Related
I have a running website, where users already have accounts. And I am trying to create a Google Assistant agent, accessible on Android, to help users access their information.
My issue is that I can't detect returning users on Android Smartphones, each time they have to sign in.
I tried Anonymous User Identity, but it is soon to be deprecated.
Is there an other way to keep track of users?Using some kind of userId that I can store, so I can make "my own Acount Linking" linking the person/Smartphone with already existing user accounts.
There are a few angles to your question.
Is there any way to keep track of users?
Yes... but...
You can store a userId that you generate in the user storage area. You do need to treat this like you would a cookie, so some jurisdictions might impose restrictions on this, but this is one approach to moving from the anonymous ID that is being turned off soon.
But...
How do I let them log into my service through the Action?
That is the problem. The General Policies states the following limitation for collecting user data:
Authentication Data
(including passwords, PINs, and answers to security questions)
Don't collect authentication data via the conversational interface (text or speech).
After a user's account has been linked, PINs or passwords may be used as part of a second verification process.
So you need to use Account Linking to connect to the existing account on your service.
How can I do Account Linking if I don't require Google Sign-In?
You can still use Google Sign-In for Assistant if it will (or may) provide the information as part of the profile that match what you have. So it doesn't need to use the same account - just have the same email (for example).
But that still may not be enough.
For other cases, you can look into setting things up to work with an OAuth server that you control.
So why use Google Sign-In if I setup an OAuth server that uses Google Sign-In?
Google Sign-In is good for a more streamlined flow, if you can use it. It can be done completely with voice, such as with a smart speaker, instead of requiring the user to go to a phone to complete the login. So if you have the user's email address in your account system, and you also get this from Google Sign In, then you can connect the two accounts.
In some cases, such as if the user is expected to have logged into the account on your website first, they won't even need to do that. If both the voice client and web client use the same Google project, then authentication will take place automatically.
I wish there was a central, fully customizable, open source, universal login system that allowed you to login and manage all of your online accounts (maybe there is?)...
I just found RPXNow today after starting to build a Sinatra app to login to Google, Facebook, Twitter, Amazon, OpenID, and EventBrite, and it looks like it might save some time.
But I keep wondering, not being an authentication guru, why couldn't I just have a sleek login page saying "Enter username and password, and check your login service", and then in the background either scrape the login page from say EventBrite and programmatically submit the form with Mechanize, or use an API if there was one? It would be so much cleaner and such a better user experience if they didn't have to go through popups and redirects and they could use any previously existing accounts.
My question is:
What are the reasons why I shouldn't do something like that?
I don't know much about the serious details of cookies/sessions/security, so if you could be descriptive or point me to some helpful links that would be awesome. Thanks!
Edit:
I'm familiar with OpenID and the APIs. I was really wondering about the security/legal/confidentiality side of things. I understand the confidentiality part totally, don't know if there's anything legally written down about this, but assuming it's under ssl, and I don't store any of the data (will store the cookies and tokens), what are the security implications?
If I come to your website and give you my gmail password, what guarantee do I have that you won't read all my emails and even send a few of your own? And what if you become a little smarter and say 'people reuse passwords, I might just as well try if this password works for his bank account'.
As a user, I don't trust your site with my password. Period.
The whole point of Open Id and OAuth (that's what RPX uses) is to get around the above issue. I can give your website restricted, revocable and configurable access to my facebook account, all without giving your website my facebook password.
The UI is confusing, I agree. But with time people will understand what its all about, and it will be a lot better.
As already said above:
The site (or the site owner) accessing your {google|yahoo|etc} account cannot be trusted not to change your password and kick you out of your account.
But I feel there are other good reasons:
Many people use the same password on more than one site ore account (some could have the same password on gmail and paypal) and the site owner could abuse that
The site owner doesn't want to be held liable for other site owners abusing your account
The site owner could not be able to store your username and password in secure fashion. The site needs to be able to access them automatically. So on the server hosting there is stored everything needed to access those credentials.
