Description of the false positive
- ASP.NET (and ASP.NET Core) projects can have two sets of endpoint handlers (controller actions) which separately handle...
- Requests originating from web-browsers, which are authenticated using browser cookies or browser-managed HTTP Basic/Digest authentication; including XHR/
fetch-based requests, as well as ordinary document navigation. These are the kinds of requests that are vulnerable to CSRF attacks and so should use a CSRF validation token or other approach.
- Requests originating from non-browser-based clients (e.g. daemon processes; cron jobs running curl, etc); these are authenticated using HTTP
Authorization header (e.g. Bearer tokens). It is not possible for a CSRF attack to succeed in this case (see https://security.stackexchange.com/questions/170388/do-i-need-csrf-token-if-im-using-bearer-jwt ).
Assuming that this code is the actual CodeQL analysis rule for this alert (CWE-352/MissingAntiForgeryTokenValidation.ql), then the problem is...
- The rule is only activated if the project uses CSRF at least once, anywhere (see the comment where it says "Verify that validate anti forgery token attributes are used somewhere within this project").
- So it assumes that if at least one controller-action in a project uses CSRF, then all controller-actions in the same project should also use CSRF...
- This assumption is incorrect: as mentioned above, it's possible for a project to serve both browser-based requests and non-browser requests - with entirely different authentication schemes and policies such that non-browser-based endpoint-actions cannot be invoked in a browser-based CSRF scenario.
Code samples or links to source code
If the two controller-classes are built in a single project, then the fact BrowserAjaxController uses [ValidateAntiForgeryToken] will cause MissingAntiForgeryTokenValidation.q to think that WebServiceController should also use [ValidateAntiForgeryToken] even though it doesn't use browser-cookies based authentication (due to the different Scheme value).
class BrowserAjaxController : Controller
{
[HttpPost("/ajax/exec-rm-rf-root" )]
[Authorize( AuthenticationSchemes = MySchemeNames.BrowserCookiesScheme, Policy = "SomePolicy1" )]
[ValidateAntiForgeryToken]
public IActionResult DoTheThing()
{
return this.Ok();
}
}
class WebServiceController : Controller
{
[HttpPost("/api/arbitrary-operation" )]
[Authorize( AuthenticationSchemes = MySchemeNames.BearerTokenScheme, Policy = "SomePolicy2" )]
public IActionResult DoTheOtherThing()
{
return this.Ok();
}
}
Description of the false positive
fetch-based requests, as well as ordinary document navigation. These are the kinds of requests that are vulnerable to CSRF attacks and so should use a CSRF validation token or other approach.Authorizationheader (e.g. Bearer tokens). It is not possible for a CSRF attack to succeed in this case (see https://security.stackexchange.com/questions/170388/do-i-need-csrf-token-if-im-using-bearer-jwt ).Assuming that this code is the actual CodeQL analysis rule for this alert (
CWE-352/MissingAntiForgeryTokenValidation.ql), then the problem is...Code samples or links to source code
If the two controller-classes are built in a single project, then the fact
BrowserAjaxControlleruses[ValidateAntiForgeryToken]will causeMissingAntiForgeryTokenValidation.qto think thatWebServiceControllershould also use[ValidateAntiForgeryToken]even though it doesn't use browser-cookies based authentication (due to the different Scheme value).