How do you create a custom AuthorizeAttribute in ASP.NET Core?

The approach recommended by the ASP.Net Core team is to use the new policy design which is fully documented here. The basic idea behind the new approach is to use the new [Authorize] attribute to designate a "policy" (e.g. [Authorize( Policy = "YouNeedToBe18ToDoThis")] where the policy is registered in the application's Startup.cs to execute some block of code (i.e. ensure the user has an age claim where the age is 18 or older).

The policy design is a great addition to the framework and the ASP.Net Security Core team should be commended for its introduction. That said, it isn't well-suited for all cases. The shortcoming of this approach is that it fails to provide a convenient solution for the most common need of simply asserting that a given controller or action requires a given claim type. In the case where an application may have hundreds of discrete permissions governing CRUD operations on individual REST resources ("CanCreateOrder", "CanReadOrder", "CanUpdateOrder", "CanDeleteOrder", etc.), the new approach either requires repetitive one-to-one mappings between a policy name and a claim name (e.g. options.AddPolicy("CanUpdateOrder", policy => policy.RequireClaim(MyClaimTypes.Permission, "CanUpdateOrder));), or writing some code to perform these registrations at run time (e.g. read all claim types from a database and perform the aforementioned call in a loop). The problem with this approach for the majority of cases is that it's unnecessary overhead.

While the ASP.Net Core Security team recommends never creating your own solution, in some cases this may be the most prudent option with which to start.

The following is an implementation which uses the IAuthorizationFilter to provide a simple way to express a claim requirement for a given controller or action:

public class ClaimRequirementAttribute : TypeFilterAttribute
{
    public ClaimRequirementAttribute(string claimType, string claimValue) : base(typeof(ClaimRequirementFilter))
    {
        Arguments = new object[] {new Claim(claimType, claimValue) };
    }
}

public class ClaimRequirementFilter : IAuthorizationFilter
{
    readonly Claim _claim;

    public ClaimRequirementFilter(Claim claim)
    {
        _claim = claim;
    }

    public void OnAuthorization(AuthorizationFilterContext context)
    {
        var hasClaim = context.HttpContext.User.Claims.Any(c => c.Type == _claim.Type && c.Value == _claim.Value);
        if (!hasClaim)
        {
            context.Result = new ForbidResult();
        }
    }
}


[Route("api/resource")]
public class MyController : Controller
{
    [ClaimRequirement(MyClaimTypes.Permission, "CanReadResource")]
    [HttpGet]
    public IActionResult GetResource()
    {
        return Ok();
    }
}

I'm the asp.net security person. Firstly let me apologize that none of this is documented yet outside of the music store sample or unit tests, and it's all still being refined in terms of exposed APIs. Detailed documentation is here.

We don't want you writing custom authorize attributes. If you need to do that we've done something wrong. Instead, you should be writing authorization requirements.

Authorization acts upon Identities. Identities are created by authentication.

You say in comments you want to check a session ID in a header. Your session ID would be the basis for identity. If you wanted to use the Authorize attribute you'd write an authentication middleware to take that header and turn it into an authenticated ClaimsPrincipal. You would then check that inside an authorization requirement. Authorization requirements can be as complicated as you like, for example here's one that takes a date of birth claim on the current identity and will authorize if the user is over 18;

public class Over18Requirement : AuthorizationHandler<Over18Requirement>, IAuthorizationRequirement
{
  public override void Handle(AuthorizationHandlerContext context, Over18Requirement requirement)
  {
    if (!context.User.HasClaim(c => c.Type == ClaimTypes.DateOfBirth))
    {
      context.Fail();
      return;
    }

    var dobVal = context.User.FindFirst(c => c.Type == ClaimTypes.DateOfBirth).Value;
    var dateOfBirth = Convert.ToDateTime(dobVal);
    int age = DateTime.Today.Year - dateOfBirth.Year;
    if (dateOfBirth > DateTime.Today.AddYears(-age))
    {
      age--;
    }

    if (age >= 18)
    {
      context.Succeed(requirement);
    }
    else
    {
      context.Fail();
    }
  }
}

Then in your ConfigureServices() function you'd wire it up

services.AddAuthorization(options =>
{
    options.AddPolicy("Over18", 
        policy => policy.Requirements.Add(new Authorization.Over18Requirement()));
});

