Prior to OpenIddict RC3, the events model used by the OpenID Connect server middleware (i.e the OIDC server framework behind OpenIddict) was deliberately not accessible due to the nature of OpenIddict: being initially designed for non-experts, exposing such a powerful API – that allows altering the way OpenID Connect requests are processed – didn't seem like a good idea at first sight.

With time, the core audience of OpenIddict has evolved a bit to not only include beginners but also developers who were already familiar with OAuth/OpenID Connect or had used OAuthAuthorizationServerMiddleware or ASOS in the past. For them, the fact OpenIddict didn't allow them to take control of the request processing pipeline was often a blocker. With the introduction of OpenIddict RC3, we're changing that.

Introducing event handlers

The events model is structured around IOpenIddictServerEventHandler<TEvent> and IOpenIddictValidationEventHandler<TEvent>. These 2 interfaces represent handlers that are invoked every time an event of type TEvent is triggered by the OpenIddict server or validation handlers.

At the time of writing, OpenIddict exposes 44 server events – grouped into the OpenIddictServerEvents static class to make them easier to find – and 5 validation events, exposed under OpenIddictValidationEvents.

Each event represents a specific moment in the request processing pipeline (e.g the moment the OpenIddict server determines whether the request is an OpenID Connect request it should handle, the moment it extracts it, handles it or returns a response).

Creating and registering a custom event handler

Creating an event handler is straightforward: pick the event you need and add a class that implements either IOpenIddictServerEventHandler<TEvent> (for a handler that receives events triggered by the OpenIddict server services) or IOpenIddictValidationEventHandler<TEvent> (for a handler that receives events triggered by the OpenIddict token validation services):

For instance, to return custom metadata in the discovery document, you'll need to implement IOpenIddictServerEventHandler<OpenIddictServerEvents.HandleConfigurationRequest>:

1
2
3
4
5
6
7
8
9

public class MyEventHandler : IOpenIddictServerEventHandler<OpenIddictServerEvents.HandleConfigurationRequest>
{
    public Task HandleAsync(OpenIddictServerEvents.HandleConfigurationRequest notification, CancellationToken cancellationToken)
    {
        notification.Context.Metadata["company_name"] = "Contoso";

        return Task.CompletedTask;
    }
}

To register it, use options.AddEventHandler<TEvent, THandler>() (by default, the handler is registered as a scoped service):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21

public void ConfigureServices(IServiceCollection services)
{
    services.AddOpenIddict()

        // Register the OpenIddict core services.
        .AddCore(options =>
        {
            // ...
        })

        // Register the OpenIddict server handler.
        .AddServer(options =>
        {
            // ...

            options.AddEventHandler<OpenIddictServerEvents.HandleConfigurationRequest, MyEventHandler>();
        })

        // Register the OpenIddict validation handler.
        .AddValidation();
}

Read more

OpenIddict RC3 is now available on NuGet.org:

• OpenIddict – 1.0.0-rc3-final (for ASP.NET Core 1.x)
• OpenIddict – 2.0.0-rc3-final (for ASP.NET Core 2.x)

What's new in this release?

The OpenIddict services registration APIs have been revamped

In this release, we focused on reworking the OpenIddict registration APIs to offer a better user experience.

As part of this change, we split the OpenIddict services into three areas - Core, Server and Validation - and the IServiceCollection APIs have been updated to reflect that:

Each specialized builder only exposes the options that are relevant to its specific area:

Of course, the calls to AddCore(), AddServer() and AddValidation() can be chained:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39

services.AddOpenIddict()

    // Register the OpenIddict core services.
    .AddCore(options =>
    {
        // Register the Entity Framework stores and models.
        options.UseEntityFrameworkCore()
               .UseDbContext<ApplicationDbContext>();
    })

    // Register the OpenIddict server handler.
    .AddServer(options =>
    {
        // Register the ASP.NET Core MVC binder used by OpenIddict.
        // Note: if you don't call this method, you won't be able to
        // bind OpenIdConnectRequest or OpenIdConnectResponse parameters.
        options.UseMvc();

