We are trying to embed Autodesk Forge Viewer with access to BIM 360 Team data in our existing app. We are having trouble with executing the 3-Legged authorisation workflow inside an iframe. Our justification for iframe instead of pop-up (window.open()) or redirecting current window (location.assign()) is following:
Redirecting current app window to authorization would disrupt the app workflow. Authorisation on app start is not an option. Viewer could be required at any point in existing workflow or not at all and authorisation should be triggered only just before the Viewer is required.
Opening the authorisation in new window (pop-up) is disabled by default in many (desktop) browsers. Requiring the user to enable pop-ups is an option though definitely not user-friendly. Furthermore, multi-window browsing in mobile environments is not always implemented (e.g. in in-app browsers), making this not-an-option.
Authorisation in an iframe would overcome these shortcomings while providing easy, unobtrusive and consistent user experience.
We are following the workflow as described in here and here. The login page (2nd link > step 1 > 1st picture) displays fine in the iframe and the user is able to enter his/her Autodesk ID credentials. After that the user is redirected to the OAuth consent page. The consent page (2nd link > step 1 > 2nd picture) is served with header X-Frame-Options: SAMEORIGIN which in an iframe obviously fails to load: Refused to display 'https://auth.autodesk.com/as/NH3Mc/resume/as/authorization.ping?opentoken=...' in a frame because it set 'X-Frame-Options' to 'deny'. However, if the user has consented beforehand (e.g. in a separate window) the consent page is skipped and user is immediately redirected to the provided callback uri along with the authorisation code as desired (response code 302 Found, header Location: callback uri + code).
This raises the question if the X-Frame-Options: SAMEORIGIN is by-design and deliberate? What is the benefit in this scenario to disable OAuth consent page in iframe while allowing login page? Would it somehow be possible to disable the header or add an exception for the origin defined in the Autodesk Forge App as Callback URL?
Are we missing any other options to provide unobtrusive and consistent authorisation that would work on desktop as well as mobile platforms?
Related
PWA Freezing after OpenID authorizes user [iOS Safari Standalone]
I have built a simple PWA for our security staff that allows employees to click links and view content about our company's policies on various matters. The app uses Microsoft's OWIN middleware library to authorize our employees' access into the app via their enterprise Microsoft login creds. When a user clicks the 'Employee Sign in' prompt on our login page, they are redirected to Microsoft's domain to complete the sign-in process. Once sign in is complete, they are redirected back to our app's home page.
The Problem
The problem appears only to arise when iOS users (v13) pin the app to their homescreen and then launch the app in standalone mode, and only after the user has completely terminated the app and then returned. We've tested the app on Chrome, Safari (non-standalone mode), Firefox, and the issue does not present in those browswers. The app functions seamlessly up until the point that the user has completed their Microsoft signin and been redirected back to the home page. At this point, if a user clicks a link to another page (within the app), the app completely locks up, doesn't respond to further button clicks, and doesn't load the page prompted by the user. No errors are thrown in the console.
What we've found immediately kicks everything back into gear is if the user switches to another app (even just for a second) and then switches back to our PWA when it's locked up. At this point, the page that the user attempted to navigate to loads immediately without further prompting and the app works 100% seamlessly after this point. It's only the initial version of the default page that freezes.
Potential Causes
My current working theory is that the problem is being caused by some combination of the following:
Redirection to Microsoft's sign in portal. When the user is sent to Microsoft for auth and then sent back to our domain, there could be issues with session/cookie continuity.
iOS's standalone mode. In conjunction with the above, is it possible that using third-party authentication and briefly leaving the domain of the PWA is causing problems with future page navigation. This is supported by the idea that no other browsers or devices have this issue, and my research suggests that Apple support for PWAs is still in its early stages.
