1; i < 5; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t2, &t1)
-	for i = 1; i < 10; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t2, &t2, &t1)
-	feSquare(&t3, &t2)
-	for i = 1; i < 20; i++ {
-		feSquare(&t3, &t3)
-	}
-	feMul(&t2, &t3, &t2)
-	feSquare(&t2, &t2)
-	for i = 1; i < 10; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t2, &t1)
-	for i = 1; i < 50; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t2, &t2, &t1)
-	feSquare(&t3, &t2)
-	for i = 1; i < 100; i++ {
-		feSquare(&t3, &t3)
-	}
-	feMul(&t2, &t3, &t2)
-	feSquare(&t2, &t2)
-	for i = 1; i < 50; i++ {
-		feSquare(&t2, &t2)
-	}
-	feMul(&t1, &t2, &t1)
-	feSquare(&t1, &t1)
-	for i = 1; i < 5; i++ {
-		feSquare(&t1, &t1)
-	}
-	feMul(out, &t1, &t0)
-}
-
-func scalarMultGeneric(out, in, base *[32]byte) {
-	var e [32]byte
-
-	copy(e[:], in[:])
-	e[0] &= 248
-	e[31] &= 127
-	e[31] |= 64
-
-	var x1, x2, z2, x3, z3, tmp0, tmp1 fieldElement
-	feFromBytes(&x1, base)
-	feOne(&x2)
-	feCopy(&x3, &x1)
-	feOne(&z3)
-
-	swap := int32(0)
-	for pos := 254; pos >= 0; pos-- {
-		b := e[pos/8] >> uint(pos&7)
-		b &= 1
-		swap ^= int32(b)
-		feCSwap(&x2, &x3, swap)
-		feCSwap(&z2, &z3, swap)
-		swap = int32(b)
-
-		feSub(&tmp0, &x3, &z3)
-		feSub(&tmp1, &x2, &z2)
-		feAdd(&x2, &x2, &z2)
-		feAdd(&z2, &x3, &z3)
-		feMul(&z3, &tmp0, &x2)
-		feMul(&z2, &z2, &tmp1)
-		feSquare(&tmp0, &tmp1)
-		feSquare(&tmp1, &x2)
-		feAdd(&x3, &z3, &z2)
-		feSub(&z2, &z3, &z2)
-		feMul(&x2, &tmp1, &tmp0)
-		feSub(&tmp1, &tmp1, &tmp0)
-		feSquare(&z2, &z2)
-		feMul121666(&z3, &tmp1)
-		feSquare(&x3, &x3)
-		feAdd(&tmp0, &tmp0, &z3)
-		feMul(&z3, &x1, &z2)
-		feMul(&z2, &tmp1, &tmp0)
-	}
-
-	feCSwap(&x2, &x3, swap)
-	feCSwap(&z2, &z3, swap)
-
-	feInvert(&z2, &z2)
-	feMul(&x2, &x2, &z2)
-	feToBytes(out, &x2)
-}

vendor/golang.org/x/crypto/curve25519/curve25519_noasm.go

@@ -1,12 +0,0 @@
-// Copyright 2019 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.
-
-//go:build !amd64 || !gc || purego
-// +build !amd64 !gc purego
-
-package curve25519
-
-func scalarMult(out, in, base *[32]byte) {
-	scalarMultGeneric(out, in, base)
-}

vendor/golang.org/x/crypto/curve25519/internal/field/README

@@ -0,0 +1,7 @@
+This package is kept in sync with crypto/ed25519/internal/edwards25519/field in
+the standard library.
+
+If there are any changes in the standard library that need to be synced to this
+package, run sync.sh. It will not overwrite any local changes made since the
+previous sync, so it's ok to land changes in this package first, and then sync
+to the standard library later.