And finally, apply it to a controller or action method with

[Authorize(Policy = "Over18")]

It seems that with ASP.NET Core 2, you can again inherit AuthorizeAttribute, you just need to also implement IAuthorizationFilter (or IAsyncAuthorizationFilter):

[AttributeUsage(AttributeTargets.Class | AttributeTargets.Method, AllowMultiple = true, Inherited = true)]
public class CustomAuthorizeAttribute : AuthorizeAttribute, IAuthorizationFilter
{
    private readonly string _someFilterParameter;

    public CustomAuthorizeAttribute(string someFilterParameter)
    {
        _someFilterParameter = someFilterParameter;
    }

    public void OnAuthorization(AuthorizationFilterContext context)
    {
        var user = context.HttpContext.User;

        if (!user.Identity.IsAuthenticated)
        {
            // it isn't needed to set unauthorized result 
            // as the base class already requires the user to be authenticated
            // this also makes redirect to a login page work properly
            // context.Result = new UnauthorizedResult();
            return;
        }

        // you can also use registered services
        var someService = context.HttpContext.RequestServices.GetService<ISomeService>();

        var isAuthorized = someService.IsUserAuthorized(user.Identity.Name, _someFilterParameter);
        if (!isAuthorized)
        {
            context.Result = new StatusCodeResult((int)System.Net.HttpStatusCode.Forbidden);
            return;
        }
    }
}

Based on Derek Greer GREAT answer, i did it with enums.

Here is an example of my code:

public enum PermissionItem
{
    User,
    Product,
    Contact,
    Review,
    Client
}

public enum PermissionAction
{
    Read,
    Create,
}


public class AuthorizeAttribute : TypeFilterAttribute
{
    public AuthorizeAttribute(PermissionItem item, PermissionAction action)
    : base(typeof(AuthorizeActionFilter))
    {
        Arguments = new object[] { item, action };
    }
}

public class AuthorizeActionFilter : IAuthorizationFilter
{
    private readonly PermissionItem _item;
    private readonly PermissionAction _action;
    public AuthorizeActionFilter(PermissionItem item, PermissionAction action)
    {
        _item = item;
        _action = action;
    }
    public void OnAuthorization(AuthorizationFilterContext context)
    {
        bool isAuthorized = MumboJumboFunction(context.HttpContext.User, _item, _action); // :)

        if (!isAuthorized)
        {
            context.Result = new ForbidResult();
        }
    }
}

public class UserController : BaseController
{
    private readonly DbContext _context;

    public UserController( DbContext context) :
        base()
    {
        _logger = logger;
    }

    [Authorize(PermissionItem.User, PermissionAction.Read)]
    public async Task<IActionResult> Index()
    {
        return View(await _context.User.ToListAsync());
    }
}

You can create your own AuthorizationHandler that will find custom attributes on your Controllers and Actions, and pass them to the HandleRequirementAsync method.

public abstract class AttributeAuthorizationHandler<TRequirement, TAttribute> : AuthorizationHandler<TRequirement> where TRequirement : IAuthorizationRequirement where TAttribute : Attribute
{
    protected override Task HandleRequirementAsync(AuthorizationHandlerContext context, TRequirement requirement)
    {
        var attributes = new List<TAttribute>();

        var action = (context.Resource as AuthorizationFilterContext)?.ActionDescriptor as ControllerActionDescriptor;
        if (action != null)
        {
            attributes.AddRange(GetAttributes(action.ControllerTypeInfo.UnderlyingSystemType));
            attributes.AddRange(GetAttributes(action.MethodInfo));
        }

        return HandleRequirementAsync(context, requirement, attributes);
    }

    protected abstract Task HandleRequirementAsync(AuthorizationHandlerContext context, TRequirement requirement, IEnumerable<TAttribute> attributes);

    private static IEnumerable<TAttribute> GetAttributes(MemberInfo memberInfo)
    {
        return memberInfo.GetCustomAttributes(typeof(TAttribute), false).Cast<TAttribute>();
    }
}

Then you can use it for any custom attributes you need on your controllers or actions. For example to add permission requirements. Just create your custom attribute.