Service worker failure. We have done significant testing to ensure that a service worker is being properly installed and registered when a user first enters the site. We have checks to re-register the SW just in case it is dropped at any point in page navigation. We are confident that at the time a user is redirected back to our home page after authentication that there is an active service worker that handles page GET requests. I have also tested explicitly caching the linked pages accessible from our home page during the service worker's registration to see if serving the page from the cache would alleviate the issue. It did not. This is the code in sw.js that handles fetch requests (taken from Google's handy guide):
// "cache-first" approach for requests from client. Will try to get the file from the cache.
// If no match found, it will send the request onto the network. If both fail serve fallback page.
self.addEventListener("fetch", function (event) {
if (event.request.method !== "GET") return;
event.respondWith(
// Try the cache
caches.match(event.request).then(function (response) {
console.log("[service worker] attempting to fetch file from cache...");
return response || fetch(event.request);
}).catch(function () {
// If both fail, show a generic fallback:
return caches.match(offlineFallbackPage);
})
);
});
I have remotely debugged the PWA in standalone using a Mac, and what I have verified is that the click event that fires when a user clicks a link to navigate to a new page IS being properly handled, so the problem truly appears to lie in the loading of the linked pages themselves. Beyond that, debugging remotely has confirmed that there are no HTTP GET errors (or any other errors) firing at all when attempting to navigate to other pages on the site.
This is the first PWA I've ever built and I'm a novice with all this stuff. So I'd love to know if I'm missing anything or where I can go from here. I've scoured all the forums and can't seem to find answers anywhere. Thanks!
I had a very similar problem in my very specific case. but my pwa (packaged with PwaBuilder) froze on oidc signout, when redirect to applications home url.
In XCode I observed an error stating:
could not signal service com.apple.webkit.webcontent 113 could not find specified service
The problem did not occur with my Identity Provider redirect back, but with the following redirect which initiated the OIDC client library which I am using oidc-client-ts. It turns out that there are two possible ways to set the location/url of a window, assign or setting href. And the library uses assign by default. Changing assign to replace href lead to my iOS PWA not to freeze anymore. Very specific use case but it might help somebody else...
auth.signoutRedirect({
post_logout_redirect_uri: process.env.BASE_URI,
redirectMethod: "replace",
});
I have a website that is iframed into a 3rd party webpage, which is itself embedded in a WkWebView in an iOS app. Mobile Safari and the WkWebView reject the session (http-only) cookies being sent for my website, breaking basically everything. The work-around in mobile Safari is to either enable all cookies in settings (yuck) or instruct users to visit my site directly (so it counts as a "site I visited", as far as Safari is concerned); neither of these is particularly palatable. I have found no work-arounds for the WkWebView.
This question is two-part:
1) I am thinking of implementing a redirect service in my website, that takes a destination URL as a parameter, and simply redirects the user to that URL on page load. The 3rd party site can then link to my redirect page with the URL set to send users right back, with the hope that this will count as "visiting" my domain, enabling cookies to be loaded.
Alternatively, the 3rd party site could open a new tab to my site, that closes immediately on load. I expect that this would be a less optimal user experience, however, and so would prefer not to go this route.
Best of all would be for the "POST to a hidden iframe" trick (3rd party page POSTs to my domain in a hidden iframe), but as SO questions indicate that trick no longer works.
Are either of these viable solutions, or has Apple blocked these methods of getting the session cookies set as well? Is there a better solution that I have not considered?
2) Is there a way to set the cookie acceptance policy with WkWebViews like could be done with UiWebViews? My searches of StackOverflow suggest not, but the answers I read could be based on older versions of iOS (the app requires iOS 9+).
If there is no app-code solution for WkWebViews, would the solutions for mobile Safari also work with WkWebViews?
I just had a similar issue. I have a WkWebView which loads my web app that has an iframe loading a login screen from a specific server. The login page would complain that the iframe did not allow cookies.
When I would load the login page directly in the web view, it would work and it would also curiously start working as well when I tested it afterwards again inside the iframe.
The best explanation I found for this is, cookies are only allowed to be saved in the iframe if the web view has directly loaded the domain of the iframe at least once. Knowing this, I was able to implement a workaround.