And the hosting usually happens in a shared or virtual server with the hosting company administrators (and sometimes - if the hosting company isn't too conscious - fellow users) able to access them.
Security and Confidentiality. Period.
Even some websites like Facebook discourage using this approach in their TOS i believe. If so, it will be illegal to do so.
Closed. This question is off-topic. It is not currently accepting answers.
Want to improve this question? Update the question so it's on-topic for Stack Overflow.
Closed 10 years ago.
Improve this question
I'd like to know what people consider best practice for securing the Admin sections of websites, specifically from an authentication/access point of view.
Of course there are obvious things, such as using SSL and logging all access, but I'm wondering just where above these basic steps people consider the bar to be set.
For example:
Are you just relying on the same authentication mechanism that you use for normal users? If not, what?
Are you running the Admin section in the same 'application domain'?
What steps do you take to make the admin section undiscovered? (or do you reject the whole 'obscurity' thing)
So far, suggestions from answerers include:
Introduce an artificial server-side pause into each admin password check to prevent brute force attacks [Developer Art]
Use separate login pages for users and admin using the same DB table (to stop XSRF and session-stealing granting access to admin areas) [Thief Master]
Consider also adding webserver native authentication to the admin area (e.g. via .htaccess) [Thief Master]
Consider blocking users IP after a number of failed admin login attempts [Thief Master]
Add captcha after failed admin login attempts [Thief Master]
Provide equally strong mechanisms (using the above techniques) for users as well as admins (e.g. don't treat admins specially) [Lo'oris]
Consider Second level authentication (e.g. client certificates, smart cards, cardspace, etc.) [JoeGeeky]
Only allow access from trusted IPs/Domains, add check to basic HTTP pipeline (via e.g. HttpModules) if possible. [JoeGeeky]
[ASP.NET] Lock down IPrincipal & Principal (make them immutable and non-enumerable) [JoeGeeky]
Federate Rights Elevation - e.g. email other admins when any admin's rights are upgraded.
[JoeGeeky]
Consider fine-grained rights for admins - e.g. rather than roles based rights, define rights for indicidual actions per admin [JoeGeeky]
Restrict creation of admins - e.g. Admins cannot change or create other admin accounts. Use a locked-down 'superadmin' client for this. [JoeGeeky]
Consider Client Side SSL Certificates, or RSA type keyfobs (electronic tokens) [Daniel Papasian]
If using cookies for Authentication, use separate cookies for admin and normal pages, by e.g. putting the admin section on a different domain. [Daniel Papasian]
If practical, consider keeping the admin site on a private subnet, off the public internet. [John Hartsock]
Reissue auth/session tickets when moving between admin/normal usage contexts of the website [Richard JP Le Guen]
These are all good answers... I generally like to add a couple additional layers for my administrative sections. Although I've used a few variations on a theme, they generally include one of the following:
Second level authentication: This could include client certificates (Ex. x509 certs), smart cards, cardspace, etc...
Domain/IP restrictions: In this case, only clients coming from trusted/verifiable domains; such as internal subnets; are allowed into the admin area. Remote admins often go through trusted VPN entrypoints so their session would be verifiable and is often protected with RSA keys as well. If you're using ASP.NET you can easily perform these checks in the HTTP Pipeline via HTTP Modules which will prevent your application from ever receiving any requests if security checks are not satisfied.
Locked down IPrincipal & Principal-based Authorization: Creating custom Principles is a common practice, although a common mistake is making them modifiable and/or rights enumerable. Although its not just an admin issue, it's more important since here is where users are likely to have elevated rights. Be sure they're immutable and not enumerable. Additionally, make sure all assessments for Authorization are made based on the Principal.
Federate Rights Elevation: When any account receives a select number of rights, all the admins and the security officer are immediately notified via email. This makes sure that if an attacker elevates rights we know right away. These rights generally revolve around priviledged rights, rights to see privacy protected information, and/or financial information (e.g. credit cards).