        // Enable the authorization, logout, token and userinfo endpoints.
        options.EnableAuthorizationEndpoint("/connect/authorize")
               .EnableLogoutEndpoint("/connect/logout")
               .EnableTokenEndpoint("/connect/token")
               .EnableUserinfoEndpoint("/api/userinfo");

        // Note: the Mvc.Client sample only uses the code flow and the password flow, but you
        // can enable the other flows if you need to support implicit or client credentials.
        options.AllowAuthorizationCodeFlow()
               .AllowPasswordFlow()
               .AllowRefreshTokenFlow();

        // During development, you can disable the HTTPS requirement.
        options.DisableHttpsRequirement();
    })

    // Register the OpenIddict validation handler.
    // Note: the OpenIddict validation handler is only compatible with the
    // default token format or with reference tokens and cannot be used with
    // JWT tokens. For JWT tokens, use the Microsoft JWT bearer handler.
    .AddValidation();

Introducing these specialized builders was also a great opportunity to revisit how the OpenIddict entities are registered. In the RC2 bits, this is controlled by the services.AddOpenIddict<...>() method, that determines which entities are used depending on the overload.

In RC3, the generic services.AddOpenIddict<...>() methods have been removed and replaced by a more explicit pattern:

Read more

Today, I'm really pleased to announce that the aspnet-contrib social providers are now RTM.

I'd like to thank all the contributors – 47 at the time of writing – who helped make that possible, with a special mention to Chino Chang, who did a fantastic job by manually porting most of the social providers to the ASP.NET Core 2.0 authentication stack: sir, you rock!

Each provider comes with 2 versions: one for ASP.NET Core 1.x and one for ASP.NET Core 2.x. You can find the complete list at the end of this post.

OAuth 2.0 social providers

OpenID 2.0 social providers

Earlier today, new OpenIddict packages were pushed to NuGet.org:

• OpenIddict – 1.0.0-rc2-final (for ASP.NET Core 1.x)
• OpenIddict – 2.0.0-rc2-final (for ASP.NET Core 2.x)

What's new?

Starting with RC2, using OpenIddict with third-party client applications (i.e applications you don't own and are managed by someone else) is officially supported. For that, new features – that were still work in progress in the previous iterations – have been added:

• A new application permissions feature was added, which allows controlling and limiting the features a client application can use. For instance, to allow a client application to use only the authorization code flow and the logout endpoint, the following permissions can be granted:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

await _applicationManager.CreateAsync(new OpenIddictApplicationDescriptor
{
    ClientId = "mvc",
    ClientSecret = "901564A5-E7FE-42CB-B10D-61EF6A8F3654",
    DisplayName = "MVC client application",
    PostLogoutRedirectUris = { new Uri("http://localhost:53507/signout-callback-oidc") },
    RedirectUris = { new Uri("http://localhost:53507/signin-oidc") },
    Permissions =
    {
        OpenIddictConstants.Permissions.Endpoints.Authorization,
        OpenIddictConstants.Permissions.Endpoints.Logout,
        OpenIddictConstants.Permissions.Endpoints.Token,

        OpenIddictConstants.Permissions.GrantTypes.AuthorizationCode
    }
});

For more information about this feature, you can read the corresponding documentation.

• A new opt-in scope validation option was added. When it is enabled, OpenIddict automatically rejects authorization and token requests that specify unregistered scopes. Scopes can be registered statically using options.RegisterScopes([list of authorized scopes]) or dynamically, using OpenIddictScopeManager.CreateAsync():

1
2
3
4
5
6
7
8
9
10
11
12

services.AddOpenIddict(options =>
{
    // ...

    // Mark the "email" and "profile" scopes as valid scopes.
    options.RegisterScopes(OpenIdConnectConstants.Scopes.Email,
                           OpenIdConnectConstants.Scopes.Profile);

    // Enable scope validation, so that authorization and token requests
    // that specify unregistered scopes are automatically rejected.
    options.EnableScopeValidation();
});

1
2
3
4
5
6

await _scopeManager.CreateAsync(new OpenIddictScopeDescriptor
{
    Description = "Grants access to the reporting API",
    DisplayName = "Reporting API",
    Name = "reporting"
});

• The introspection endpoint was updated to reject access tokens that don't have any audience (since OpenIddict can no longer assume all the registered applications are fully trusted). This change requires updating your code to explicitly attach a resource to your tokens.