vendor/golang.org/x/crypto/curve25519/internal/field/fe.go

@@ -0,0 +1,416 @@
+// Copyright (c) 2017 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 field implements fast arithmetic modulo 2^255-19.
+package field
+
+import (
+	"crypto/subtle"
+	"encoding/binary"
+	"math/bits"
+)
+
+// Element represents an element of the field GF(2^255-19). Note that this
+// is not a cryptographically secure group, and should only be used to interact
+// with edwards25519.Point coordinates.
+//
+// This type works similarly to math/big.Int, and all arguments and receivers
+// are allowed to alias.
+//
+// The zero value is a valid zero element.
+type Element struct {
+	// An element t represents the integer
+	//     t.l0 + t.l1*2^51 + t.l2*2^102 + t.l3*2^153 + t.l4*2^204
+	//
+	// Between operations, all limbs are expected to be lower than 2^52.
+	l0 uint64
+	l1 uint64
+	l2 uint64
+	l3 uint64
+	l4 uint64
+}
+
+const maskLow51Bits uint64 = (1 << 51) - 1
+
+var feZero = &Element{0, 0, 0, 0, 0}
+
+// Zero sets v = 0, and returns v.
+func (v *Element) Zero() *Element {
+	*v = *feZero
+	return v
+}
+
+var feOne = &Element{1, 0, 0, 0, 0}
+
+// One sets v = 1, and returns v.
+func (v *Element) One() *Element {
+	*v = *feOne
+	return v
+}
+
+// reduce reduces v modulo 2^255 - 19 and returns it.
+func (v *Element) reduce() *Element {
+	v.carryPropagate()
+
+	// After the light reduction we now have a field element representation
+	// v < 2^255 + 2^13 * 19, but need v < 2^255 - 19.
+
+	// If v >= 2^255 - 19, then v + 19 >= 2^255, which would overflow 2^255 - 1,
+	// generating a carry. That is, c will be 0 if v < 2^255 - 19, and 1 otherwise.
+	c := (v.l0 + 19) >> 51
+	c = (v.l1 + c) >> 51
+	c = (v.l2 + c) >> 51
+	c = (v.l3 + c) >> 51
+	c = (v.l4 + c) >> 51
+
+	// If v < 2^255 - 19 and c = 0, this will be a no-op. Otherwise, it's
+	// effectively applying the reduction identity to the carry.
+	v.l0 += 19 * c
+
+	v.l1 += v.l0 >> 51
+	v.l0 = v.l0 & maskLow51Bits
+	v.l2 += v.l1 >> 51
+	v.l1 = v.l1 & maskLow51Bits
+	v.l3 += v.l2 >> 51
+	v.l2 = v.l2 & maskLow51Bits
+	v.l4 += v.l3 >> 51
+	v.l3 = v.l3 & maskLow51Bits
+	// no additional carry
+	v.l4 = v.l4 & maskLow51Bits
+
+	return v
+}
+
+// Add sets v = a + b, and returns v.
+func (v *Element) Add(a, b *Element) *Element {
+	v.l0 = a.l0 + b.l0
+	v.l1 = a.l1 + b.l1
+	v.l2 = a.l2 + b.l2
+	v.l3 = a.l3 + b.l3
+	v.l4 = a.l4 + b.l4
+	// Using the generic implementation here is actually faster than the
+	// assembly. Probably because the body of this function is so simple that
+	// the compiler can figure out better optimizations by inlining the carry
+	// propagation. TODO
+	return v.carryPropagateGeneric()
+}
+
+// Subtract sets v = a - b, and returns v.
+func (v *Element) Subtract(a, b *Element) *Element {
+	// We first add 2 * p, to guarantee the subtraction won't underflow, and
+	// then subtract b (which can be up to 2^255 + 2^13 * 19).
+	v.l0 = (a.l0 + 0xFFFFFFFFFFFDA) - b.l0
+	v.l1 = (a.l1 + 0xFFFFFFFFFFFFE) - b.l1
+	v.l2 = (a.l2 + 0xFFFFFFFFFFFFE) - b.l2
+	v.l3 = (a.l3 + 0xFFFFFFFFFFFFE) - b.l3
+	v.l4 = (a.l4 + 0xFFFFFFFFFFFFE) - b.l4
+	return v.carryPropagate()
+}
+
+// Negate sets v = -a, and returns v.
+func (v *Element) Negate(a *Element) *Element {
+	return v.Subtract(feZero, a)
+}
+
+// Invert sets v = 1/z mod p, and returns v.
+//
+// If z == 0, Invert returns v = 0.
+func (v *Element) Invert(z *Element) *Element {
+	// Inversion is implemented as exponentiation with exponent p − 2. It uses the
+	// same sequence of 255 squarings and 11 multiplications as [Curve25519].
+	var z2, z9, z11, z2_5_0, z2_10_0, z2_20_0, z2_50_0, z2_100_0, t Element
+
+	z2.Square(z)             // 2
+	t.Square(&z2)            // 4
+	t.Square(&t)             // 8
+	z9.Multiply(&t, z)       // 9
+	z11.Multiply(&z9, &z2)   // 11
+	t.Square(&z11)           // 22
+	z2_5_0.Multiply(&t, &z9) // 31 = 2^5 - 2^0
+
+	t.Square(&z2_5_0) // 2^6 - 2^1
+	for i := 0; i < 4; i++ {
+		t.Square(&t) // 2^10 - 2^5
+	}
+	z2_10_0.Multiply(&t, &z2_5_0) // 2^10 - 2^0
+
+	t.Square(&z2_10_0) // 2^11 - 2^1
+	for i := 0; i < 9; i++ {
+		t.Square(&t) // 2^20 - 2^10
+	}
+	z2_20_0.Multiply(&t, &z2_10_0) // 2^20 - 2^0
+
+	t.Square(&z2_20_0) // 2^21 - 2^1
+	for i := 0; i < 19; i++ {
+		t.Square(&t) // 2^40 - 2^20
+	}
+	t.Multiply(&t, &z2_20_0) // 2^40 - 2^0
+
+	t.Square(&t) // 2^41 - 2^1
+	for i := 0; i < 9; i++ {
+		t.Square(&t) // 2^50 - 2^10
+	}
+	z2_50_0.Multiply(&t, &z2_10_0) // 2^50 - 2^0
+
+	t.Square(&z2_50_0) // 2^51 - 2^1
+	for i := 0; i < 49; i++ {
+		t.Square(&t) // 2^100 - 2^50
+	}
+	z2_100_0.Multiply(&t, &z2_50_0) // 2^100 - 2^0
+
+	t.Square(&z2_100_0) // 2^101 - 2^1
+	for i := 0; i < 99; i++ {
+		t.Square(&t) // 2^200 - 2^100
+	}
+	t.Multiply(&t, &z2_100_0) // 2^200 - 2^0
+
+	t.Square(&t) // 2^201 - 2^1
+	for i := 0; i < 49; i++ {
+		t.Square(&t) // 2^250 - 2^50
+	}
+	t.Multiply(&t, &z2_50_0) // 2^250 - 2^0
+
+	t.Square(&t) // 2^251 - 2^1
+	t.Square(&t) // 2^252 - 2^2
+	t.Square(&t) // 2^253 - 2^3
+	t.Square(&t) // 2^254 - 2^4
+	t.Square(&t) // 2^255 - 2^5
+
+	return v.Multiply(&t, &z11) // 2^255 - 21
+}
+
+// Set sets v = a, and returns v.
+func (v *Element) Set(a *Element) *Element {
+	*v = *a
+	return v
+}
+
+// SetBytes sets v to x, which must be a 32-byte little-endian encoding.
+//
+// Consistent with RFC 7748, the most significant bit (the high bit of the
+// last byte) is ignored, and non-canonical values (2^255-19 through 2^255-1)
+// are accepted. Note that this is laxer than specified by RFC 8032.
+func (v *Element) SetBytes(x []byte) *Element {
+	if len(x) != 32 {
+		panic("edwards25519: invalid field element input size")
+	}
+
+	// Bits 0:51 (bytes 0:8, bits 0:64, shift 0, mask 51).
+	v.l0 = binary.LittleEndian.Uint64(x[0:8])
+	v.l0 &= maskLow51Bits
+	// Bits 51:102 (bytes 6:14, bits 48:112, shift 3, mask 51).
+	v.l1 = binary.LittleEndian.Uint64(x[6:14]) >> 3
+	v.l1 &= maskLow51Bits
+	// Bits 102:153 (bytes 12:20, bits 96:160, shift 6, mask 51).
+	v.l2 = binary.LittleEndian.Uint64(x[12:20]) >> 6
+	v.l2 &= maskLow51Bits
+	// Bits 153:204 (bytes 19:27, bits 152:216, shift 1, mask 51).