Issue rights sparingly, even to Admins: Finally, and this can be a bit more advanced for some shops. Authorization rights should be as discreet as possible and should surround real functional behaviours. Typical Role-Based Security (RBS) approaches tend to have a Group mentality. From a security perspective this is not the best pattern. Instead of 'Groups' like 'User Manager', try breaking it down further (Ex. Create User, Authorize User, Elevate/Revoke access rights, etc...). This can have a little more overhead in terms of administration, but this gives you the flexibility to only assign rights that are actually needed by the larger admin group. If access is compromised at least they may not get all rights. I like to wrap this in Code Access Security (CAS) permissions supported by .NET and Java, but that is beyond the scope of this answer. One more thing... in one app, admins cannot manage change other admin accounts, or make a users an admin. That can only be done via a locked down client which only a couple people can access.
If the website requires a login for both regular activities and admins, e.g. a forum, I'd use separate logins which use the same user database. This ensures that XSRF and session-stealing won't allow the attacker to access administrative areas.
Additionally, if the admin section is in a separate subdirectory, securing that one with the webserver's authentication (.htaccess in Apache for example) might be a good idea - then someone needs both that password and the user password.
Obscuring the admin path yields almost no security gain - if someone knows valid login data he's most likely also able to find out the path of the admin tool since he either phished it or keylogged you or got it via social engineering (which would probably reveal the path, too).
A brute-force protection like blocking the user's IP after 3 failed logins or requiring a CAPTCHA after a failed login (not for the first login as that's just extremely annoying for legit users) might also be useful.
I reject obscurity
Using two authentication systems instead of one is overkill
The artificial pause between attempts should be done for users too
Blocking IPs of failed attempts should be done for users too
Strong passwords should be used by users too
If you consider captchas ok, guess what, you could use them for users too
Yes, after writing it, I realize that this answer could be summarized as a "nothing special for the admin login, they are all security features that should be used for any login".
If you do use only a single login for users who have both normal-user privileges and admin privileges, regenerate their session identifier (be it in a cookie or a GET parameter or whatever...) when there is a change in the level of priviledge... at the very least.
So if I log in, do a bunch of normal user stuff and then visit an admin page, regenerate my session ID. If I then navigate away from an admin page(s) to a normal user page, regenerate my ID again.
Have a good admin password.
Not "123456" but a sequence of letters, digits and special characters long enough, say, 15-20 characters. Like "ksd83,'|4d#rrpp0%27&lq(go43$sd{3>".
Add a pause for each password check to prevent brute force attack.
Here are some other things to consider:
One option to consider, especially if you manage the admin's computers or they are technically competent, is to use something based on SSL certificates for client authentication. RSA keyfobs and whatnot can also be used for added security.
If you're using cookies at all - perhaps for an authentication/session token - you probably want to ensure that the cookies are only sent to the admin pages. This helps mitigate the risks posed to your site by stealing cookies, by either layer 1/2 compromise or XSS. This can be done easily by having the admin portion being on a different hostname or domain as well as setting the secure flag with the cookie.
Restricting by IP can be smart as well, and if you have users throughout the internet you can still do this, if there is a trusted VPN that they can join.
We use Windows Authentication for admin access. This is most practical way of protecting admin areas while keeping the authentication seperate from what applies to general end-users. The system admin manages the Admin user access credentials and enforces password policies on the domain user account.
The strict way is to have two complete different "farms" including databases, servers and all and move the data from one farm to the other. Most modern, large scale, systems use this approach (Vignette, SharePoint, etc.). It's normally refered to as having different stages "editing stage" -> "preview stage" -> "delivery stage". This method lets you treat content/config the same way you treat code (dev->qa->prod).
If you are less paranoid you can have a single database but only have your admin section available on the "editing" servers. I mean, only have the editing scripts/files placed on the editing server.
Naturally the editing stage should only be available on a local intranet and/or using a VPN.
This may seem a bit of an overkill and may not be the easiest solution for all usage cases, but it is definetly the most robust way of doing things.
Note that things like "have strong admin passwords" are nice, but still leave your admin open to smart attacts of all sorts.