1
2
3
4
5
6

var ticket = new AuthenticationTicket(
    new ClaimsPrincipal(identity),
    new AuthenticationProperties(),
    OpenIdConnectServerDefaults.AuthenticationScheme);

ticket.SetResources("reporting");

Read more

Earlier today, Microsoft released two security advisories for vulnerabilities I discovered in the ASP.NET Core 2.0 VS2017 templates and reported late October:

• CVE-2018-0785 - ASP.NET Core Templates enable Cross Site Request Forgery: a CSRF vulnerability in ManageController.GenerateRecoveryCodes() allows re-generating the recovery codes associated with a victim's account, which may result in a definitive account lockout (aka per-user denial of service).

• CVE-2018-0784 - ASP.NET Core Templates enable Elevation Of Privilege Vulnerability.

While the first one is a very classic case of cross-site request forgery (CSRF), the second one is a bit more interesting as it relies on a specificity of ASP.NET Core MVC to be exploited.

Where was the vulnerability located?

When you create a new project based on the ASP.NET Core 2.0 templates offered by Visual Studio 2017 and opt for individual authentication (that uses ASP.NET Core Identity under the hood), a new ManageController and a bunch of views are automatically added to the resulting solution.

One of the actions exposed by this controller, EnableAuthenticator, allows you to generate a shared secret you can use in your favorite TOTP-based application (like Microsoft or Google Authenticator) to enable 2-factor authentication:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

[HttpGet]
public async Task<IActionResult> EnableAuthenticator()
{
    var user = await _userManager.GetUserAsync(User);
    if (user == null)
    {
        throw new ApplicationException($"Unable to load user with ID '{_userManager.GetUserId(User)}'.");
    }

    var unformattedKey = await _userManager.GetAuthenticatorKeyAsync(user);
    if (string.IsNullOrEmpty(unformattedKey))
    {
        await _userManager.ResetAuthenticatorKeyAsync(user);
        unformattedKey = await _userManager.GetAuthenticatorKeyAsync(user);
    }

    var model = new EnableAuthenticatorViewModel
    {
        SharedKey = FormatKey(unformattedKey),
        AuthenticatorUri = GenerateQrCodeUri(user.Email, unformattedKey)
    };

    return View(model);
}

For that, a random value is generated server-side by Identity, stored in the database and used to build a special otpauth://totp URI which is rendered in the Razor view as an HTML attribute (that can be optionally read by a JS library to generate a QR code image client-side):

1
2
3
4
5
6

<li>
    <p>Scan the QR Code or enter this key <kbd>@Model.SharedKey</kbd> into your two factor authenticator app. Spaces and casing do not matter.</p>
    <div class="alert alert-info">To enable QR code generation please read our <a href="https://go.microsoft.com/fwlink/?Linkid=852423">documentation</a>.</div>
    <div id="qrCode"></div>
    <div id="qrCodeData" data-url="@Html.Raw(Model.AuthenticatorUri)"></div>
</li>

1
2
3
4
5
6

<li>
    <p>Scan the QR Code or enter this key <kbd>xuht bbzd 3juv 4kt6 glpb l5tc jbc6 yjsn</kbd> into your two factor authenticator app. Spaces and casing do not matter.</p>
    <div class="alert alert-info">To enable QR code generation please read our <a href="https://go.microsoft.com/fwlink/?Linkid=852423">documentation</a>.</div>
    <div id="qrCode"></div>
    <div id="qrCodeData" data-url="otpauth://totp/myappname:alice@bob.com?secret=XUHTBBZD3JUV4KT6GLPBL5TCJBC6YJSN&issuer=myappname&digits=6"></div>
</li>

What does the vulnerability consist in?

As you can see, the view uses Html.Raw to render the generated authenticator URI, which is confirmed by the fact the & character is not properly HTML-encoded (it should be rendered in the HTML document as & since it's a special character).