+	v.l3 = binary.LittleEndian.Uint64(x[19:27]) >> 1
+	v.l3 &= maskLow51Bits
+	// Bits 204:251 (bytes 24:32, bits 192:256, shift 12, mask 51).
+	// Note: not bytes 25:33, shift 4, to avoid overread.
+	v.l4 = binary.LittleEndian.Uint64(x[24:32]) >> 12
+	v.l4 &= maskLow51Bits
+
+	return v
+}
+
+// Bytes returns the canonical 32-byte little-endian encoding of v.
+func (v *Element) Bytes() []byte {
+	// This function is outlined to make the allocations inline in the caller
+	// rather than happen on the heap.
+	var out [32]byte
+	return v.bytes(&out)
+}
+
+func (v *Element) bytes(out *[32]byte) []byte {
+	t := *v
+	t.reduce()
+
+	var buf [8]byte
+	for i, l := range [5]uint64{t.l0, t.l1, t.l2, t.l3, t.l4} {
+		bitsOffset := i * 51
+		binary.LittleEndian.PutUint64(buf[:], l<<uint(bitsOffset%8))
+		for i, bb := range buf {
+			off := bitsOffset/8 + i
+			if off >= len(out) {
+				break
+			}
+			out[off] |= bb
+		}
+	}
+
+	return out[:]
+}
+
+// Equal returns 1 if v and u are equal, and 0 otherwise.
+func (v *Element) Equal(u *Element) int {
+	sa, sv := u.Bytes(), v.Bytes()
+	return subtle.ConstantTimeCompare(sa, sv)
+}
+
+// mask64Bits returns 0xffffffff if cond is 1, and 0 otherwise.
+func mask64Bits(cond int) uint64 { return ^(uint64(cond) - 1) }
+
+// Select sets v to a if cond == 1, and to b if cond == 0.
+func (v *Element) Select(a, b *Element, cond int) *Element {
+	m := mask64Bits(cond)
+	v.l0 = (m & a.l0) | (^m & b.l0)
+	v.l1 = (m & a.l1) | (^m & b.l1)
+	v.l2 = (m & a.l2) | (^m & b.l2)
+	v.l3 = (m & a.l3) | (^m & b.l3)
+	v.l4 = (m & a.l4) | (^m & b.l4)
+	return v
+}
+
+// Swap swaps v and u if cond == 1 or leaves them unchanged if cond == 0, and returns v.
+func (v *Element) Swap(u *Element, cond int) {
+	m := mask64Bits(cond)
+	t := m & (v.l0 ^ u.l0)
+	v.l0 ^= t
+	u.l0 ^= t
+	t = m & (v.l1 ^ u.l1)
+	v.l1 ^= t
+	u.l1 ^= t
+	t = m & (v.l2 ^ u.l2)
+	v.l2 ^= t
+	u.l2 ^= t
+	t = m & (v.l3 ^ u.l3)
+	v.l3 ^= t
+	u.l3 ^= t
+	t = m & (v.l4 ^ u.l4)
+	v.l4 ^= t
+	u.l4 ^= t
+}
+
+// IsNegative returns 1 if v is negative, and 0 otherwise.
+func (v *Element) IsNegative() int {
+	return int(v.Bytes()[0] & 1)
+}
+
+// Absolute sets v to |u|, and returns v.
+func (v *Element) Absolute(u *Element) *Element {
+	return v.Select(new(Element).Negate(u), u, u.IsNegative())
+}
+
+// Multiply sets v = x * y, and returns v.
+func (v *Element) Multiply(x, y *Element) *Element {
+	feMul(v, x, y)
+	return v
+}
+
+// Square sets v = x * x, and returns v.
+func (v *Element) Square(x *Element) *Element {
+	feSquare(v, x)
+	return v
+}
+
+// Mult32 sets v = x * y, and returns v.
+func (v *Element) Mult32(x *Element, y uint32) *Element {
+	x0lo, x0hi := mul51(x.l0, y)
+	x1lo, x1hi := mul51(x.l1, y)
+	x2lo, x2hi := mul51(x.l2, y)
+	x3lo, x3hi := mul51(x.l3, y)
+	x4lo, x4hi := mul51(x.l4, y)
+	v.l0 = x0lo + 19*x4hi // carried over per the reduction identity
+	v.l1 = x1lo + x0hi
+	v.l2 = x2lo + x1hi
+	v.l3 = x3lo + x2hi
+	v.l4 = x4lo + x3hi
+	// The hi portions are going to be only 32 bits, plus any previous excess,
+	// so we can skip the carry propagation.
+	return v
+}
+
+// mul51 returns lo + hi * 2⁵¹ = a * b.
+func mul51(a uint64, b uint32) (lo uint64, hi uint64) {
+	mh, ml := bits.Mul64(a, uint64(b))
+	lo = ml & maskLow51Bits
+	hi = (mh << 13) | (ml >> 51)
+	return
+}
+
+// Pow22523 set v = x^((p-5)/8), and returns v. (p-5)/8 is 2^252-3.
+func (v *Element) Pow22523(x *Element) *Element {
+	var t0, t1, t2 Element
+
+	t0.Square(x)             // x^2
+	t1.Square(&t0)           // x^4
+	t1.Square(&t1)           // x^8
+	t1.Multiply(x, &t1)      // x^9
+	t0.Multiply(&t0, &t1)    // x^11
+	t0.Square(&t0)           // x^22
+	t0.Multiply(&t1, &t0)    // x^31
+	t1.Square(&t0)           // x^62
+	for i := 1; i < 5; i++ { // x^992
+		t1.Square(&t1)
+	}
+	t0.Multiply(&t1, &t0)     // x^1023 -> 1023 = 2^10 - 1
+	t1.Square(&t0)            // 2^11 - 2
+	for i := 1; i < 10; i++ { // 2^20 - 2^10
+		t1.Square(&t1)
+	}
+	t1.Multiply(&t1, &t0)     // 2^20 - 1
+	t2.Square(&t1)            // 2^21 - 2
+	for i := 1; i < 20; i++ { // 2^40 - 2^20
+		t2.Square(&t2)
+	}
+	t1.Multiply(&t2, &t1)     // 2^40 - 1
+	t1.Square(&t1)            // 2^41 - 2
+	for i := 1; i < 10; i++ { // 2^50 - 2^10
+		t1.Square(&t1)
+	}
+	t0.Multiply(&t1, &t0)     // 2^50 - 1
+	t1.Square(&t0)            // 2^51 - 2
+	for i := 1; i < 50; i++ { // 2^100 - 2^50
+		t1.Square(&t1)
+	}
+	t1.Multiply(&t1, &t0)      // 2^100 - 1
+	t2.Square(&t1)             // 2^101 - 2
+	for i := 1; i < 100; i++ { // 2^200 - 2^100
+		t2.Square(&t2)
+	}
+	t1.Multiply(&t2, &t1)     // 2^200 - 1
+	t1.Square(&t1)            // 2^201 - 2
+	for i := 1; i < 50; i++ { // 2^250 - 2^50
+		t1.Square(&t1)
+	}
+	t0.Multiply(&t1, &t0)     // 2^250 - 1
+	t0.Square(&t0)            // 2^251 - 2
+	t0.Square(&t0)            // 2^252 - 4
+	return v.Multiply(&t0, x) // 2^252 - 3 -> x^(2^252-3)
+}
+
+// sqrtM1 is 2^((p-1)/4), which squared is equal to -1 by Euler's Criterion.
+var sqrtM1 = &Element{1718705420411056, 234908883556509,
+	2233514472574048, 2117202627021982, 765476049583133}
+
+// SqrtRatio sets r to the non-negative square root of the ratio of u and v.
+//
+// If u/v is square, SqrtRatio returns r and 1. If u/v is not square, SqrtRatio
+// sets r according to Section 4.3 of draft-irtf-cfrg-ristretto255-decaf448-00,
+// and returns r and 0.
+func (r *Element) SqrtRatio(u, v *Element) (rr *Element, wasSquare int) {
+	var a, b Element
+
+	// r = (u * v3) * (u * v7)^((p-5)/8)
+	v2 := a.Square(v)
+	uv3 := b.Multiply(u, b.Multiply(v2, v))
+	uv7 := a.Multiply(uv3, a.Square(v2))
+	r.Multiply(uv3, r.Pow22523(uv7))
+
+	check := a.Multiply(v, a.Square(r)) // check = v * r^2
+
+	uNeg := b.Negate(u)
+	correctSignSqrt := check.Equal(u)
+	flippedSignSqrt := check.Equal(uNeg)
+	flippedSignSqrtI := check.Equal(uNeg.Multiply(uNeg, sqrtM1))
+
+	rPrime := b.Multiply(r, sqrtM1) // r_prime = SQRT_M1 * r
+	// r = CT_SELECT(r_prime IF flipped_sign_sqrt | flipped_sign_sqrt_i ELSE r)
+	r.Select(rPrime, r, flippedSignSqrt|flippedSignSqrtI)
+
+	r.Absolute(r) // Choose the nonnegative square root.
+	return r, correctSignSqrt | flippedSignSqrt
+}

