Implement RSA helper functions

This commit is contained in:
Hanno Becker 2017-08-23 14:06:45 +01:00
parent a3ebec2423
commit e2e8b8da1d

View File

@ -71,6 +71,407 @@ static void mbedtls_zeroize( void *v, size_t n ) {
volatile unsigned char *p = (unsigned char*)v; while( n-- ) *p++ = 0;
}
/*
* Context-independent RSA helper functions.
*
* The following three functions
* - mbedtls_rsa_deduce_moduli
* - mbedtls_rsa_deduce_private
* - mbedtls_rsa_check_params
* are helper functions operating on the core RSA parameters
* (represented as MPI's). They do not use the RSA context structure
* and therefore need not be replaced when providing an alternative
* RSA implementation.
*
* Their purpose is to provide common MPI operations in the context
* of RSA that can be easily shared across multiple implementations.
*/
/*
* mbedtls_rsa_deduce_moduli
*
* Given the modulus N=PQ and a pair of public and private
* exponents E and D, respectively, factor N.
*
* Setting F := lcm(P-1,Q-1), the idea is as follows:
*
* (a) For any 1 <= X < N with gcd(X,N)=1, we have X^F = 1 modulo N, so X^(F/2)
* is a square root of 1 in Z/NZ. Since Z/NZ ~= Z/PZ x Z/QZ by CRT and the
* square roots of 1 in Z/PZ and Z/QZ are +1 and -1, this leaves the four
* possibilities X^(F/2) = (+-1, +-1). If it happens that X^(F/2) = (-1,+1)
* or (+1,-1), then gcd(X^(F/2) + 1, N) will be equal to one of the prime
* factors of N.
*
* (b) If we don't know F/2 but (F/2) * K for some odd (!) K, then the same
* construction still applies since (-)^K is the identity on the set of
* roots of 1 in Z/NZ.
*
* The public and private key primitives (-)^E and (-)^D are mutually inverse
* bijections on Z/NZ if and only if (-)^(DE) is the identity on Z/NZ, i.e.
* if and only if DE - 1 is a multiple of F, say DE - 1 = F * L.
* Splitting L = 2^t * K with K odd, we have
*
* DE - 1 = FL = (F/2) * (2^(t+1)) * K,
*
* so (F / 2) * K is among the numbers
*
* (DE - 1) >> 1, (DE - 1) >> 2, ..., (DE - 1) >> ord
*
* where ord is the order of 2 in (DE - 1).
* We can therefore iterate through these numbers apply the construction
* of (a) and (b) above to attempt to factor N.
*
*/
int mbedtls_rsa_deduce_moduli( mbedtls_mpi *N, mbedtls_mpi *D, mbedtls_mpi *E,
int (*f_rng)(void *, unsigned char *, size_t), void *p_rng,
mbedtls_mpi *P, mbedtls_mpi *Q )
{
/* Implementation note:
*
* Space-efficiency is given preference over time-efficiency here:
* several calculations are done in place and temporarily change
* the values of D and E.
*
* Specifically, D is replaced the largest odd divisor of DE - 1
* throughout the calculations.
*/
int ret = 0;
uint16_t attempt; /* Number of current attempt */
uint16_t iter; /* Number of squares computed in the current attempt */
uint16_t bitlen_half; /* Half the bitsize of the modulus N */
uint16_t order; /* Order of 2 in DE - 1 */
mbedtls_mpi K; /* Temporary used for two purposes:
* - During factorization attempts, stores a andom integer
* in the range of [0,..,N]
* - During verification, holding intermediate results.