By simply pinging the login page once with the webview, I use the WKNavigationDelegate to wait until I start receiving some data from the server. Once this happens, I make the web view load my page that contains the iframe. Now the iframe is able to consistently load the login screen.
So I'm weighing up the pros/cons of the different auth mechanisims for a SalesForce Canvas application.
First, a little background. I've got an existing web app - let's call it "myapp". I setup a VisualForce tab containing a VisualForce page which embeds my canvas app by the following approach:
<apex:page>
<apex:canvasApp applicationName="___________" />
</apex:page>
I first implemented access via the signed request method, which was great, as the user only had to accept my app's permissions on first access, and subsequent attempts to access my canvas app could jump right in to myapp homepage.
From reading this article about packaging canvas apps for different SalesForce editions, signed request method has a limitation:
Most typical Group and Professional Edition customers are not going tobe able to use your Signed Request Connected App unless they upgrade to EE or higher or purchase these features as an add-on.
So I decided to switch to the oauth workflows.
Looking at the user agent oauth flow documentation, it says:
The user must always approve access for this authentication flow. After approving access, the application receives the callback from Salesforce.
This is also not desirable, however the web server oauth flow does not have that requirement - once a user accepts the app's permission requirements, they don't need to be prompted to do that again. It also makes things like adding a "Login with SalesForce" option on myapp's login page super easy to add in.
So I setup the web server oauth flow, and have everything functioning well, and as an added benefit added a "Login with SalesForce" option to my login page - great.
Next step was to set the canvas app in the VisualForce tab up to kick off the web server oauth flow.
The problem I want to solve:
I want to use the web server oauth flow to gain access to the homepage of myapp from that VisualForce tab, with as few user interactions as possible in the process.
Base setup
To initiate the oauth flow, I've setup a URL with the following logic in there to make it simple to craft links that kick off the process from elsewhere. This is used from each of the below examples, and is referenced by links to /salesforce/oauth/......
// Allow links to this page to specify "state" and "prompt" paramaters.
$state = isset($_GET['state']) ? $_GET['state'] : 'login';
$prompt = isset($_GET['prompt']) ? $_GET['prompt'] : '';
// Canvas app contextual information provides the right SalesForce endpoint domain, so provide a way for that to be passed in here, or fallback to standard login.salesforce.com for other workflows.
$authDomain = isset($_GET['authDomain']) ? $_GET['authDomain'] : 'https://login.salesforce.com/services/oauth2/authorize';
$url = $authDomain.'?'.http_build_query(array(
'response_type' => 'code',
'client_id' => 'XXXXXXX_MY_APP_CLIENT_ID',
'redirect_uri' => 'https://'.$_SERVER['HTTP_HOST'].'/salesforce/authorize/',
'state' => $state,
'prompt' => $prompt,
));
header('Location: '.$url);
die;
Failed attempt 1
Using the javascript canvas sdk, redirect the canvas app to initiate my web server oauth flow:
location.href = '/salesforce/oauth?authUrl='+encodeURIComponent(Sfdc.canvas.oauth.loginUrl())+'&state=canvas';
I ran into two problems with this approach:
The canvas app looses the contextual information provided by the #query fragment in it's URL.
X-Frame-Options header of the SalesForce page which has the accept/decline permissions prevent it from displaying within an iframe, even on SalesForce domains.
I feel this would be the best way to accomplish my goal if these issues could be overcome.
I actually did an experiment which resolves issue #1 by loading that URL in another iframe, within my canvas app, and if I'd already accepted myapps permission requirements that worked flawlessly, however I was still stuck with issue #2 when the permissions screen popped up and the whole process failed to complete.
Current solution
I've made the initial canvas app page include a button to click, which opens the web server oauth flow in a new window where it completes successfully. Upon completion, the canvas app iframe is redirected to my app homepage.
I'm unhappy with this, because every time I click on my VisualForce tab, there's a step in there requiring the user to click a button, and a new window pops up to run the oauth workflow. It closes itself automatically with no extra user interaction if the user had already accepted the app permissions in the past, or prompts them to accept if they hadn't yet, and then closes itself.