vendor/golang.org/x/crypto/curve25519/internal/field/fe_amd64.go

@@ -0,0 +1,13 @@
+// Code generated by command: go run fe_amd64_asm.go -out ../fe_amd64.s -stubs ../fe_amd64.go -pkg field. DO NOT EDIT.
+
+// +build amd64,gc,!purego
+
+package field
+
+// feMul sets out = a * b. It works like feMulGeneric.
+//go:noescape
+func feMul(out *Element, a *Element, b *Element)
+
+// feSquare sets out = a * a. It works like feSquareGeneric.
+//go:noescape
+func feSquare(out *Element, a *Element)

vendor/golang.org/x/crypto/curve25519/internal/field/fe_amd64.s

@@ -0,0 +1,379 @@
+// Code generated by command: go run fe_amd64_asm.go -out ../fe_amd64.s -stubs ../fe_amd64.go -pkg field. DO NOT EDIT.
+
+//go:build amd64 && gc && !purego
+// +build amd64,gc,!purego
+
+#include "textflag.h"
+
+// func feMul(out *Element, a *Element, b *Element)
+TEXT ·feMul(SB), NOSPLIT, $0-24
+	MOVQ a+8(FP), CX
+	MOVQ b+16(FP), BX
+
+	// r0 = a0×b0
+	MOVQ (CX), AX
+	MULQ (BX)
+	MOVQ AX, DI
+	MOVQ DX, SI
+
+	// r0 += 19×a1×b4
+	MOVQ   8(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   32(BX)
+	ADDQ   AX, DI
+	ADCQ   DX, SI
+
+	// r0 += 19×a2×b3
+	MOVQ   16(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   24(BX)
+	ADDQ   AX, DI
+	ADCQ   DX, SI
+
+	// r0 += 19×a3×b2
+	MOVQ   24(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   16(BX)
+	ADDQ   AX, DI
+	ADCQ   DX, SI
+
+	// r0 += 19×a4×b1
+	MOVQ   32(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   8(BX)
+	ADDQ   AX, DI
+	ADCQ   DX, SI
+
+	// r1 = a0×b1
+	MOVQ (CX), AX
+	MULQ 8(BX)
+	MOVQ AX, R9
+	MOVQ DX, R8
+
+	// r1 += a1×b0
+	MOVQ 8(CX), AX
+	MULQ (BX)
+	ADDQ AX, R9
+	ADCQ DX, R8
+
+	// r1 += 19×a2×b4
+	MOVQ   16(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   32(BX)
+	ADDQ   AX, R9
+	ADCQ   DX, R8
+
+	// r1 += 19×a3×b3
+	MOVQ   24(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   24(BX)
+	ADDQ   AX, R9
+	ADCQ   DX, R8
+
+	// r1 += 19×a4×b2
+	MOVQ   32(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   16(BX)
+	ADDQ   AX, R9
+	ADCQ   DX, R8
+
+	// r2 = a0×b2
+	MOVQ (CX), AX
+	MULQ 16(BX)
+	MOVQ AX, R11
+	MOVQ DX, R10
+
+	// r2 += a1×b1
+	MOVQ 8(CX), AX
+	MULQ 8(BX)
+	ADDQ AX, R11
+	ADCQ DX, R10
+
+	// r2 += a2×b0
+	MOVQ 16(CX), AX
+	MULQ (BX)
+	ADDQ AX, R11
+	ADCQ DX, R10
+
+	// r2 += 19×a3×b4
+	MOVQ   24(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   32(BX)
+	ADDQ   AX, R11
+	ADCQ   DX, R10
+
+	// r2 += 19×a4×b3
+	MOVQ   32(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   24(BX)
+	ADDQ   AX, R11
+	ADCQ   DX, R10
+
+	// r3 = a0×b3
+	MOVQ (CX), AX
+	MULQ 24(BX)
+	MOVQ AX, R13
+	MOVQ DX, R12
+
+	// r3 += a1×b2
+	MOVQ 8(CX), AX
+	MULQ 16(BX)
+	ADDQ AX, R13
+	ADCQ DX, R12
+
+	// r3 += a2×b1
+	MOVQ 16(CX), AX
+	MULQ 8(BX)
+	ADDQ AX, R13
+	ADCQ DX, R12
+
+	// r3 += a3×b0
+	MOVQ 24(CX), AX
+	MULQ (BX)
+	ADDQ AX, R13
+	ADCQ DX, R12
+
+	// r3 += 19×a4×b4
+	MOVQ   32(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   32(BX)
+	ADDQ   AX, R13
+	ADCQ   DX, R12
+
+	// r4 = a0×b4
+	MOVQ (CX), AX
+	MULQ 32(BX)
+	MOVQ AX, R15
+	MOVQ DX, R14
+
+	// r4 += a1×b3
+	MOVQ 8(CX), AX
+	MULQ 24(BX)
+	ADDQ AX, R15
+	ADCQ DX, R14
+
+	// r4 += a2×b2
+	MOVQ 16(CX), AX
+	MULQ 16(BX)
+	ADDQ AX, R15
+	ADCQ DX, R14
+
+	// r4 += a3×b1
+	MOVQ 24(CX), AX
+	MULQ 8(BX)
+	ADDQ AX, R15
+	ADCQ DX, R14
+
+	// r4 += a4×b0
+	MOVQ 32(CX), AX
+	MULQ (BX)
+	ADDQ AX, R15
+	ADCQ DX, R14
+
+	// First reduction chain
+	MOVQ   $0x0007ffffffffffff, AX
+	SHLQ   $0x0d, DI, SI
+	SHLQ   $0x0d, R9, R8
+	SHLQ   $0x0d, R11, R10
+	SHLQ   $0x0d, R13, R12
+	SHLQ   $0x0d, R15, R14
+	ANDQ   AX, DI
+	IMUL3Q $0x13, R14, R14
+	ADDQ   R14, DI
+	ANDQ   AX, R9
+	ADDQ   SI, R9
+	ANDQ   AX, R11
+	ADDQ   R8, R11
+	ANDQ   AX, R13
+	ADDQ   R10, R13
+	ANDQ   AX, R15
+	ADDQ   R12, R15
+
+	// Second reduction chain (carryPropagate)
+	MOVQ   DI, SI
+	SHRQ   $0x33, SI
+	MOVQ   R9, R8
+	SHRQ   $0x33, R8
+	MOVQ   R11, R10
+	SHRQ   $0x33, R10
+	MOVQ   R13, R12
+	SHRQ   $0x33, R12
+	MOVQ   R15, R14
+	SHRQ   $0x33, R14
+	ANDQ   AX, DI
+	IMUL3Q $0x13, R14, R14
+	ADDQ   R14, DI
+	ANDQ   AX, R9
+	ADDQ   SI, R9
+	ANDQ   AX, R11
+	ADDQ   R8, R11
+	ANDQ   AX, R13
+	ADDQ   R10, R13
+	ANDQ   AX, R15
+	ADDQ   R12, R15
+
+	// Store output
+	MOVQ out+0(FP), AX
+	MOVQ DI, (AX)
+	MOVQ R9, 8(AX)
+	MOVQ R11, 16(AX)
+	MOVQ R13, 24(AX)
+	MOVQ R15, 32(AX)
+	RET
+
+// func feSquare(out *Element, a *Element)
+TEXT ·feSquare(SB), NOSPLIT, $0-16
+	MOVQ a+8(FP), CX
+
+	// r0 = l0×l0
+	MOVQ (CX), AX
+	MULQ (CX)
+	MOVQ AX, SI
+	MOVQ DX, BX
+
+	// r0 += 38×l1×l4
+	MOVQ   8(CX), AX
+	IMUL3Q $0x26, AX, AX
+	MULQ   32(CX)
+	ADDQ   AX, SI
+	ADCQ   DX, BX
+
+	// r0 += 38×l2×l3
+	MOVQ   16(CX), AX
+	IMUL3Q $0x26, AX, AX
+	MULQ   24(CX)
+	ADDQ   AX, SI
+	ADCQ   DX, BX
+
+	// r1 = 2×l0×l1
+	MOVQ (CX), AX
+	SHLQ $0x01, AX
+	MULQ 8(CX)
+	MOVQ AX, R8
+	MOVQ DX, DI
+
+	// r1 += 38×l2×l4
+	MOVQ   16(CX), AX
+	IMUL3Q $0x26, AX, AX
+	MULQ   32(CX)
+	ADDQ   AX, R8
+	ADCQ   DX, DI
+
+	// r1 += 19×l3×l3
+	MOVQ   24(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   24(CX)
+	ADDQ   AX, R8
+	ADCQ   DX, DI
+
+	// r2 = 2×l0×l2
+	MOVQ (CX), AX
+	SHLQ $0x01, AX
+	MULQ 16(CX)
+	MOVQ AX, R10
+	MOVQ DX, R9
+
+	// r2 += l1×l1
+	MOVQ 8(CX), AX
+	MULQ 8(CX)
+	ADDQ AX, R10
+	ADCQ DX, R9
+
+	// r2 += 38×l3×l4
+	MOVQ   24(CX), AX
+	IMUL3Q $0x26, AX, AX
+	MULQ   32(CX)
+	ADDQ   AX, R10
+	ADCQ   DX, R9
+
+	// r3 = 2×l0×l3
+	MOVQ (CX), AX
+	SHLQ $0x01, AX
+	MULQ 24(CX)
+	MOVQ AX, R12
+	MOVQ DX, R11
+
+	// r3 += 2×l1×l2
+	MOVQ   8(CX), AX
+	IMUL3Q $0x02, AX, AX
+	MULQ   16(CX)
+	ADDQ   AX, R12
+	ADCQ   DX, R11
+
+	// r3 += 19×l4×l4
+	MOVQ   32(CX), AX
+	IMUL3Q $0x13, AX, AX
+	MULQ   32(CX)
+	ADDQ   AX, R12
+	ADCQ   DX, R11
+
+	// r4 = 2×l0×l4
+	MOVQ (CX), AX
+	SHLQ $0x01, AX
+	MULQ 32(CX)
+	MOVQ AX, R14
+	MOVQ DX, R13
+
+	// r4 += 2×l1×l3
+	MOVQ   8(CX), AX
+	IMUL3Q $0x02, AX, AX
+	MULQ   24(CX)
+	ADDQ   AX, R14
+	ADCQ   DX, R13
+
+	// r4 += l2×l2
+	MOVQ 16(CX), AX
+	MULQ 16(CX)
+	ADDQ AX, R14
+	ADCQ DX, R13
+
+	// First reduction chain
+	MOVQ   $0x0007ffffffffffff, AX
+	SHLQ   $0x0d, SI, BX
+	SHLQ   $0x0d, R8, DI
+	SHLQ   $0x0d, R10, R9
+	SHLQ   $0x0d, R12, R11
+	SHLQ   $0x0d, R14, R13
+	ANDQ   AX, SI
+	IMUL3Q $0x13, R13, R13
+	ADDQ   R13, SI
+	ANDQ   AX, R8
+	ADDQ   BX, R8
+	ANDQ   AX, R10
+	ADDQ   DI, R10
+	ANDQ   AX, R12
+	ADDQ   R9, R12
+	ANDQ   AX, R14
+	ADDQ   R11, R14
+
+	// Second reduction chain (carryPropagate)
+	MOVQ   SI, BX
+	SHRQ   $0x33, BX
+	MOVQ   R8, DI
+	SHRQ   $0x33, DI
+	MOVQ   R10, R9
+	SHRQ   $0x33, R9
+	MOVQ   R12, R11
+	SHRQ   $0x33, R11
+	MOVQ   R14, R13
+	SHRQ   $0x33, R13
+	ANDQ   AX, SI
+	IMUL3Q $0x13, R13, R13
+	ADDQ   R13, SI
+	ANDQ   AX, R8
+	ADDQ   BX, R8
+	ANDQ   AX, R10
+	ADDQ   DI, R10
+	ANDQ   AX, R12
+	ADDQ   R9, R12
+	ANDQ   AX, R14
+	ADDQ   R11, R14
+
+	// Store output
+	MOVQ out+0(FP), AX
+	MOVQ SI, (AX)
+	MOVQ R8, 8(AX)
+	MOVQ R10, 16(AX)
+	MOVQ R12, 24(AX)
+	MOVQ R14, 32(AX)
+	RET

