I would like to report a possible bug in rsa_prepare_blinding function in rsa.c (https://github.com/ARMmbed/mbedtls/blob/v2.26.0/library/rsa.c). I am not sure if it is a real issue, but I think that there is a possibility to use uninitialized variable ret:

static int rsa_prepare_blinding( mbedtls_rsa_context *ctx,
                 int (*f_rng)(void *, unsigned char *, size_t), void *p_rng )
{
    int ret, count = 0; <--- uninitialized variable ret
    mbedtls_mpi R;

    mbedtls_mpi_init( &R );

    if( ctx->Vf.p != NULL )
    {
        /* We already have blinding values, just update them by squaring */
        MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &ctx->Vi, &ctx->Vi, &ctx->Vi ) );
        MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &ctx->Vi, &ctx->Vi, &ctx->N ) );
        MBEDTLS_MPI_CHK( mbedtls_mpi_mul_mpi( &ctx->Vf, &ctx->Vf, &ctx->Vf ) );
        MBEDTLS_MPI_CHK( mbedtls_mpi_mod_mpi( &ctx->Vf, &ctx->Vf, &ctx->N ) );

        goto cleanup; <--- going to cleanup without setting a value of ret
    }

(Skipping lines for readability)

 cleanup:
    mbedtls_mpi_free( &R );

    return( ret ); <--- returning uninitialized variable ret
}

Best regards,

grapix121