It very much depends on what kind of data you want to protect (legal requirements and such).
Alot of suggestions is about authentication.. I think you just should consider using OpenId / Facebook authentication as login. (They will most likely spend more resources on authentication security then you)
Save changes as well as updating values in the database. That way you can rollback changes from user X or between date X and Y.
I didn't notice anyone mention storage/validation of the admin password. Please please please do not store the PW in plain text, and preferably not even something that can be reversed - use something like a salted MD5 hash so that at the very least if someone happens to retrieve the stored "password" they don't have anything terribly useful, unless they also have your salt scheme.
Add a password field and a security question that the Administrator will know, e.g. what was your first girlfriend name, or randomize the questions everytime viewing the admin panel.
Perhaps you could always put the administration section in a big directory, e.g.
http://domain.com/sub/sub/sub/sub/sub/index.php
But that's not really good hah.
Perhaps you could include a query string in the home page, like:
http://domain.com/index.php?display=true
When it does, the username and password field will appear.
When supporting a new web app in an enterprise environment, it is often necessary to log in as a specific user in order to diagnose a real or perceived problem they are having. Two opposing issues apply here:
Best practice is to use hashed or encrypted passwords, not clear text. Sometimes, there is a third-party SSO (single sign-on) in the middle. There is no way to retrieve the user's password. Unless the user provides it (not encouraged), there is no way to log in as that user.
Many web app's have personalization and complex authorization. Different users have different roles (admin, manager, user) with different permissions. Sometimes users can only see their data -- their customers or tasks. Some users have read-only access, while others can edit. So, each user's view of the web app is unique.
Assume that in an enterprise environment, it isn't feasible to go to the user's desk, or to connect directly to their machine.
How do you handle this situation?
Edit: I want to reiterate that in a large financial institution or typical Fortune 500 company with hundreds of thousands of employees all of the country, and around the world, it is not possible for a mere developer in some IT unit to be able to directly access a user's machine. Some of those are public-facing web apps used by customers (such as online banking and stock trading). And, many of those are intranet applications rely on Active Directory or an SSO, meaning that user credentials are the same for many applications. I do thank you all for your suggestions; some may be highly useful in other kinds of environments.
A number of these ideas inconvenience the user, either by forcing them to change their password, or by occupying their desktop for your debugging session.
Markc's idea is the best: augment your authentication logic to allow superusers to log in as a particular user by supplying not the user's credentials, but the user's name plus their superuser credentials.
I've done it like this in the past (pseudo-ish python):
if is_user_authenticated(username, userpassword):
login the user
else if ':' in userpassword:
supername, superpassword = userpassword.split(':')
if is_superuser_authenticated(supername, superpassword):
login the user
In other words, if the username and password don't authenticate, if the password has a colon, then it's actually the admin username and admin password joined by a colon, so login as the username if they are the right admin username and password.
This means you can login as the user without knowing their secrets, and without inconveniencing them.
For our web applications we use a process that for lack of a better term is defined as 'hijacking' a user's account.
Basically, administrators can 'hijack' a user's account with a simple button click. In the code, you simply use a unique identifier (user id works in a less secure environment) that then establishes the necessary credentials in the session so that they can then work within that user's profile. For a more secure environment you could use a unique hash for each user.
In order to ensure that this hijack method is secure, it always first verifies that the request is being made by an authenticated administrator with the appropriate rights. Because of this it becomes necessary for either the administrator's session to be hijacked or for their authentication credentials to be captured in order for someone to ever exploit the hijack function within the application.
I had 4 ideas. While I was typing 3 of them were already suggested (so I upvoted them)
Variant on idea 3 - impersonation:
To make this as "identical as possible" to a normal login with minimal code changes, you might add the ability to impersonate directly at login by supplying Admin credentials plus an alternate username, e.g. login as Admin:user, adminpassword. The system would treat this exactly as logging in as user with userpassword.