vendor/golang.org/x/crypto/curve25519/internal/field/fe_amd64_noasm.go

@@ -0,0 +1,12 @@
+// Copyright (c) 2019 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.
+
+//go:build !amd64 || !gc || purego
+// +build !amd64 !gc purego
+
+package field
+
+func feMul(v, x, y *Element) { feMulGeneric(v, x, y) }
+
+func feSquare(v, x *Element) { feSquareGeneric(v, x) }

vendor/golang.org/x/crypto/curve25519/internal/field/fe_arm64.go

@@ -0,0 +1,16 @@
+// Copyright (c) 2020 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.
+
+//go:build arm64 && gc && !purego
+// +build arm64,gc,!purego
+
+package field
+
+//go:noescape
+func carryPropagate(v *Element)
+
+func (v *Element) carryPropagate() *Element {
+	carryPropagate(v)
+	return v
+}

vendor/golang.org/x/crypto/curve25519/internal/field/fe_arm64.s

@@ -0,0 +1,43 @@
+// Copyright (c) 2020 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.
+
+//go:build arm64 && gc && !purego
+// +build arm64,gc,!purego
+
+#include "textflag.h"
+
+// carryPropagate works exactly like carryPropagateGeneric and uses the
+// same AND, ADD, and LSR+MADD instructions emitted by the compiler, but
+// avoids loading R0-R4 twice and uses LDP and STP.
+//
+// See https://golang.org/issues/43145 for the main compiler issue.
+//
+// func carryPropagate(v *Element)
+TEXT ·carryPropagate(SB),NOFRAME|NOSPLIT,$0-8
+	MOVD v+0(FP), R20
+
+	LDP 0(R20), (R0, R1)
+	LDP 16(R20), (R2, R3)
+	MOVD 32(R20), R4
+
+	AND $0x7ffffffffffff, R0, R10
+	AND $0x7ffffffffffff, R1, R11
+	AND $0x7ffffffffffff, R2, R12
+	AND $0x7ffffffffffff, R3, R13
+	AND $0x7ffffffffffff, R4, R14
+
+	ADD R0>>51, R11, R11
+	ADD R1>>51, R12, R12
+	ADD R2>>51, R13, R13
+	ADD R3>>51, R14, R14
+	// R4>>51 * 19 + R10 -> R10
+	LSR $51, R4, R21
+	MOVD $19, R22
+	MADD R22, R10, R21, R10
+
+	STP (R10, R11), 0(R20)
+	STP (R12, R13), 16(R20)
+	MOVD R14, 32(R20)
+
+	RET

vendor/golang.org/x/crypto/curve25519/internal/field/fe_arm64_noasm.go

@@ -0,0 +1,12 @@
+// Copyright (c) 2021 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.
+
+//go:build !arm64 || !gc || purego
+// +build !arm64 !gc purego
+
+package field
+
+func (v *Element) carryPropagate() *Element {
+	return v.carryPropagateGeneric()
+}