[AttributeUsage(AttributeTargets.Class | AttributeTargets.Method, AllowMultiple = true)]
public class PermissionAttribute : AuthorizeAttribute
{
    public string Name { get; }

    public PermissionAttribute(string name) : base("Permission")
    {
        Name = name;
    }
}

Then create a Requirement to add to your Policy

public class PermissionAuthorizationRequirement : IAuthorizationRequirement
{
    //Add any custom requirement properties if you have them
}

Then create the AuthorizationHandler for your custom attribute, inheriting the AttributeAuthorizationHandler that we created earlier. It will be passed an IEnumerable for all your custom attributes in the HandleRequirementsAsync method, accumulated from your Controller and Action.

public class PermissionAuthorizationHandler : AttributeAuthorizationHandler<PermissionAuthorizationRequirement, PermissionAttribute>
{
    protected override async Task HandleRequirementAsync(AuthorizationHandlerContext context, PermissionAuthorizationRequirement requirement, IEnumerable<PermissionAttribute> attributes)
    {
        foreach (var permissionAttribute in attributes)
        {
            if (!await AuthorizeAsync(context.User, permissionAttribute.Name))
            {
                return;
            }
        }

        context.Succeed(requirement);
    }

    private Task<bool> AuthorizeAsync(ClaimsPrincipal user, string permission)
    {
        //Implement your custom user permission logic here
    }
}

And finally, in your Startup.cs ConfigureServices method, add your custom AuthorizationHandler to the services, and add your Policy.

        services.AddSingleton<IAuthorizationHandler, PermissionAuthorizationHandler>();

        services.AddAuthorization(options =>
        {
            options.AddPolicy("Permission", policyBuilder =>
            {
                policyBuilder.Requirements.Add(new PermissionAuthorizationRequirement());
            });
        });

Now you can simply decorate your Controllers and Actions with your custom attribute.

[Permission("AccessCustomers")]
public class CustomersController
{
    [Permission("AddCustomer")]
    IActionResult AddCustomer([FromBody] Customer customer)
    {
        //Add customer
    }
}

What is the current approach to make a custom AuthorizeAttribute

For pure authorization scenarios (like restricting access to specific users only), the recommended approach is to use the new authorization block: https://github.com/aspnet/MusicStore/blob/1c0aeb08bb1ebd846726232226279bbe001782e1/samples/MusicStore/Startup.cs#L84-L92

public class Startup
{
    public void ConfigureServices(IServiceCollection services)
    {
        services.Configure<AuthorizationOptions>(options =>
        {
            options.AddPolicy("ManageStore", policy => policy.RequireClaim("Action", "ManageStore"));
        });
    }
}

public class StoreController : Controller
{
    [Authorize(Policy = "ManageStore"), HttpGet]
    public async Task<IActionResult> Manage() { ... }
}

For authentication, it's best handled at the middleware level.

What are you trying to achieve exactly?

What?!

I decided to add another simple answer. B/c I find most of these answers a little overengineered. And also because I needed a way to GRANT authorization, not just DENY it. Most of the answers here offer a way to "tighten" security, but I wanted to "loosen" it. For example: "if some application setting is configured, then allow access to anonymous users".

public class MyAuthAttribute : Attribute, IAuthorizationFilter
{
    public void OnAuthorization(AuthorizationFilterContext context)
    {
        //check access 
        if (CheckPermissions())
        {
            //all good, add optional code if you want. Or don't
        }
        else
        {
            //DENIED!
            //return "ChallengeResult" to redirect to login page (for example)
            context.Result = new ChallengeResult(CookieAuthenticationDefaults.AuthenticationScheme);
        }
    }
}

That's it. No need to mess with "policies", "claims", "handlers" and other [beep]

Usage:

// GET api/Get/5
[MyAuth]
public ActionResult<string> Get(int id)
{
    return "blahblah";
}

