تحرير

مشاركة عبر

TLS/SSL best practices

  • مقالة

TLS (Transport Layer Security) is a cryptographic protocol designed to secure communication between two computers over the internet. The TLS protocol is exposed in .NET via the SslStream class.

This article presents best practices for setting up secure communication between client and server and assumes use of .NET. For best practices with .NET Framework, see Transport Layer Security (TLS) best practices with the .NET Framework.

Select TLS version

While it is possible to specify the version of the TLS protocol to be used via the EnabledSslProtocols property, it is recommended to defer to the operating system settings by using None value (this is the default).

Deferring the decision to the OS automatically uses the most recent version of TLS available and lets the application pick up changes after OS upgrades. The operating system may also prevent use of TLS versions which are no longer considered secure.

Select cipher suites

SslStream allows users to specify which cipher suites can be negotiated by the TLS handshake via the CipherSuitesPolicy class. As with TLS versions, it's recommended to let the OS decide which are the best cipher suites to negotiate with, and, therefore, it's recommended to avoid using CipherSuitesPolicy.

Note

CipherSuitesPolicy is not supported on Windows and attempts to instantiate it will cause NotSupportedException to be thrown.

Specify a server certificate

When authenticating as a server, SslStream requires an X509Certificate2 instance. It is recommended to always use an X509Certificate2 instance which also contains the private key.

There are multiple ways that a server certificate can be passed to SslStream:

The recommended approach is to use the SslServerAuthenticationOptions.ServerCertificateContext property. When the certificate is obtained by one of the other two ways, a SslStreamCertificateContext instance is created internally by the SslStream implementation. Creating a SslStreamCertificateContext involves building an X509Chain which is a CPU intensive operation. It is more efficient to create a SslStreamCertificateContext once and reuse it for multiple SslStream instances.

Reusing SslStreamCertificateContext instances also enables additional features such us TLS session resumption on Linux servers.

Custom X509Certificate validation

There are certain scenarios in which the default certificate validation procedure isn't adequate and some custom validation logic is required. Parts of the validation logic can be customized by specifying SslClientAuthenticationOptions.CertificateChainPolicy or SslServerAuthenticationOptions.CertificateChainPolicy. Alternatively, completely custom logic can be provided via the <System.Net.Security.SslClientAuthenticationOptions.RemoteCertificateValidationCallback> property. For more information, see Custom certificate trust.

Custom certificate trust

When encountering a certificate that wasn't issued by any of the certificate authorities trusted by the machine (including self-signed certificates), the default certificate validation procedure will fail. One possible way to resolve this is to add the necessary issuer certificates to the machine's trusted store. That, however, might affect other applications on the system and is not always possible.

The alternative solution is to specify custom trusted root certificates via an X509ChainPolicy. To specify a custom trust list that will be used instead of the system trust list during validation, consider the following example:

SslClientAuthenticationOptions clientOptions = new();

clientOptions.CertificateChainPolicy = new X509ChainPolicy()
{
    TrustMode = X509ChainTrustMode.CustomRootTrust,
    CustomTrustStore =
    {
        customIssuerCert
    }
};

Clients configured with the preceding policy would only accept certificates trusted by customIssuerCert.

Ignore specific validation errors

Consider an IoT device without a persistent clock. After powering on, the clock of the device would start many years in the past and, therefore, all certificates would be considered "not yet valid". Consider the following code that shows a validation callback implementation ignoring validity period violations.

static bool CustomCertificateValidationCallback(
    object sender,
    X509Certificate? certificate,
    X509Chain? chain,
    SslPolicyErrors sslPolicyErrors)
{
    // Anything that would have been accepted by default is OK
    if (sslPolicyErrors == SslPolicyErrors.None)
    {
        return true;
    }
    
    // If there is something wrong other than a chain processing error, don't trust it.
    if (sslPolicyErrors != SslPolicyErrors.RemoteCertificateChainErrors)
    {
        return false;
    }
    
    Debug.Assert(chain is not null);

    // If the reason for RemoteCertificateChainError is that the chain built empty, don't trust it.
    if (chain.ChainStatus.Length == 0)
    {
        return false;
    }

    foreach (X509ChainStatus status in chain.ChainStatus)
    {
        // If an error other than `NotTimeValid` (or `NoError`) is present, don't trust it.
        if ((status.Status & ~X509ChainStatusFlags.NotTimeValid) != X509ChainStatusFlags.NoError)
        {
            return false;
        }
    }

    return true;
}

Certificate pinning

Another situation where custom certificate validation is necessary is when clients expect servers to use a specific certificate, or a certificate from a small set of known certificates. This practice is known as certificate pinning. The following code snippet shows a validation callback which checks that the server presents a certificate with a specific known public key.

static bool CustomCertificateValidationCallback(
    object sender,
    X509Certificate? certificate,
    X509Chain? chain,
    SslPolicyErrors sslPolicyErrors)
{
    // If there is something wrong other than a chain processing error, don't trust it.
    if ((sslPolicyErrors & ~SslPolicyErrors.RemoteCertificateChainErrors) != 0)
    {
        return false;
    }
    
    Debug.Assert(certificate is not null);

    const string ExpectedPublicKey =
        "3082010A0282010100C204ECF88CEE04C2B3D850D57058CC9318EB5C" +
        "A86849B022B5F9959EB12B2C763E6CC04B604C4CEAB2B4C00F80B6B0" +
        "F972C98602F95C415D132B7F71C44BBCE9942E5037A6671C618CF641" +
        "42C546D31687279F74EB0A9D11522621736C844C7955E4D16BE8063D" +
        "481552ADB328DBAAFF6EFF60954A776B39F124D131B6DD4DC0C4FC53" +
        "B96D42ADB57CFEAEF515D23348E72271C7C2147A6C28EA374ADFEA6C" +
        "B572B47E5AA216DC69B15744DB0A12ABDEC30F47745C4122E19AF91B" +
        "93E6AD2206292EB1BA491C0C279EA3FB8BF7407200AC9208D98C5784" +
        "538105CBE6FE6B5498402785C710BB7370EF6918410745557CF9643F" +
        "3D2CC3A97CEB931A4C86D1CA850203010001";

    return certificate.GetPublicKeyString().Equals(ExpectedPublicKey);
}

Considerations for client certificate validation

Server applications need to be careful when requiring and validating client certificates. Certificates may contain the AIA (Authority Information Access) extension which specifies where the issuer certificate can be downloaded. The server may therefore attempt to download the issuer certificate from external server when building the X509Chain for the client certificate. Similarly, servers may need to contact external servers to ensure that the client certificate has not been revoked.

The need to contact external servers when building and validating the X509Chain may expose the application to denial of service attacks if the external servers are slow to respond. Therefore, server applications should configure the X509Chain building behavior using the CertificateChainPolicy.