vendor/golang.org/x/crypto/curve25519/internal/field/fe_generic.go

@@ -0,0 +1,264 @@
+// Copyright (c) 2017 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 field
+
+import "math/bits"
+
+// uint128 holds a 128-bit number as two 64-bit limbs, for use with the
+// bits.Mul64 and bits.Add64 intrinsics.
+type uint128 struct {
+	lo, hi uint64
+}
+
+// mul64 returns a * b.
+func mul64(a, b uint64) uint128 {
+	hi, lo := bits.Mul64(a, b)
+	return uint128{lo, hi}
+}
+
+// addMul64 returns v + a * b.
+func addMul64(v uint128, a, b uint64) uint128 {
+	hi, lo := bits.Mul64(a, b)
+	lo, c := bits.Add64(lo, v.lo, 0)
+	hi, _ = bits.Add64(hi, v.hi, c)
+	return uint128{lo, hi}
+}
+
+// shiftRightBy51 returns a >> 51. a is assumed to be at most 115 bits.
+func shiftRightBy51(a uint128) uint64 {
+	return (a.hi << (64 - 51)) | (a.lo >> 51)
+}
+
+func feMulGeneric(v, a, b *Element) {
+	a0 := a.l0
+	a1 := a.l1
+	a2 := a.l2
+	a3 := a.l3
+	a4 := a.l4
+
+	b0 := b.l0
+	b1 := b.l1
+	b2 := b.l2
+	b3 := b.l3
+	b4 := b.l4
+
+	// Limb multiplication works like pen-and-paper columnar multiplication, but
+	// with 51-bit limbs instead of digits.
+	//
+	//                          a4   a3   a2   a1   a0  x
+	//                          b4   b3   b2   b1   b0  =
+	//                         ------------------------
+	//                        a4b0 a3b0 a2b0 a1b0 a0b0  +
+	//                   a4b1 a3b1 a2b1 a1b1 a0b1       +
+	//              a4b2 a3b2 a2b2 a1b2 a0b2            +
+	//         a4b3 a3b3 a2b3 a1b3 a0b3                 +
+	//    a4b4 a3b4 a2b4 a1b4 a0b4                      =
+	//   ----------------------------------------------
+	//      r8   r7   r6   r5   r4   r3   r2   r1   r0
+	//
+	// We can then use the reduction identity (a * 2²⁵⁵ + b = a * 19 + b) to
+	// reduce the limbs that would overflow 255 bits. r5 * 2²⁵⁵ becomes 19 * r5,
+	// r6 * 2³⁰⁶ becomes 19 * r6 * 2⁵¹, etc.
+	//
+	// Reduction can be carried out simultaneously to multiplication. For
+	// example, we do not compute r5: whenever the result of a multiplication
+	// belongs to r5, like a1b4, we multiply it by 19 and add the result to r0.
+	//
+	//            a4b0    a3b0    a2b0    a1b0    a0b0  +
+	//            a3b1    a2b1    a1b1    a0b1 19×a4b1  +
+	//            a2b2    a1b2    a0b2 19×a4b2 19×a3b2  +
+	//            a1b3    a0b3 19×a4b3 19×a3b3 19×a2b3  +
+	//            a0b4 19×a4b4 19×a3b4 19×a2b4 19×a1b4  =
+	//           --------------------------------------
+	//              r4      r3      r2      r1      r0
+	//
+	// Finally we add up the columns into wide, overlapping limbs.
+
+	a1_19 := a1 * 19
+	a2_19 := a2 * 19
+	a3_19 := a3 * 19
+	a4_19 := a4 * 19
+
+	// r0 = a0×b0 + 19×(a1×b4 + a2×b3 + a3×b2 + a4×b1)
+	r0 := mul64(a0, b0)
+	r0 = addMul64(r0, a1_19, b4)
+	r0 = addMul64(r0, a2_19, b3)
+	r0 = addMul64(r0, a3_19, b2)
+	r0 = addMul64(r0, a4_19, b1)
+
+	// r1 = a0×b1 + a1×b0 + 19×(a2×b4 + a3×b3 + a4×b2)
+	r1 := mul64(a0, b1)
+	r1 = addMul64(r1, a1, b0)
+	r1 = addMul64(r1, a2_19, b4)
+	r1 = addMul64(r1, a3_19, b3)
+	r1 = addMul64(r1, a4_19, b2)
+
+	// r2 = a0×b2 + a1×b1 + a2×b0 + 19×(a3×b4 + a4×b3)
+	r2 := mul64(a0, b2)
+	r2 = addMul64(r2, a1, b1)
+	r2 = addMul64(r2, a2, b0)
+	r2 = addMul64(r2, a3_19, b4)
+	r2 = addMul64(r2, a4_19, b3)
+
+	// r3 = a0×b3 + a1×b2 + a2×b1 + a3×b0 + 19×a4×b4
+	r3 := mul64(a0, b3)
+	r3 = addMul64(r3, a1, b2)
+	r3 = addMul64(r3, a2, b1)
+	r3 = addMul64(r3, a3, b0)
+	r3 = addMul64(r3, a4_19, b4)
+
+	// r4 = a0×b4 + a1×b3 + a2×b2 + a3×b1 + a4×b0
+	r4 := mul64(a0, b4)
+	r4 = addMul64(r4, a1, b3)
+	r4 = addMul64(r4, a2, b2)
+	r4 = addMul64(r4, a3, b1)
+	r4 = addMul64(r4, a4, b0)
+
+	// After the multiplication, we need to reduce (carry) the five coefficients
+	// to obtain a result with limbs that are at most slightly larger than 2⁵¹,
+	// to respect the Element invariant.
+	//
+	// Overall, the reduction works the same as carryPropagate, except with
+	// wider inputs: we take the carry for each coefficient by shifting it right
+	// by 51, and add it to the limb above it. The top carry is multiplied by 19
+	// according to the reduction identity and added to the lowest limb.
+	//
+	// The largest coefficient (r0) will be at most 111 bits, which guarantees
+	// that all carries are at most 111 - 51 = 60 bits, which fits in a uint64.
+	//
+	//     r0 = a0×b0 + 19×(a1×b4 + a2×b3 + a3×b2 + a4×b1)
+	//     r0 < 2⁵²×2⁵² + 19×(2⁵²×2⁵² + 2⁵²×2⁵² + 2⁵²×2⁵² + 2⁵²×2⁵²)
+	//     r0 < (1 + 19 × 4) × 2⁵² × 2⁵²
+	//     r0 < 2⁷ × 2⁵² × 2⁵²
+	//     r0 < 2¹¹¹
+	//
+	// Moreover, the top coefficient (r4) is at most 107 bits, so c4 is at most
+	// 56 bits, and c4 * 19 is at most 61 bits, which again fits in a uint64 and
+	// allows us to easily apply the reduction identity.
+	//
+	//     r4 = a0×b4 + a1×b3 + a2×b2 + a3×b1 + a4×b0
+	//     r4 < 5 × 2⁵² × 2⁵²
+	//     r4 < 2¹⁰⁷
+	//
+
+	c0 := shiftRightBy51(r0)
+	c1 := shiftRightBy51(r1)
+	c2 := shiftRightBy51(r2)
+	c3 := shiftRightBy51(r3)
+	c4 := shiftRightBy51(r4)
+
+	rr0 := r0.lo&maskLow51Bits + c4*19
+	rr1 := r1.lo&maskLow51Bits + c0
+	rr2 := r2.lo&maskLow51Bits + c1
+	rr3 := r3.lo&maskLow51Bits + c2
+	rr4 := r4.lo&maskLow51Bits + c3
+
+	// Now all coefficients fit into 64-bit registers but are still too large to
+	// be passed around as a Element. We therefore do one last carry chain,
+	// where the carries will be small enough to fit in the wiggle room above 2⁵¹.
+	*v = Element{rr0, rr1, rr2, rr3, rr4}
+	v.carryPropagate()
+}
+
+func feSquareGeneric(v, a *Element) {
+	l0 := a.l0
+	l1 := a.l1
+	l2 := a.l2
+	l3 := a.l3
+	l4 := a.l4
+
+	// Squaring works precisely like multiplication above, but thanks to its
+	// symmetry we get to group a few terms together.
+	//
+	//                          l4   l3   l2   l1   l0  x
+	//                          l4   l3   l2   l1   l0  =
+	//                         ------------------------
+	//                        l4l0 l3l0 l2l0 l1l0 l0l0  +
+	//                   l4l1 l3l1 l2l1 l1l1 l0l1       +
+	//              l4l2 l3l2 l2l2 l1l2 l0l2            +
+	//         l4l3 l3l3 l2l3 l1l3 l0l3                 +
+	//    l4l4 l3l4 l2l4 l1l4 l0l4                      =
+	//   ----------------------------------------------
+	//      r8   r7   r6   r5   r4   r3   r2   r1   r0
+	//
+	//            l4l0    l3l0    l2l0    l1l0    l0l0  +
+	//            l3l1    l2l1    l1l1    l0l1 19×l4l1  +
+	//            l2l2    l1l2    l0l2 19×l4l2 19×l3l2  +
+	//            l1l3    l0l3 19×l4l3 19×l3l3 19×l2l3  +
+	//            l0l4 19×l4l4 19×l3l4 19×l2l4 19×l1l4  =
+	//           --------------------------------------
+	//              r4      r3      r2      r1      r0
+	//
+	// With precomputed 2×, 19×, and 2×19× terms, we can compute each limb with
+	// only three Mul64 and four Add64, instead of five and eight.
+
+	l0_2 := l0 * 2
+	l1_2 := l1 * 2
+
+	l1_38 := l1 * 38
+	l2_38 := l2 * 38
+	l3_38 := l3 * 38
+
+	l3_19 := l3 * 19
+	l4_19 := l4 * 19
+
+	// r0 = l0×l0 + 19×(l1×l4 + l2×l3 + l3×l2 + l4×l1) = l0×l0 + 19×2×(l1×l4 + l2×l3)
+	r0 := mul64(l0, l0)
+	r0 = addMul64(r0, l1_38, l4)
+	r0 = addMul64(r0, l2_38, l3)
+
+	// r1 = l0×l1 + l1×l0 + 19×(l2×l4 + l3×l3 + l4×l2) = 2×l0×l1 + 19×2×l2×l4 + 19×l3×l3
+	r1 := mul64(l0_2, l1)
+	r1 = addMul64(r1, l2_38, l4)
+	r1 = addMul64(r1, l3_19, l3)
+
+	// r2 = l0×l2 + l1×l1 + l2×l0 + 19×(l3×l4 + l4×l3) = 2×l0×l2 + l1×l1 + 19×2×l3×l4
+	r2 := mul64(l0_2, l2)
+	r2 = addMul64(r2, l1, l1)
+	r2 = addMul64(r2, l3_38, l4)
+
+	// r3 = l0×l3 + l1×l2 + l2×l1 + l3×l0 + 19×l4×l4 = 2×l0×l3 + 2×l1×l2 + 19×l4×l4
+	r3 := mul64(l0_2, l3)
+	r3 = addMul64(r3, l1_2, l2)
+	r3 = addMul64(r3, l4_19, l4)
+
+	// r4 = l0×l4 + l1×l3 + l2×l2 + l3×l1 + l4×l0 = 2×l0×l4 + 2×l1×l3 + l2×l2
+	r4 := mul64(l0_2, l4)
+	r4 = addMul64(r4, l1_2, l3)
+	r4 = addMul64(r4, l2, l2)
+
+	c0 := shiftRightBy51(r0)
+	c1 := shiftRightBy51(r1)
+	c2 := shiftRightBy51(r2)
+	c3 := shiftRightBy51(r3)
+	c4 := shiftRightBy51(r4)
+
+	rr0 := r0.lo&maskLow51Bits + c4*19
+	rr1 := r1.lo&maskLow51Bits + c0
+	rr2 := r2.lo&maskLow51Bits + c1
+	rr3 := r3.lo&maskLow51Bits + c2
+	rr4 := r4.lo&maskLow51Bits + c3
+
+	*v = Element{rr0, rr1, rr2, rr3, rr4}
+	v.carryPropagate()
+}
+
+// carryPropagate brings the limbs below 52 bits by applying the reduction
+// identity (a * 2²⁵⁵ + b = a * 19 + b) to the l4 carry. TODO inline
+func (v *Element) carryPropagateGeneric() *Element {
+	c0 := v.l0 >> 51
+	c1 := v.l1 >> 51
+	c2 := v.l2 >> 51
+	c3 := v.l3 >> 51
+	c4 := v.l4 >> 51
+
+	v.l0 = v.l0&maskLow51Bits + c4*19
+	v.l1 = v.l1&maskLow51Bits + c0
+	v.l2 = v.l2&maskLow51Bits + c1
+	v.l3 = v.l3&maskLow51Bits + c2
+	v.l4 = v.l4&maskLow51Bits + c3
+
+	return v
+}

vendor/golang.org/x/crypto/go.mod

@@ -1,9 +1,11 @@
 module golang.org/x/crypto
 
-go 1.11
+go 1.17
 
 require (
-	golang.org/x/net v0.0.0-20210226172049-e18ecbb05110
+	golang.org/x/net v0.0.0-20211112202133-69e39bad7dc2
-	golang.org/x/sys v0.0.0-20201119102817-f84b799fce68
+	golang.org/x/sys v0.0.0-20210615035016-665e8c7367d1
 	golang.org/x/term v0.0.0-20201126162022-7de9c90e9dd1
 )
+
+require golang.org/x/text v0.3.6 // indirect

vendor/golang.org/x/crypto/internal/poly1305/poly1305.go

@@ -15,7 +15,7 @@
 // used with a fixed key in order to generate one-time keys from an nonce.
 // However, in this package AES isn't used and the one-time key is specified
 // directly.
-package poly1305 // import "golang.org/x/crypto/poly1305"
+package poly1305
 
 import "crypto/subtle"
 

vendor/golang.org/x/crypto/internal/poly1305/sum_amd64.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build gc && !purego
 // +build gc,!purego
 