If I'm stuck with this solution it's not the end of the world - we'll make that button into a splash screen with some marketing crap in there and a big obnoxious "Continue to MyApp" button somewhere.
The question (...finally)
Is there some way I can remove that necessary step of clicking the button every single time the canvas app is loaded, but continue using the web server oauth flow here? (remembering I don't want to use the user agent oauth flow because that has a similar requirement of accepting the permissions every time the user accesses it anyway).
That extra step is particularly annoying when the canvas app is being embedded within a SalesForce account or contact screen - as it stops my app from loading and displaying data to my users until the user clicks on the button.
Either I don't get it, or that's really simple. You need to use oAuth immediate parameter:
immediate — Determines whether the user should be prompted for login and approval. This parameter is optional. The value must be true or false if specified. Default value is false. Note the following:
If set to true, and if the user is currently logged in and has previously approved the client_id, Salesforce skips the approval step.
If set to true and the user is not logged in or has not previously approved the client, Salesforce immediately terminates with the immediate_unsuccessful error code.
Source: 1, 2
The only catch is the last part: you will receive error if user hasn't authorized your app yet. In my experience with oAuth it's really easier to use Javascript to run your requests from the client's browser itself. You can fire off immediate = true request, and right after it immediate = false in case first one failed. Then you send the access_token to your app by creating third request – to your own server.
My organization had a web app that worked perfectly in iOS 6. You'd visit the website, the website would tell you to add the page to your homescreen, and boom, a nice HTML5 web app was added to the home screen.
Because we're processing sensitive data, the web app used HTTP authentication (via the native WebKit auth dialog) to authenticate user/passes. It worked without a hitch until iOS 7. Now when someone tries to summon the HTTP auth dialog, nothing happens. It's clearly trying to load something, as the spinner in the status bar appears, but no dialog ever pops up, essentially breaking the "app."
Has anyone else run into this? Is this something you'd consider to be a bug on Apple's end? Any workaround?
My company ran into this last fall, starting with iOS 6, and what we have been able to ascertain is that it is a genuine Apple Safari bug as part of its security "enhancements". No real explanation from them for rationale, but here is what we see in the debug and packet sniffers.
In normal operation, the Safari browser will request a page (or an object in the page) from the server on a GET. If that asset is protected with an Access Control List, in our case Apache Basic Auth, and it is the first request on that host in the session, the server will respond with a 401 HTTP response header indicating to the client (the browser) that it needs to request again, this time adding a basic auth header that has authorization credentials. The browser then presents a login dialog to the user, where they can enter user and pass credentials, and either submit or cancel the request. On submit, the client re-requests with those credentials in the auth header.
Assuming the credentials are accepted on the second GET request, the proper asset will be returned on the response, and the document in the browser will proceed with loading the rest of the page (assuming it was a page you requested). If you have embedded assets that reside on a different host, and that host requires authentication for that asset, the process is repeated as the page loads.
Here's where it gets broken. If you embed calls to objects from more than 2 hosts total on the same page, which require basic authentication, the 3rd authentication prompt on that page is suppressed, so the browser spins forever waiting for you to enter credentials on a prompt that you never see. Your Safari browser is now hung up on that stalled authentication prompt, on this and any other tab, even on a reload, and you will not get another prompt unless and until you hard-close your browser or restart your device.
This does not affect Chrome, just Safari, and it is both on an iPhone and an iPad with iOS 6 or later. I have the latest iOS version as of this writing (7.0.6), and the problem is still there.
We had a workaround last year, where we would create an internal page that had an array of each of the embedded hosts, which we would then loop through with an iframe embedding a call to the favicon.ico at that host's location. That worked until recently, where now, perhaps because of the iOS 7 feature of freezing background tabs, the auth prompts are frozen up again.