Innocently, you might think it's not a big deal since AuthenticatorUri is generated server-side (and thus can't be directly set by an attacker).

Unfortunately, that's not exact: while it's true that adding an AuthenticatorUri paramater will have initially no effect on the GET EnableAuthenticator action (since the value will be always overridden when setting EnableAuthenticatorViewModel.AuthenticatorUri before returning the view), the query string value will be used if the form is re-displayed.

Yet, that's exactly what the POST action does if the model state is not valid (e.g because the 2FA confirmation code was not correctly typed by the user or was invalid, which may happen if the date/time differ between the server and the device that generated the 2FA code):

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31

[HttpPost]
[ValidateAntiForgeryToken]
public async Task<IActionResult> EnableAuthenticator(EnableAuthenticatorViewModel model)
{
    if (!ModelState.IsValid)
    {
        return View(model);
    }

    var user = await _userManager.GetUserAsync(User);
    if (user == null)
    {
        throw new ApplicationException($"Unable to load user with ID '{_userManager.GetUserId(User)}'.");
    }

    // Strip spaces and hypens
    var verificationCode = model.Code.Replace(" ", string.Empty).Replace("-", string.Empty);

    var is2faTokenValid = await _userManager.VerifyTwoFactorTokenAsync(
        user, _userManager.Options.Tokens.AuthenticatorTokenProvider, verificationCode);

    if (!is2faTokenValid)
    {
        ModelState.AddModelError("model.Code", "Verification code is invalid.");
        return View(model);
    }

    await _userManager.SetTwoFactorEnabledAsync(user, true);
    _logger.LogInformation("User with ID {UserId} has enabled 2FA with an authenticator app.", user.Id);
    return RedirectToAction(nameof(GenerateRecoveryCodes));
}

The bad news is that using this specificity alongside Html.Raw can result in a XSS vulnerability being exploitable since an attacker can craft a special URL containing a malicious JavaScript payload that will be executed by the victim's browser if he or she sends an invalid 2FA confirmation code.

For instance, if a victim visits https://localhost:44370/Manage/EnableAuthenticator?AuthenticatorUri=%22%3E%3C/div%3E%00%00%00%00%00%00%00%3Cscript%3Ealert(%22XSS%22)%3C/script%3E (which uses a special pattern to bypass Chrome 61's XSS auditor feature) and enters an invalid code, the following HTML source will be rendered:

1
2
3
4
5
6

<li>
    <p>Scan the QR Code or enter this key <kbd></kbd> into your two factor authenticator app. Spaces and casing do not matter.</p>
    <div class="alert alert-info">To enable QR code generation please read our <a href="https://go.microsoft.com/fwlink/?Linkid=852423">documentation</a>.</div>
    <div id="qrCode"></div>
    <div id="qrCodeData" data-url=""></div>       <script>alert("XSS")</script>"></div>
</li>

How can I fix the vulnerability?

Detailed instructions are listed on Microsoft's announcement.

So, what's wrong with Html.Raw?

In some cases, disabling HTML encoding is just unavoidable, for instance if the content is already encoded: not doing that would result in double-encoding.

So, why shouldn't you use Html.Raw in your views? It's actually all about intent: Html.Raw encourages developers to "disable" HTML encoding at the wrong layer (i.e in the view rather than in the controller itself) by declaring already-encoded HTML properties as simple string properties in their view model rather than as HtmlString properties.

Using this type instead of Html.Raw has two big advantages, that would have helped avoid this vulnerability in the first place:

• By marking your view model properties as HtmlString, you force the developer in charge of maintaining the controller code to manually create a HtmlString instance from an existing string: new HtmlString(existingEncodedString). Technically, the result is exactly the same, but it's now clear by looking at either your view model or your controller code that a property is not a simple string, but something hightly susceptible of containing HTML payloads you should treat with extreme care.

• HtmlString is not bindable: a form model containing a HtmlString will never be populated with user input. In this specific case, exploiting the bug wouldn't have been possible with HtmlString.

Please, do yourself a favor: stop using Html.Raw.