 #include "textflag.h"

vendor/golang.org/x/crypto/internal/poly1305/sum_ppc64le.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build gc && !purego
 // +build gc,!purego
 
 #include "textflag.h"
@@ -82,7 +83,7 @@ multiply:
 	BGE loop
 
 bytes_between_0_and_15:
-	CMP  $0, R5
+	CMP  R5, $0
 	BEQ  done
 	MOVD $0, R16 // h0
 	MOVD $0, R17 // h1
@@ -122,7 +123,7 @@ just1:
 	// Exactly 8
 	MOVD (R4), R16
 
-	CMP $0, R17
+	CMP R17, $0
 
 	// Check if we've already set R17; if not
 	// set 1 to indicate end of msg.
@@ -151,7 +152,7 @@ less4:
 	ADD   $2, R4
 
 less2:
-	CMP   $0, R5
+	CMP   R5, $0
 	BEQ   insert1
 	MOVBZ (R4), R21
 	SLD   R22, R21, R21
@@ -166,12 +167,12 @@ insert1:
 
 carry:
 	// Add new values to h0, h1, h2
-	ADDC R16, R8
+	ADDC  R16, R8
-	ADDE R17, R9
+	ADDE  R17, R9
-	ADDE $0, R10
+	ADDZE R10, R10
-	MOVD $16, R5
+	MOVD  $16, R5
-	ADD  R5, R4
+	ADD   R5, R4
-	BR   multiply
+	BR    multiply
 
 done:
 	// Save h0, h1, h2 in state

vendor/golang.org/x/crypto/internal/poly1305/sum_s390x.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build gc && !purego
 // +build gc,!purego
 
 #include "textflag.h"
@@ -17,7 +18,7 @@
 // value. These limbs are, for the most part, zero extended and
 // placed into 64-bit vector register elements. Each vector
 // register is 128-bits wide and so holds 2 of these elements.
-// Using 26-bit limbs allows us plenty of headroom to accomodate
+// Using 26-bit limbs allows us plenty of headroom to accommodate
 // accumulations before and after multiplication without
 // overflowing either 32-bits (before multiplication) or 64-bits
 // (after multiplication).

vendor/golang.org/x/crypto/nacl/secretbox/secretbox.go

@@ -35,8 +35,8 @@ This package is interoperable with NaCl: https://nacl.cr.yp.to/secretbox.html.
 package secretbox // import "golang.org/x/crypto/nacl/secretbox"
 
 import (
+	"golang.org/x/crypto/internal/poly1305"
 	"golang.org/x/crypto/internal/subtle"
-	"golang.org/x/crypto/poly1305"
 	"golang.org/x/crypto/salsa20/salsa"
 )
 

vendor/golang.org/x/crypto/salsa20/salsa/salsa20_amd64.s

@@ -2,6 +2,7 @@
 // Use of this source code is governed by a BSD-style
 // license that can be found in the LICENSE file.
 
+//go:build amd64 && !purego && gc
 // +build amd64,!purego,gc
 
 // This code was translated into a form compatible with 6a from the public

vendor/golang.org/x/crypto/ssh/certs.go

@@ -14,7 +14,7 @@ import (
 	"time"
 )
 
-// These constants from [PROTOCOL.certkeys] represent the algorithm names
+// These constants from [PROTOCOL.certkeys] represent the key algorithm names
 // for certificate types supported by this package.
 const (
 	CertAlgoRSAv01        = "ssh-rsa-cert-v01@openssh.com"
@@ -27,6 +27,14 @@ const (
 	CertAlgoSKED25519v01  = "sk-ssh-ed25519-cert-v01@openssh.com"
 )
 
+// These constants from [PROTOCOL.certkeys] represent additional signature
+// algorithm names for certificate types supported by this package.
+const (
+	CertSigAlgoRSAv01        = "ssh-rsa-cert-v01@openssh.com"
+	CertSigAlgoRSASHA2256v01 = "rsa-sha2-256-cert-v01@openssh.com"
+	CertSigAlgoRSASHA2512v01 = "rsa-sha2-512-cert-v01@openssh.com"
+)
+
 // Certificate types distinguish between host and user
 // certificates. The values can be set in the CertType field of
 // Certificate.
@@ -423,6 +431,12 @@ func (c *Certificate) SignCert(rand io.Reader, authority Signer) error {
 	}
 	c.SignatureKey = authority.PublicKey()
 
+	if v, ok := authority.(AlgorithmSigner); ok {
+		if v.PublicKey().Type() == KeyAlgoRSA {
+			authority = &rsaSigner{v, SigAlgoRSASHA2512}
+		}
+	}
+
 	sig, err := authority.Sign(rand, c.bytesForSigning())
 	if err != nil {
 		return err
@@ -431,8 +445,14 @@ func (c *Certificate) SignCert(rand io.Reader, authority Signer) error {
 	return nil
 }
 
+// certAlgoNames includes a mapping from signature algorithms to the
+// corresponding certificate signature algorithm. When a key type (such
+// as ED25516) is associated with only one algorithm, the KeyAlgo
+// constant is used instead of the SigAlgo.
 var certAlgoNames = map[string]string{
-	KeyAlgoRSA:        CertAlgoRSAv01,
+	SigAlgoRSA:        CertSigAlgoRSAv01,
+	SigAlgoRSASHA2256: CertSigAlgoRSASHA2256v01,
+	SigAlgoRSASHA2512: CertSigAlgoRSASHA2512v01,
 	KeyAlgoDSA:        CertAlgoDSAv01,
 	KeyAlgoECDSA256:   CertAlgoECDSA256v01,
 	KeyAlgoECDSA384:   CertAlgoECDSA384v01,

vendor/golang.org/x/crypto/ssh/cipher.go

@@ -18,7 +18,7 @@ import (
 	"io/ioutil"
 
 	"golang.org/x/crypto/chacha20"
-	"golang.org/x/crypto/poly1305"
+	"golang.org/x/crypto/internal/poly1305"
 )
 
 const (
@@ -394,6 +394,10 @@ func (c *gcmCipher) readCipherPacket(seqNum uint32, r io.Reader) ([]byte, error)
 	}
 	c.incIV()
 
+	if len(plain) == 0 {
+		return nil, errors.New("ssh: empty packet")
+	}
+
 	padding := plain[0]
 	if padding < 4 {
 		// padding is a byte, so it automatically satisfies
@@ -710,6 +714,10 @@ func (c *chacha20Poly1305Cipher) readCipherPacket(seqNum uint32, r io.Reader) ([
 	plain := c.buf[4:contentEnd]
 	s.XORKeyStream(plain, plain)
 
+	if len(plain) == 0 {
+		return nil, errors.New("ssh: empty packet")
+	}
+
 	padding := plain[0]
 	if padding < 4 {
 		// padding is a byte, so it automatically satisfies

vendor/golang.org/x/crypto/ssh/client.go

@@ -77,7 +77,7 @@ func NewClientConn(c net.Conn, addr string, config *ClientConfig) (Conn, <-chan
 	}
 
 	conn := &connection{
-		sshConn: sshConn{conn: c},
+		sshConn: sshConn{conn: c, user: fullConf.User},
 	}
 
 	if err := conn.clientHandshake(addr, &fullConf); err != nil {
@@ -115,12 +115,25 @@ func (c *connection) clientHandshake(dialAddress string, config *ClientConfig) e
 