Here was the JavaScript sample:
hosts=["store","profile","www","secure-store","images","m","modules"];
devhost=location.hostname;
var i=0;
while (hosts[i])
{
newhost=devhost.replace('store.mydomain',hosts[i]+'.mydomain');
document.write("<iframe Xhidden seamless=seamless width=0 height=0 src=http://"+newhost+"/favicon.ico><img height='16' width='20' alt='NOT' title='NOT AUTHENTICATED' src=http://"+newhost+"/favicon.ico> Authenticated on "+newhost+"</a></br></iframe>");
document.write("<img height='16' width='20' alt='NOT' title='NOT AUTHENTICATED' src="+(newhost.indexOf('secure')>0?'https://':'http://')+newhost+"/favicon.ico> Authenticated on "+newhost+"</a></br>");
i++;
}
The second set in the document.write would give a visual indication of which hosts have been authenticated, as their favicon is now displayed. It also lets you know which host might be stalled, as its icon is missing.
Since this workaround stopped working on iOS 7, the only cumbersome solution we have is to pre-open a separate tab for each of the favicons (directly in the URL), enter the auth, go back, go to the next one in the list, and repeat until you have cached all of the auth credentials for all of the hosts used on the page. At that point, you can load the original page since your creds are now cached. Cruddy, and completely unreasonable for an end consumer, but is what we need to do for testing sites that are behind a public CDN, as we need to protect assets on that development site with an ACL.
As of today, we are still figuring out a better workaround. Not an issue on Android, Windows, or any other iOS.
Sure worked better when Jobs was alive.
Hope some of this helps.
I have the exact same problem. Basic authentication worked with previous iOS versions but not with iOS 7 in combination with web apps added to the home screen. I think this may be related to the dialog problem described here.
Standard dialogs are not working at all, such as alert, confirm or prompt.
The login prompt that is shown to authenticate the user is probably blocked (does not work or is not visible) and that is why the web app does not pass through the authentication phase.
I suppose Apple will have to fix this bug in a future release.
Edit: After upgrading to iOS 7.0.3 basic authentication suddenly started to work again also in home screen web app mode. Login prompt is displayed and everything works as expected.
I have a BlackBerry java-based app that works with salesforce data and uses oauth 2 for authentication. It has been working fine until recently when users have been complaining that once they get to the accept/deny salesforce oauth page, clicking on Accept button displays a salesforce 404 page.
I attached a debugger to the bb simulator (OS7) and found that everything was working fine up to the point where a POST is made to salesforce on pressing the ACCEPT button in the final stage of authentication/authorisation via salesforce oauth2 user agent flow. At this point, the post would succeed and the response body would contain some markup with a small JavaScript block that just performs a document.location.href assignment to force the browser to navigate to a new URL. I noticed that the URL was not what I was expecting - I was expecting to see the redirect_uri callback address + the access token etc data appended to the end of it (as per the sf oauth2 documentation). What I saw instead was a URL that when navigated to leads the user back to a login page with a message saying that they are attempting to access a resource that requires authentication.
If I perform the same flow in say chrome the URL after the post is what I expect. This all used to work just fine so I am confused as to why it would stop working all of a sudden. I've heard that Salesforce changed their oauth 2 implementation back in late 2011 slightly so I wonder if this has broken my client although I don't see how.
As an experiment I removed all of my code except for an embedded BrowserField and made it navigate to the oauth URL as constructed by combining consumer key and various other values as per the documentation and it still fails with a 404 page at the point of clicking on the accept button after a successful login.
If I try the same bootstrap oauth URL in the built in bb browser the full authentication flow works fine and I am eventually redirected to the final callback URL with access token information appended.
Has anyone else out there had trouble with oauth2 user agent flow on BrowserField (from field2 package)?
Have you tried using the "scope=mobile" parameter? This parameter puts you into a very simplified oauth flow that is optimized for blackberry...we've found they generally have trouble getting through the richer flows designed for iOS/Android. That said, I haven't heard of them having issues on BB7...only the 5 and sometimes 6 series
If that fails, could you open a support case with us?
Try also "display=mobile" and "scope=api web" (with a space), depending on what you are trying to access after OAuth.