The modern way is AuthenticationHandlers

in startup.cs add

services.AddAuthentication("BasicAuthentication").AddScheme<AuthenticationSchemeOptions, BasicAuthenticationHandler>("BasicAuthentication", null);

public class BasicAuthenticationHandler : AuthenticationHandler<AuthenticationSchemeOptions>
    {
        private readonly IUserService _userService;

        public BasicAuthenticationHandler(
            IOptionsMonitor<AuthenticationSchemeOptions> options,
            ILoggerFactory logger,
            UrlEncoder encoder,
            ISystemClock clock,
            IUserService userService)
            : base(options, logger, encoder, clock)
        {
            _userService = userService;
        }

        protected override async Task<AuthenticateResult> HandleAuthenticateAsync()
        {
            if (!Request.Headers.ContainsKey("Authorization"))
                return AuthenticateResult.Fail("Missing Authorization Header");

            User user = null;
            try
            {
                var authHeader = AuthenticationHeaderValue.Parse(Request.Headers["Authorization"]);
                var credentialBytes = Convert.FromBase64String(authHeader.Parameter);
                var credentials = Encoding.UTF8.GetString(credentialBytes).Split(new[] { ':' }, 2);
                var username = credentials[0];
                var password = credentials[1];
                user = await _userService.Authenticate(username, password);
            }
            catch
            {
                return AuthenticateResult.Fail("Invalid Authorization Header");
            }

            if (user == null)
                return AuthenticateResult.Fail("Invalid User-name or Password");

            var claims = new[] {
                new Claim(ClaimTypes.NameIdentifier, user.Id.ToString()),
                new Claim(ClaimTypes.Name, user.Username),
            };
            var identity = new ClaimsIdentity(claims, Scheme.Name);
            var principal = new ClaimsPrincipal(identity);
            var ticket = new AuthenticationTicket(principal, Scheme.Name);

            return AuthenticateResult.Success(ticket);
        }
    }

IUserService is a service that you make where you have user name and password. basically it returns a user class that you use to map your claims on.

var claims = new[] {
                new Claim(ClaimTypes.NameIdentifier, user.Id.ToString()),
                new Claim(ClaimTypes.Name, user.Username),
            }; 

Then you can query these claims and her any data you mapped, ther are quite a few, have a look at ClaimTypes class

you can use this in an extension method an get any of the mappings

public int? GetUserId()
{
   if (context.User.Identity.IsAuthenticated)
    {
       var id=context.User.FindFirst(ClaimTypes.NameIdentifier);
       if (!(id is null) && int.TryParse(id.Value, out var userId))
            return userId;
     }
      return new Nullable<int>();
 }

This new way, i think is better than the old way as shown here, both work

public class BasicAuthenticationAttribute : AuthorizationFilterAttribute
{
    public override void OnAuthorization(HttpActionContext actionContext)
    {
        if (actionContext.Request.Headers.Authorization != null)
        {
            var authToken = actionContext.Request.Headers.Authorization.Parameter;
            // decoding authToken we get decode value in 'Username:Password' format
            var decodeauthToken = System.Text.Encoding.UTF8.GetString(Convert.FromBase64String(authToken));
            // spliting decodeauthToken using ':'
            var arrUserNameandPassword = decodeauthToken.Split(':');
            // at 0th postion of array we get username and at 1st we get password
            if (IsAuthorizedUser(arrUserNameandPassword[0], arrUserNameandPassword[1]))
            {
                // setting current principle
                Thread.CurrentPrincipal = new GenericPrincipal(new GenericIdentity(arrUserNameandPassword[0]), null);
            }
            else
            {
                actionContext.Response = actionContext.Request.CreateResponse(HttpStatusCode.Unauthorized);
            }
        }
        else
        {
            actionContext.Response = actionContext.Request.CreateResponse(HttpStatusCode.Unauthorized);
        }
    }

    public static bool IsAuthorizedUser(string Username, string Password)
    {
        // In this method we can handle our database logic here...
        return Username.Equals("test") && Password == "test";
    }
}

