linux/lib/mpi/mpi-mul.c
Ignat Korchagin f145d411a6 crypto: rsa - implement Chinese Remainder Theorem for faster private key operations
Changes from v1:
  * exported mpi_sub and mpi_mul, otherwise the build fails when RSA is a module

The kernel RSA ASN.1 private key parser already supports only private keys with
additional values to be used with the Chinese Remainder Theorem [1], but these
values are currently not used.

This rudimentary CRT implementation speeds up RSA private key operations for the
following Go benchmark up to ~3x.

This implementation also tries to minimise the allocation of additional MPIs,
so existing MPIs are reused as much as possible (hence the variable names are a
bit weird).

The benchmark used:

```
package keyring_test

import (
	"crypto"
	"crypto/rand"
	"crypto/rsa"
	"crypto/x509"
	"io"
	"syscall"
	"testing"
	"unsafe"
)

type KeySerial int32
type Keyring int32

const (
	KEY_SPEC_PROCESS_KEYRING Keyring = -2
	KEYCTL_PKEY_SIGN                 = 27
)

var (
	keyTypeAsym = []byte("asymmetric\x00")
	sha256pkcs1 = []byte("enc=pkcs1 hash=sha256\x00")
)

func (keyring Keyring) LoadAsym(desc string, payload []byte) (KeySerial, error) {
	cdesc := []byte(desc + "\x00")
	serial, _, errno := syscall.Syscall6(syscall.SYS_ADD_KEY, uintptr(unsafe.Pointer(&keyTypeAsym[0])), uintptr(unsafe.Pointer(&cdesc[0])), uintptr(unsafe.Pointer(&payload[0])), uintptr(len(payload)), uintptr(keyring), uintptr(0))
	if errno == 0 {
		return KeySerial(serial), nil
	}

	return KeySerial(serial), errno
}

type pkeyParams struct {
	key_id         KeySerial
	in_len         uint32
	out_or_in2_len uint32
	__spare        [7]uint32
}

// the output signature buffer is an input parameter here, because we want to
// avoid Go buffer allocation leaking into our benchmarks
func (key KeySerial) Sign(info, digest, out []byte) error {
	var params pkeyParams
	params.key_id = key
	params.in_len = uint32(len(digest))
	params.out_or_in2_len = uint32(len(out))

	_, _, errno := syscall.Syscall6(syscall.SYS_KEYCTL, KEYCTL_PKEY_SIGN, uintptr(unsafe.Pointer(&params)), uintptr(unsafe.Pointer(&info[0])), uintptr(unsafe.Pointer(&digest[0])), uintptr(unsafe.Pointer(&out[0])), uintptr(0))
	if errno == 0 {
		return nil
	}

	return errno
}

func BenchmarkSign(b *testing.B) {
	priv, err := rsa.GenerateKey(rand.Reader, 2048)
	if err != nil {
		b.Fatalf("failed to generate private key: %v", err)
	}

	pkcs8, err := x509.MarshalPKCS8PrivateKey(priv)
	if err != nil {
		b.Fatalf("failed to serialize the private key to PKCS8 blob: %v", err)
	}

	serial, err := KEY_SPEC_PROCESS_KEYRING.LoadAsym("test rsa key", pkcs8)
	if err != nil {
		b.Fatalf("failed to load the private key into the keyring: %v", err)
	}

	b.Logf("loaded test rsa key: %v", serial)

	digest := make([]byte, 32)
	_, err = io.ReadFull(rand.Reader, digest)
	if err != nil {
		b.Fatalf("failed to generate a random digest: %v", err)
	}

	sig := make([]byte, 256)
	for n := 0; n < b.N; n++ {
		err = serial.Sign(sha256pkcs1, digest, sig)
		if err != nil {
			b.Fatalf("failed to sign the digest: %v", err)
		}
	}

	err = rsa.VerifyPKCS1v15(&priv.PublicKey, crypto.SHA256, digest, sig)
	if err != nil {
		b.Fatalf("failed to verify the signature: %v", err)
	}
}
```

[1]: https://en.wikipedia.org/wiki/RSA_(cryptosystem)#Using_the_Chinese_remainder_algorithm

Signed-off-by: Ignat Korchagin <ignat@cloudflare.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
2022-06-24 17:12:29 +08:00

93 lines
2.2 KiB
C

/* mpi-mul.c - MPI functions
* Copyright (C) 1994, 1996, 1998, 2001, 2002,
* 2003 Free Software Foundation, Inc.
*
* This file is part of Libgcrypt.
*
* Note: This code is heavily based on the GNU MP Library.
* Actually it's the same code with only minor changes in the
* way the data is stored; this is to support the abstraction
* of an optional secure memory allocation which may be used
* to avoid revealing of sensitive data due to paging etc.
*/
#include "mpi-internal.h"
void mpi_mul(MPI w, MPI u, MPI v)
{
mpi_size_t usize, vsize, wsize;
mpi_ptr_t up, vp, wp;
mpi_limb_t cy;
int usign, vsign, sign_product;
int assign_wp = 0;
mpi_ptr_t tmp_limb = NULL;
if (u->nlimbs < v->nlimbs) {
/* Swap U and V. */
usize = v->nlimbs;
usign = v->sign;
up = v->d;
vsize = u->nlimbs;
vsign = u->sign;
vp = u->d;
} else {
usize = u->nlimbs;
usign = u->sign;
up = u->d;
vsize = v->nlimbs;
vsign = v->sign;
vp = v->d;
}
sign_product = usign ^ vsign;
wp = w->d;
/* Ensure W has space enough to store the result. */
wsize = usize + vsize;
if (w->alloced < wsize) {
if (wp == up || wp == vp) {
wp = mpi_alloc_limb_space(wsize);
assign_wp = 1;
} else {
mpi_resize(w, wsize);
wp = w->d;
}
} else { /* Make U and V not overlap with W. */
if (wp == up) {
/* W and U are identical. Allocate temporary space for U. */
up = tmp_limb = mpi_alloc_limb_space(usize);
/* Is V identical too? Keep it identical with U. */
if (wp == vp)
vp = up;
/* Copy to the temporary space. */
MPN_COPY(up, wp, usize);
} else if (wp == vp) {
/* W and V are identical. Allocate temporary space for V. */
vp = tmp_limb = mpi_alloc_limb_space(vsize);
/* Copy to the temporary space. */
MPN_COPY(vp, wp, vsize);
}
}
if (!vsize)
wsize = 0;
else {
mpihelp_mul(wp, up, usize, vp, vsize, &cy);
wsize -= cy ? 0:1;
}
if (assign_wp)
mpi_assign_limb_space(w, wp, wsize);
w->nlimbs = wsize;
w->sign = sign_product;
if (tmp_limb)
mpi_free_limb_space(tmp_limb);
}
EXPORT_SYMBOL_GPL(mpi_mul);
void mpi_mulm(MPI w, MPI u, MPI v, MPI m)
{
mpi_mul(w, u, v);
mpi_tdiv_r(w, w, m);
}
EXPORT_SYMBOL_GPL(mpi_mulm);