*/
if( P == NULL || Q == NULL || P->p != NULL || Q->p != NULL )
return( MBEDTLS_ERR_MPI_BAD_INPUT_DATA );
if( mbedtls_mpi_cmp_int( N, 0 ) <= 0 ||
mbedtls_mpi_cmp_int( D, 1 ) <= 0 ||
mbedtls_mpi_cmp_mpi( D, N ) >= 0 ||
mbedtls_mpi_cmp_int( E, 1 ) <= 0 ||
mbedtls_mpi_cmp_mpi( E, N ) >= 0 )
{
return( MBEDTLS_ERR_MPI_BAD_INPUT_DATA );
}
/*
* Initializations and temporary changes
*/
mbedtls_mpi_init( &K );
mbedtls_mpi_init( P );
mbedtls_mpi_init( Q );
/* Replace D by DE - 1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( D, D, E ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( D, D, 1 ) );
if( ( order = mbedtls_mpi_lsb( D ) ) == 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
/* After this operation, D holds the largest odd divisor
* of DE - 1 for the original values of D and E. */
MBEDTLS_MPI_CHK( mbedtls_mpi_shift_r( D, order ) );
/* This is used to generate a few numbers around N / 2
* if no PRNG is provided. */
if( f_rng == NULL )
bitlen_half = mbedtls_mpi_bitlen( N ) / 2;
/*
* Actual work
*/
for( attempt = 0; attempt < 30; ++attempt )
{
/* Generate some number in [0,N], either randomly
* if a PRNG is given, or try numbers around N/2 */
if( f_rng != NULL )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_fill_random( &K,
mbedtls_mpi_size( N ),
f_rng, p_rng ) );
}
else
{
MBEDTLS_MPI_CHK( mbedtls_mpi_lset( &K, 1 ) ) ;
MBEDTLS_MPI_CHK( mbedtls_mpi_shift_l( &K, bitlen_half ) ) ;
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( &K, &K, attempt + 1 ) );
}
/* Check if gcd(K,N) = 1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_gcd( P, &K, N ) );
if( mbedtls_mpi_cmp_int( P, 1 ) != 0 )
continue;
/* Go through K^X + 1, K^(2X) + 1, K^(4X) + 1, ...
* and check whether they have nontrivial GCD with N. */
MBEDTLS_MPI_CHK( mbedtls_mpi_exp_mod( &K, &K, D, N,
Q /* temporarily use Q for storing Montgomery
* multiplication helper values */ ) );
for( iter = 1; iter < order; ++iter )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( &K, &K, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_gcd( P, &K, N ) );
if( mbedtls_mpi_cmp_int( P, 1 ) == 1 &&
mbedtls_mpi_cmp_mpi( P, N ) == -1 )
{
/*
* Have found a nontrivial divisor P of N.
* Set Q := N / P and verify D, E.
*/
MBEDTLS_MPI_CHK( mbedtls_mpi_div_mpi( Q, &K, N, P ) );
/*
* Verify that DE - 1 is indeed a multiple of
* lcm(P-1, Q-1), i.e. that it's a multiple of both
* P-1 and Q-1.
*/
/* Restore DE - 1 and temporarily replace P, Q by P-1, Q-1. */
MBEDTLS_MPI_CHK( mbedtls_mpi_shift_l( D, order ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( Q, Q, 1 ) );
/* Compute DE-1 mod P-1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &K, D, P ) );
if( mbedtls_mpi_cmp_int( &K, 0 ) != 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
/* Compute DE-1 mod Q-1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &K, D, Q ) );
if( mbedtls_mpi_cmp_int( &K, 0 ) != 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
/*
* All good, restore P, Q and D and return.
*/
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( Q, Q, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( D, D, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_div_mpi( D, NULL, D, E ) );
goto cleanup;
}
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( &K, &K, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &K, &K, &K ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &K, &K, N ) );
}
}
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
cleanup:
mbedtls_mpi_free( &K );
return( ret );
}
/*
* Given P, Q and the public exponent E, deduce D.
* This is essentially a modular inversion.