If anyone just wants to validate a bearer token in the authorize phase using the current security practices you can,

add this to your Startup/ConfigureServices

    services.AddSingleton<IAuthorizationHandler, BearerAuthorizationHandler>();
    services.AddAuthentication(JwtBearerDefaults.AuthenticationScheme).AddJwtBearer();

    services.AddAuthorization(options => options.AddPolicy("Bearer",
        policy => policy.AddRequirements(new BearerRequirement())
        )
    );

and this in your codebase,

public class BearerRequirement : IAuthorizationRequirement
{
    public async Task<bool> IsTokenValid(SomeValidationContext context, string token)
    {
        // here you can check if the token received is valid 
        return true;
    }
}

public class BearerAuthorizationHandler : AuthorizationHandler<BearerRequirement> 
{

    public BearerAuthorizationHandler(SomeValidationContext thatYouCanInject)
    {
       ...
    }

    protected override async Task HandleRequirementAsync(AuthorizationHandlerContext context, BearerRequirement requirement)
    {
        var authFilterCtx = (Microsoft.AspNetCore.Mvc.Filters.AuthorizationFilterContext)context.Resource;
        string authHeader = authFilterCtx.HttpContext.Request.Headers["Authorization"];
        if (authHeader != null && authHeader.Contains("Bearer"))
        {
            var token = authHeader.Replace("Bearer ", string.Empty);
            if (await requirement.IsTokenValid(thatYouCanInject, token))
            {
                context.Succeed(requirement);
            }
        }
    }
}

If the code doesn't reach context.Succeed(...) it will Fail anyway (401).

And then in your controllers you can use

 [Authorize(Policy = "Bearer", AuthenticationSchemes = JwtBearerDefaults.AuthenticationScheme)]

The accepted answer (https://stackoverflow.com/a/41348219/4974715) is not realistically maintainable or suitable because "CanReadResource" is being used as a claim (but should essentially be a policy in reality, IMO). The approach at the answer is not OK in the way it was used, because if an action method requires many different claims setups, then with that answer you would have to repeatedly write something like...

[ClaimRequirement(MyClaimTypes.Permission, "CanReadResource")] 
[ClaimRequirement(MyClaimTypes.AnotherPermision, "AnotherClaimVaue")]
//and etc. on a single action.

So, imagine how much coding that would take. Ideally, "CanReadResource" is supposed to be a policy that uses many claims to determine if a user can read a resource.

What I do is I create my policies as an enumeration and then loop through and set up the requirements like thus...

services.AddAuthorization(authorizationOptions =>
        {
            foreach (var policyString in Enum.GetNames(typeof(Enumerations.Security.Policy)))
            {
                authorizationOptions.AddPolicy(
                    policyString,
                    authorizationPolicyBuilder => authorizationPolicyBuilder.Requirements.Add(new DefaultAuthorizationRequirement((Enumerations.Security.Policy)Enum.Parse(typeof(Enumerations.Security.Policy), policyWrtString), DateTime.UtcNow)));

      /* Note that thisn does not stop you from 
          configuring policies directly against a username, claims, roles, etc. You can do the usual.
     */
            }
        }); 

The DefaultAuthorizationRequirement class looks like...

public class DefaultAuthorizationRequirement : IAuthorizationRequirement
{
    public Enumerations.Security.Policy Policy {get; set;} //This is a mere enumeration whose code is not shown.
    public DateTime DateTimeOfSetup {get; set;} //Just in case you have to know when the app started up. And you may want to log out a user if their profile was modified after this date-time, etc.
}

public class DefaultAuthorizationHandler : AuthorizationHandler<DefaultAuthorizationRequirement>
{
    private IAServiceToUse _aServiceToUse;

    public DefaultAuthorizationHandler(
        IAServiceToUse aServiceToUse
        )
    {
        _aServiceToUse = aServiceToUse;
    }

    protected async override Task HandleRequirementAsync(AuthorizationHandlerContext context, DefaultAuthorizationRequirement requirement)
    {
        /*Here, you can quickly check a data source or Web API or etc. 
           to know the latest date-time of the user's profile modification...
        */
        if (_aServiceToUse.GetDateTimeOfLatestUserProfileModication > requirement.DateTimeOfSetup)
        {
            context.Fail(); /*Because any modifications to user information, 
            e.g. if the user used another browser or if by Admin modification, 
            the claims of the user in this session cannot be guaranteed to be reliable.
            */
            return;
        }

        bool shouldSucceed = false; //This should first be false, because context.Succeed(...) has to only be called if the requirement specifically succeeds.

        bool shouldFail = false; /*This should first be false, because context.Fail() 
        doesn't have to be called if there's no security breach.
        */

        // You can do anything.
        await doAnythingAsync();