Idea 4: Can you access the password store? If so, temporarily replace the user's hash with the hash of a known password. (the passwords are often stored online in a database. A SQL Query tool can do the swaps )
An administrator should be able to change a user's password. Change the password for the user to something you know. You can then log in as that user.
Tell the user to reset his/her password after you are done debugging.
Usually by some sort of remote control software that can be used to view their desktop. If they're on a Windows terminal server, then the built in admin tools can be used for that. Otherwise I'd use something like VNC across an internal network, or an external service like LogMeIn (http://www.logmein.com/).
Could you have a testing environment where there is a regular cut of live data copied to (obviously sanitised to meet any security or data protection issues). A user similar in setup to the one having trouble could be used to troubleshoot or indeed the very user if this is allowed.
Use a remote desktop client as mentioned in other answers, but again this may not be practical for you. If you have these rights within the domain, I have heard of error handling even doing a screenscrape and including this in logs! but this sounds a little odd to me.
Could you have an admin tool to clone a user into a demo account?
The solution we have used in our web apps is to have the authN/authZ return the desired user as the effective user. We do this by having an admin feature to setup a masquerade, and then when we ask for the currently logged in user (current_user), we handle the masquerade:
def current_user_with_effective_user
if masked?
current_user_without_effective_user.masquerade_as
else
current_user_without_effective_user
end
end
alias_method_chain, :current_user, :effective_user
I am considering adding OpenID to our customer facing admin and control panel areas...
1 - Associating OpenID's With Existing Accounts
For customers that already have accounts with us, I'm thinking they would need to login using their existing account number that we issue and then I'd have a mechanism to associate their OpenID with that account in their account management area (call it 'OpenID Manager' for the sake of argument).
In the 'OpenID Manager', presuming the user already has an OpenID, would I authenticate the user against their OpenID then associate with our generated account number for future OpenID logins (assuming that they authenticated ok)?
2 - Sensitive Data
Although we don't store full credit card data in our DB there is other data that is sensitive, invoices, domain reg details etc. After reading this article http://idcorner.org/2007/08/22/the-problems-with-openid/ I'm a little cautious about the idea of using OpenID in this way, what's the general consensus with you folks?
It seems to me that a lot of the arguments against OpenID are either made out of ignorance or by people with an axe to grind.
For example, the document you link to complains that identifying yourself with a URI is "dehumanising and more than a little frightening". Is that a legitimate complaint, or something written by somebody desperate to find things to complain about?
The two major things that get brought up are phishing and compromised accounts and these arguments have been rehashed so many times, it's hard to take somebody seriously if they bring them up yet again with no new points to make.
Phishing protection depends on the provider. Some providers offer much better security than typical websites ever would. Some providers just offer the typical username and password. Either way, if an account is compromised, that's something between the user and their provider, it's not your concern. You don't worry that the end-user has a keylogger installed on their computer, do you? That's because their local security isn't your responsibility, even though it might be used to gain access to their account. Likewise with OpenID - its security is not your responsibility.
If you compromise an OpenID, it gives you access to more than a single website. Sure, but the same is true for email. Just say you've forgotten your password, and you get sent a new one. You now have access to every account they've registered with that email address.
OpenID is no worse than the status quo, and it's significantly better in many circumstances, especially for informed users. If you are still wary of it, then just make it optional, so only the informed users use it.
I'd allow the registration of multiple OpenIDs with a particular account. That's a nice feature to have because it allows users to migrate between OpenIDs should the need ever arise.
That said, the idcorner link raises a good point. I think he massively overblows the security issue and makes many idiotic assumptions about how OpenID providers work, but that OpenID really isn't intended to replace all forms of user authentication. It's designed to make it easy for "drive-by" users to interact with a site with some form of basic authentication.
Ever been to somebody's blog, want to post a comment, but first you have to step through a 3-page registration? OpenID solves that problem.
Want to post a quick bug report on a public tracker but need an account first? OpenID to the rescue.
Want to store sensitive proprietary data in a web-accessible way and provide access only to people who are trusted? OpenID is not the solution.