 // verifyHostKeySignature verifies the host key obtained in the key
 // exchange.
-func verifyHostKeySignature(hostKey PublicKey, result *kexResult) error {
+func verifyHostKeySignature(hostKey PublicKey, algo string, result *kexResult) error {
 	sig, rest, ok := parseSignatureBody(result.Signature)
 	if len(rest) > 0 || !ok {
 		return errors.New("ssh: signature parse error")
 	}
 
+	// For keys, underlyingAlgo is exactly algo. For certificates,
+	// we have to look up the underlying key algorithm that SSH
+	// uses to evaluate signatures.
+	underlyingAlgo := algo
+	for sigAlgo, certAlgo := range certAlgoNames {
+		if certAlgo == algo {
+			underlyingAlgo = sigAlgo
+		}
+	}
+	if sig.Format != underlyingAlgo {
+		return fmt.Errorf("ssh: invalid signature algorithm %q, expected %q", sig.Format, underlyingAlgo)
+	}
+
 	return hostKey.Verify(result.H, sig)
 }
 

vendor/golang.org/x/crypto/ssh/common.go

@@ -69,11 +69,13 @@ var preferredKexAlgos = []string{
 // supportedHostKeyAlgos specifies the supported host-key algorithms (i.e. methods
 // of authenticating servers) in preference order.
 var supportedHostKeyAlgos = []string{
-	CertAlgoRSAv01, CertAlgoDSAv01, CertAlgoECDSA256v01,
+	CertSigAlgoRSASHA2512v01, CertSigAlgoRSASHA2256v01,
+	CertSigAlgoRSAv01, CertAlgoDSAv01, CertAlgoECDSA256v01,
 	CertAlgoECDSA384v01, CertAlgoECDSA521v01, CertAlgoED25519v01,
 
 	KeyAlgoECDSA256, KeyAlgoECDSA384, KeyAlgoECDSA521,
-	KeyAlgoRSA, KeyAlgoDSA,
+	SigAlgoRSASHA2512, SigAlgoRSASHA2256,
+	SigAlgoRSA, KeyAlgoDSA,
 
 	KeyAlgoED25519,
 }
@@ -90,16 +92,20 @@ var supportedCompressions = []string{compressionNone}
 // hashFuncs keeps the mapping of supported algorithms to their respective
 // hashes needed for signature verification.
 var hashFuncs = map[string]crypto.Hash{
-	KeyAlgoRSA:          crypto.SHA1,
+	SigAlgoRSA:               crypto.SHA1,
-	KeyAlgoDSA:          crypto.SHA1,
+	SigAlgoRSASHA2256:        crypto.SHA256,
-	KeyAlgoECDSA256:     crypto.SHA256,
+	SigAlgoRSASHA2512:        crypto.SHA512,
-	KeyAlgoECDSA384:     crypto.SHA384,
+	KeyAlgoDSA:               crypto.SHA1,
-	KeyAlgoECDSA521:     crypto.SHA512,
+	KeyAlgoECDSA256:          crypto.SHA256,
-	CertAlgoRSAv01:      crypto.SHA1,
+	KeyAlgoECDSA384:          crypto.SHA384,
-	CertAlgoDSAv01:      crypto.SHA1,
+	KeyAlgoECDSA521:          crypto.SHA512,
-	CertAlgoECDSA256v01: crypto.SHA256,
+	CertSigAlgoRSAv01:        crypto.SHA1,
-	CertAlgoECDSA384v01: crypto.SHA384,
+	CertSigAlgoRSASHA2256v01: crypto.SHA256,
-	CertAlgoECDSA521v01: crypto.SHA512,
+	CertSigAlgoRSASHA2512v01: crypto.SHA512,
+	CertAlgoDSAv01:           crypto.SHA1,
+	CertAlgoECDSA256v01:      crypto.SHA256,
+	CertAlgoECDSA384v01:      crypto.SHA384,
+	CertAlgoECDSA521v01:      crypto.SHA512,
 }
 
 // unexpectedMessageError results when the SSH message that we received didn't

vendor/golang.org/x/crypto/ssh/handshake.go

@@ -457,8 +457,15 @@ func (t *handshakeTransport) sendKexInit() error {
 
 	if len(t.hostKeys) > 0 {
 		for _, k := range t.hostKeys {
-			msg.ServerHostKeyAlgos = append(
+			algo := k.PublicKey().Type()
-				msg.ServerHostKeyAlgos, k.PublicKey().Type())
+			switch algo {
+			case KeyAlgoRSA:
+				msg.ServerHostKeyAlgos = append(msg.ServerHostKeyAlgos, []string{SigAlgoRSASHA2512, SigAlgoRSASHA2256, SigAlgoRSA}...)
+			case CertAlgoRSAv01:
+				msg.ServerHostKeyAlgos = append(msg.ServerHostKeyAlgos, []string{CertSigAlgoRSASHA2512v01, CertSigAlgoRSASHA2256v01, CertSigAlgoRSAv01}...)
+			default:
+				msg.ServerHostKeyAlgos = append(msg.ServerHostKeyAlgos, algo)
+			}
 		}
 	} else {
 		msg.ServerHostKeyAlgos = t.hostKeyAlgorithms
@@ -614,8 +621,22 @@ func (t *handshakeTransport) enterKeyExchange(otherInitPacket []byte) error {
 func (t *handshakeTransport) server(kex kexAlgorithm, algs *algorithms, magics *handshakeMagics) (*kexResult, error) {
 	var hostKey Signer
 	for _, k := range t.hostKeys {
-		if algs.hostKey == k.PublicKey().Type() {
+		kt := k.PublicKey().Type()
+		if kt == algs.hostKey {
 			hostKey = k
+		} else if signer, ok := k.(AlgorithmSigner); ok {
+			// Some signature algorithms don't show up as key types
+			// so we have to manually check for a compatible host key.
+			switch kt {
+			case KeyAlgoRSA:
+				if algs.hostKey == SigAlgoRSASHA2256 || algs.hostKey == SigAlgoRSASHA2512 {
+					hostKey = &rsaSigner{signer, algs.hostKey}
+				}
+			case CertAlgoRSAv01:
+				if algs.hostKey == CertSigAlgoRSASHA2256v01 || algs.hostKey == CertSigAlgoRSASHA2512v01 {
+					hostKey = &rsaSigner{signer, certToPrivAlgo(algs.hostKey)}
+				}
+			}
 		}
 	}
 
@@ -634,7 +655,7 @@ func (t *handshakeTransport) client(kex kexAlgorithm, algs *algorithms, magics *
 		return nil, err
 	}
 
-	if err := verifyHostKeySignature(hostKey, result); err != nil {
+	if err := verifyHostKeySignature(hostKey, algs.hostKey, result); err != nil {
 		return nil, err
 	}
 

vendor/golang.org/x/crypto/ssh/keys.go

@@ -939,6 +939,15 @@ func newDSAPrivateKey(key *dsa.PrivateKey) (Signer, error) {
 	return &dsaPrivateKey{key}, nil
 }
 
+type rsaSigner struct {
+	AlgorithmSigner
+	defaultAlgorithm string
+}
+
+func (s *rsaSigner) Sign(rand io.Reader, data []byte) (*Signature, error) {
+	return s.AlgorithmSigner.SignWithAlgorithm(rand, data, s.defaultAlgorithm)
+}
+
 type wrappedSigner struct {
 	signer crypto.Signer
 	pubKey PublicKey

vendor/golang.org/x/crypto/ssh/server.go

@@ -284,7 +284,7 @@ func (s *connection) serverHandshake(config *ServerConfig) (*Permissions, error)
 
 func isAcceptableAlgo(algo string) bool {
 	switch algo {
-	case KeyAlgoRSA, KeyAlgoDSA, KeyAlgoECDSA256, KeyAlgoECDSA384, KeyAlgoECDSA521, KeyAlgoSKECDSA256, KeyAlgoED25519, KeyAlgoSKED25519,
+	case SigAlgoRSA, SigAlgoRSASHA2256, SigAlgoRSASHA2512, KeyAlgoDSA, KeyAlgoECDSA256, KeyAlgoECDSA384, KeyAlgoECDSA521, KeyAlgoSKECDSA256, KeyAlgoED25519, KeyAlgoSKED25519,
 		CertAlgoRSAv01, CertAlgoDSAv01, CertAlgoECDSA256v01, CertAlgoECDSA384v01, CertAlgoECDSA521v01, CertAlgoSKECDSA256v01, CertAlgoED25519v01, CertAlgoSKED25519v01:
 		return true
 	}