       /*You can get the user's claims... 
          ALSO, note that if you have a way to priorly map users or users with certain claims 
          to particular policies, add those policies as claims of the user for the sake of ease. 
          BUT policies that require dynamic code (e.g. checking for age range) would have to be 
          coded in the switch-case below to determine stuff.
       */

        var claims = context.User.Claims;

        // You can, of course, get the policy that was hit...
        var policy = requirement.Policy

        //You can use a switch case to determine what policy to deal with here...
        switch (policy)
        {
            case Enumerations.Security.Policy.CanReadResource:
                 /*Do stuff with the claims and change the 
                     value of shouldSucceed and/or shouldFail.
                */
                 break;
            case Enumerations.Security.Policy.AnotherPolicy:
                 /*Do stuff with the claims and change the 
                    value of shouldSucceed and/or shouldFail.
                 */
                 break;
                // Other policies too.

            default:
                 throw new NotImplementedException();
        }

        /* Note that the following conditions are 
            so because failure and success in a requirement handler 
            are not mutually exclusive. They demand certainty.
        */

        if (shouldFail)
        {
            context.Fail(); /*Check the docs on this method to 
            see its implications.
            */
        }                

        if (shouldSucceed)
        {
            context.Succeed(requirement); 
        } 
     }
}

Note that the code above can also enable pre-mapping of a user to a policy in your data store. So, when composing claims for the user, you basically retrieve the policies that had been pre-mapped to the user directly or indirectly (e.g. because the user has a certain claim value and that claim value had been identified and mapped to a policy, such that it provides automatic mapping for users who have that claim value too), and enlist the policies as claims, such that in the authorization handler, you can simply check if the user's claims contain requirement.Policy as a Value of a Claim item in their claims. That is for a static way of satisfying a policy requirement, e.g. "First name" requirement is quite static in nature. So, for the example above (which I had forgotten to give example on Authorize attribute in my earlier updates to this answer), using the policy with Authorize attribute is like as follows, where ViewRecord is an enum member:

[Authorize(Policy = nameof(Enumerations.Security.Policy.ViewRecord))] 

A dynamic requirement can be about checking age range, etc. and policies that use such requirements cannot be pre-mapped to users.

An example of dynamic policy claims checking (e.g. to check if a user is above 18 years old) is already at the answer given by @blowdart (https://stackoverflow.com/a/31465227/4974715).

PS: I typed this on my phone. Pardon any typos and lack of formatting.

The below code worked for me in .Net Core 5

[AttributeUsage(AttributeTargets.Class | AttributeTargets.Method)]
public class AccessAuthorizationAttribute : AuthorizeAttribute, IAuthorizationFilter
{
    public string Module { get; set; } //Permission string to get from controller

    public AccessAuthorizationAttribute(string module)
    {
        Module = module;
    }
    public void OnAuthorization(AuthorizationFilterContext context)
    {
        //Validate if any permissions are passed when using attribute at controller or action level

        if (string.IsNullOrEmpty(Module))
        {
            //Validation cannot take place without any permissions so returning unauthorized
            context.Result = new UnauthorizedResult();
            return;
        }
       
        if (hasAccess)
        {
            return;
        }

        context.Result = new UnauthorizedResult();
        return;
    }
}

As of this writing I believe this can be accomplished with the IClaimsTransformation interface in asp.net core 2 and above. I just implemented a proof of concept which is sharable enough to post here.

public class PrivilegesToClaimsTransformer : IClaimsTransformation
{
    private readonly IPrivilegeProvider privilegeProvider;
    public const string DidItClaim = "http://foo.bar/privileges/resolved";

    public PrivilegesToClaimsTransformer(IPrivilegeProvider privilegeProvider)
    {
        this.privilegeProvider = privilegeProvider;
    }

    public async Task<ClaimsPrincipal> TransformAsync(ClaimsPrincipal principal)
    {
        if (principal.Identity is ClaimsIdentity claimer)
        {
            if (claimer.HasClaim(DidItClaim, bool.TrueString))
            {
                return principal;
            }

            var privileges = await this.privilegeProvider.GetPrivileges( ... );
            claimer.AddClaim(new Claim(DidItClaim, bool.TrueString));

            foreach (var privilegeAsRole in privileges)
            {
                claimer.AddClaim(new Claim(ClaimTypes.Role /*"http://schemas.microsoft.com/ws/2008/06/identity/claims/role" */, privilegeAsRole));
            }
        }

        return principal;
    }
}

To use this in your Controller just add an appropriate [Authorize(Roles="whatever")] to your methods.