*/
int mbedtls_rsa_deduce_private( mbedtls_mpi *P, mbedtls_mpi *Q,
mbedtls_mpi *D, mbedtls_mpi *E )
{
int ret = 0;
mbedtls_mpi K;
if( D == NULL || mbedtls_mpi_cmp_int( D, 0 ) != 0 )
return( MBEDTLS_ERR_MPI_BAD_INPUT_DATA );
if( mbedtls_mpi_cmp_int( P, 1 ) <= 0 ||
mbedtls_mpi_cmp_int( Q, 1 ) <= 0 ||
mbedtls_mpi_cmp_int( E, 0 ) == 0 )
{
return( MBEDTLS_ERR_MPI_BAD_INPUT_DATA );
}
mbedtls_mpi_init( &K );
/* Temporarily replace P and Q by P-1 and Q-1, respectively. */
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( Q, Q, 1 ) );
/* Temporarily compute the gcd(P-1, Q-1) in D. */
MBEDTLS_MPI_CHK( mbedtls_mpi_gcd( D, P, Q ) );
/* Compute LCM(P-1, Q-1) in K */
MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &K, P, Q ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_div_mpi( &K, NULL, &K, D ) );
/* Compute modular inverse of E in LCM(P-1, Q-1) */
MBEDTLS_MPI_CHK( mbedtls_mpi_inv_mod( D, E, &K ) );
/* Restore P and Q. */
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( Q, Q, 1 ) );
/* Double-check result */
MBEDTLS_MPI_CHK( mbedtls_rsa_check_params( NULL, P, Q, D, E, NULL, NULL ) );
cleanup:
mbedtls_mpi_free( &K );
return( ret );
}
/*
* Check that core RSA parameters are sane.
*
* Note that the inputs are not declared const and may be
* altered on an unsuccessful run.
*/
int mbedtls_rsa_check_params( mbedtls_mpi *N, mbedtls_mpi *P, mbedtls_mpi *Q,
mbedtls_mpi *D, mbedtls_mpi *E,
int (*f_rng)(void *, unsigned char *, size_t),
void *p_rng )
{
int ret = 0;
mbedtls_mpi K;
mbedtls_mpi_init( &K );
/*
* Step 1: If PRNG provided, check that P and Q are prime
*/
if( f_rng != NULL && P != NULL &&
( ret = mbedtls_mpi_is_prime( P, f_rng, p_rng ) ) != 0 )
{
goto cleanup;
}
if( f_rng != NULL && Q != NULL &&
( ret = mbedtls_mpi_is_prime( Q, f_rng, p_rng ) ) != 0 )
{
goto cleanup;
}
/*
* Step 2: Check that N = PQ
*/
if( P != NULL && Q != NULL && N != NULL )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &K, P, Q ) );
if( mbedtls_mpi_cmp_mpi( &K, N ) != 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
}
/*
* Step 3: Check that D, E are inverse modulo P-1 and Q-1
*/
if( P != NULL && Q != NULL && D != NULL && E != NULL )
{
/* Temporarily replace P, Q by P-1, Q-1. */
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( Q, Q, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &K, D, E ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( &K, &K, 1 ) );
/* Compute DE-1 mod P-1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &K, &K, P ) );
if( mbedtls_mpi_cmp_int( &K, 0 ) != 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
/* Compute DE-1 mod Q-1 */
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &K, &K, Q ) );
if( mbedtls_mpi_cmp_int( &K, 0 ) != 0 )
{
ret = MBEDTLS_ERR_MPI_BAD_INPUT_DATA;
goto cleanup;
}
/* Restore P, Q. */
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( P, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_add_int( Q, Q, 1 ) );
}
cleanup:
mbedtls_mpi_free( &K );
return( ret );
}
int mbedtls_rsa_deduce_crt( const mbedtls_mpi *P, const mbedtls_mpi *Q,
const mbedtls_mpi *D, mbedtls_mpi *DP,
mbedtls_mpi *DQ, mbedtls_mpi *QP )
{
int ret = 0;
mbedtls_mpi K;
mbedtls_mpi_init( &K );
if( DP != NULL )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( &K, P, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( DP, D, &K ) );
}
if( DQ != NULL )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_sub_int( &K, Q, 1 ) );
MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( DQ, D, &K ) );
}
if( QP != NULL )
{
MBEDTLS_MPI_CHK( mbedtls_mpi_inv_mod( QP, Q, P ) );
}
cleanup:
mbedtls_mpi_free( &K );
return( ret );
}
{
int ret = 0;
cleanup:
return( ret );
}
/*
* Initialize an RSA context
*/