[HttpGet]
[Route("poc")]
[Authorize(Roles = "plugh,blast")]
public JsonResult PocAuthorization()
{
    var result = Json(new
    {
        when = DateTime.UtcNow,
    });

    result.StatusCode = (int)HttpStatusCode.OK;

    return result;
}

In our case every request includes an Authorization header that is a JWT. This is the prototype and I believe we will do something super close to this in our production system next week.

Future voters, consider the date of writing when you vote. As of today, this works on my machine.™ You will probably want more error handling and logging on your implementation.

Just adding to the great answer from @Shawn. If you are using dotnet 5 you need to update the class to be:

public abstract class AttributeAuthorizationHandler<TRequirement, TAttribute> : AuthorizationHandler<TRequirement> where TRequirement : IAuthorizationRequirement where TAttribute : Attribute
{
    protected override Task HandleRequirementAsync(AuthorizationHandlerContext context, TRequirement requirement)
    {
        var attributes = new List<TAttribute>();
        
        if (context.Resource is HttpContext httpContext)
        {
            var endPoint = httpContext.GetEndpoint();

            var action = endPoint?.Metadata.GetMetadata<ControllerActionDescriptor>();

            if(action != null)
            {
                attributes.AddRange(GetAttributes(action.ControllerTypeInfo.UnderlyingSystemType));
                attributes.AddRange(GetAttributes(action.MethodInfo));
            }
        }
        
        return HandleRequirementAsync(context, requirement, attributes);
    }

    protected abstract Task HandleRequirementAsync(AuthorizationHandlerContext context, TRequirement requirement, IEnumerable<TAttribute> attributes);

    private static IEnumerable<TAttribute> GetAttributes(MemberInfo memberInfo) => memberInfo.GetCustomAttributes(typeof(TAttribute), false).Cast<TAttribute>();
}

Noting the way getting the ControllerActionDescriptor has changed.

I have bearer token and I can read claims. I use that attribute on controllers and actions

public class CustomAuthorizationAttribute : ActionFilterAttribute
{
    public string[] Claims;

    public override void OnActionExecuting(ActionExecutingContext context)
    {
        // check user 
        var contextUser = context?.HttpContext?.User;
        if (contextUser == null)
        {
            throw new BusinessException("Forbidden");
        }


        // check roles
        var roles = contextUser.FindAll("http://schemas.microsoft.com/ws/2008/06/identity/claims/role").Select(c => c.Value).ToList();
        if (!roles.Any(s => Claims.Contains(s)))
        {
            throw new BusinessException("Forbidden");
        }

        base.OnActionExecuting(context);
    }
}

example

[CustomAuthorization(Claims = new string[]
    {
        nameof(AuthorizationRole.HR_ADMIN),
        nameof(AuthorizationRole.HR_SETTING)
    })]
[Route("api/[controller]")]
[ApiController]
public class SomeAdminController : ControllerBase
{
    private readonly IMediator _mediator;

    public SomeAdminController(IMediator mediator)
    {
        _mediator = mediator;
    }

    [HttpGet("list/SomeList")]
    public async Task<IActionResult> SomeList()
        => Ok(await _mediator.Send(new SomeListQuery()));
}

That is Roles

public struct AuthorizationRole
{
    public static string HR_ADMIN;
    public static string HR_SETTING;
}

A lot of people here already told this, but with Policy handlers you can come really far in terms of what you could achieve with the old way in .NET Framework.

I followed a quick writeup from this answer on SO: https://stackoverflow.com/a/61963465/7081176 For me it works flawlessly after making some classes:

The EditUserRequirement:

public class EditUserRequirement : IAuthorizationRequirement
{
    public EditUserRequirement()
    {
    }
}

An abstract handler to make my life easier:

public abstract class AbstractRequirementHandler<T> : IAuthorizationHandler
    where T : IAuthorizationRequirement
{
    public async Task HandleAsync(AuthorizationHandlerContext context)
    {
        var pendingRequirements = context.PendingRequirements.ToList();
        foreach (var requirement in pendingRequirements)
        {
            if (requirement is T typedRequirement)
            {
                await HandleRequirementAsync(context, typedRequirement);
            }
        }
    }

    protected abstract Task HandleRequirementAsync(AuthorizationHandlerContext context, T requirement);
}

An implementation of the abstract handler:

public class EditUserRequirementHandler : AbstractRequirementHandler<EditUserRequirement>
{
    protected override Task HandleRequirementAsync(AuthorizationHandlerContext context, EditUserRequirement requirement)
    {
        // If the user is owner of the resource, allow it.
        if (IsOwner(context.User, g))
        {
            context.Succeed(requirement);
        }

        return Task.CompletedTask;
    }

    private static bool IsOwner(ClaimsPrincipal user, Guid userIdentifier)
    {
        return user.GetUserIdentifier() == userIdentifier;
    }
}

Registering my handler and requirement: services.AddSingleton();

        services.AddAuthorization(options =>
        {
            options.AddPolicy(Policies.Policies.EditUser, policy =>
            {
                policy.Requirements.Add(new EditUserRequirement());
            });
        });

And then using my Policy in Blazor:

<AuthorizeView Policy="@Policies.EditUser" Resource="@id">
    <NotAuthorized>
        <Unauthorized />
    </NotAuthorized>
    <Authorized Context="Auth">
        ...
    </Authorized>
</AuthorizeView>

I hope this is useful for anyone facing this issue.

I have been looking into solving a very similar issue, and settled on creating a custom ActionFilterAttribute (I'm going to call it AuthorizationFilterAttribute) instead of an AuthorizeAttribute to implement the guidance here: https://docs.microsoft.com/en-us/aspnet/core/security/authorization/resourcebased?view=aspnetcore-6.0#challenge-and-forbid-with-an-operational-resource-handler.

For authorization in our app. We had to call a service based on the parameters passed in authorization attribute.

For example, if we want to check if logged in doctor can view patient appointments we will pass "View_Appointment" to custom authorize attribute and check that right in DB service and based on results we will athorize. Here is the code for this scenario:

    public class PatientAuthorizeAttribute : TypeFilterAttribute
    {
    public PatientAuthorizeAttribute(params PatientAccessRights[] right) : base(typeof(AuthFilter)) //PatientAccessRights is an enum
    {
        Arguments = new object[] { right };
    }

    private class AuthFilter : IActionFilter
    {
        PatientAccessRights[] right;

        IAuthService authService;

        public AuthFilter(IAuthService authService, PatientAccessRights[] right)
        {
            this.right = right;
            this.authService = authService;
        }

        public void OnActionExecuted(ActionExecutedContext context)
        {
        }

        public void OnActionExecuting(ActionExecutingContext context)
        {
            var allparameters = context.ActionArguments.Values;
            if (allparameters.Count() == 1)
            {
                var param = allparameters.First();
                if (typeof(IPatientRequest).IsAssignableFrom(param.GetType()))
                {
                    IPatientRequest patientRequestInfo = (IPatientRequest)param;
                    PatientAccessRequest userAccessRequest = new PatientAccessRequest();
                    userAccessRequest.Rights = right;
                    userAccessRequest.MemberID = patientRequestInfo.PatientID;
                    var result = authService.CheckUserPatientAccess(userAccessRequest).Result; //this calls DB service to check from DB
                    if (result.Status == ReturnType.Failure)
                    {
                        //TODO: return apirepsonse
                        context.Result = new StatusCodeResult((int)System.Net.HttpStatusCode.Forbidden);
                    }
                }
                else
                {
                    throw new AppSystemException("PatientAuthorizeAttribute not supported");
                }
            }
            else
            {
                throw new AppSystemException("PatientAuthorizeAttribute not supported");
            }
        }
    }
}

And on API action we use it like this:

    [PatientAuthorize(PatientAccessRights.PATIENT_VIEW_APPOINTMENTS)] //this is enum, we can pass multiple
    [HttpPost]
    public SomeReturnType ViewAppointments()
    {

    }