How can that be done? That's between my phone and Google, so how can they "listen in" on that?
The simplified version is, Google sends the browser a one-time key, which the browser forwards to the HW token to sign with its private key. Then the browser sends this back to the web server to verify, using its copy of the HW token's public key.
This would be vulnerable to MITM attacks, as you say.
So what the protocol actually does is concatenate the nonce sent by the web server with the origin of the web page as seen by the browser and have the HW token sign that. This way the server can verify that the HW token signed the right nonce for the right origin.
See https://docs.google.com/document/d/1SjCwdrFbVPG1tYavO5RsSD1Q..., search for "origin".
So… whichever login attempt gets to confirmation stage last wins (not relevant in this situation), and the confirmation screen on (at least) my phone does not indicate anything regarding location (which is highly relevant).
This looks a little weaker than TOTP (you're basically trading a little security for the convenience of not entering a code while keeping the second factor) and a lot weaker than U2F.
^Instead of using "standard" 2-factor that generates a code on-the-fly within an app like GAuth or Authy, users receive a text message with 10 codes. The first digit of every code increases sequentially (0972,1042,2512,etc), must be used in that order (0 code on first login, 1 code on second, etc.), and the page informs the user which number they're on.
Duo offers a choice of authentication methods, depending on the usability and security requirements of your application or organization.
Duo Push is actually one of the easiest (and most secure) authentication methods, as one of the commenters pointed out:
It might be worth pinging your IT/security dept to ask about enabling Duo Push as an option or to change the policy for SMS passcodes (eg. you can just have one passcode sent instead of ten).
- Jon Oberheide, Co-Founder & CTO @ Duo
From the policy:
"Device-Specific Information: We also collect device-specific information (e.g. mobile and desktop) from you in order to provide the Services. Device-specific information includes:
attributes (e.g. hardware model, operating system, web browser version, as well as unique device identifiers and characteristics (such as, whether your device is “jailbroken,” whether you have a screen lock in place and whether your device has full disk encryption enabled));
connection information (e.g. name of your mobile operator or ISP, browser type, language and time zone, and mobile phone number); and
device locations (e.g. internet protocol addresses and Wi-Fi).
We may need to associate your device-specific information with your Personal Information on a periodic basis in order to confirm you as a user and to check the security on your device."
The policy continues to state that Duo may use this data for analytic/advertising purposes (although only in-house) as well as to comply with legal requests, subpoenas, NSLs etc.
Duo isn't collecting your data for nefarious purposes or to sell it to other companies but they still are collecting A LOT of it. Other two factor methods, like the one's used by Google and Facebook, allow clients to install their own code generators that don't collect personal data or even need access to the internet. Of course these methods don't have push requests that you can just approve rather than type in the code.
Another way to look at it: We collect security-relevant information on your device, but not your _personal_ data. In other words, we don't collect your email, photos, contacts, user-generated data, etc.
Most importantly to me, though, the system has thus far been completely reliable. I haven't yet heard of a single case where somebody couldn't log in because of Duo.
I'm not sure what our enterprise agreement is / how much this all costs, but it's a very good system for us.
My Duo hardware token (the code generator with the button and the LCD) tends to "desynchronize" after long periods where you don't use it. The internal clock gets off, so it drifts in what token it returns vs what the server thinks it should be returning, and then it stops working.
Normally, if you log in on a regular basis the server corrects for this drift. There is probably a sliding window of N valid keys (say 10) and using one of them tells the server what the internal clock state is. But if you don't use it for a long time (more than 30 days in my experience), the clock drifts, you start going outside the window and it refuses to let you log in.
If your IT desk is open, they can "resync" it by typing in a couple numbers in a row, which lets the server scan the key sequence and find where your token is.
Use-case: We don't have Duo tokens rolled out system-wide, they are only issued for admin tasks and we have separate admin accounts for these with the Duo attached. I'm an "occasional sysadmin" who administrates several stable servers that mostly don't need to be touched.
As I don't need to use it day-to-day, my key desynchronizes quite often for me, I have had it happen at least 3 times. It would be bad if I had an after-hours emergency with my Duo token, I do not trust it. The hardware tokens are not reliable, in my book.
edit: The fix for me would be for the token to automatically resynchronize on the fly. Just like the IT guys can do, but over-the-wire. If the server sees (f.ex) three sequential login attempts with valid-but-stale keys, with the proper order and timing pattern, then it accepts them and resynchronizes the key window.
To prevent replay attacks, you would also need to add a constraint that the keys be newer than one ones last used for a sucessful login, but it should be doable. You would also want to avoid causing an account lockout as you type in the invalid keys.
Thanks for the reply! I'll definitely get in contact with the school's OIT to figure out alternate options for authentication
I do plan on getting in contact with the schools OIT for enabling alternatives
I think institutions also use Duo because Duo takes care of the whole think whereas traditional 2FA isn't trivial to implement for the institution (generating tokens and all of that). At least that's what I was told by my institution when they made us start using Duo.
One of the more annoying things is that the codes are sent from a random 386 number. Out of the 7+ texts I've received thus far, only 2 were from the same number.
Apparently the company they're using is named https://duo.com/
I don't recall whether I had the option to use the app when I enabled MFA initially. However, after the fact, and as far as I can find, I cannot go back and enable the app.
I remember that configuring this is tricky, but I did eventually get user self enrollment configured with push being the default. Happy to dig more into my config, if you're curious: email@example.com
SMS for 2fa is poor to begin with. I wish people would at least implement the standard TOTP/HOTP option as well if they are going to pull stuff like that.
Like, that would prevent me from using 2FA.
Whatever happened to standards?
However I wouldn't want my second-factor to be attached to my browser. Seems way too volatile for me. Personally I'd rather keep TOTP and be vulnerable to time-of-use phishing.
Maybe if the browser had an OS API that a YubiKey could query...
1. Currently Chrome has this, Firefox is close (50.1 shipped it but it only works in the e10s mode), and there are extensions for Safari and older versions of Firefox.
> the URL is not included in the hash
What hash? Nobody even mentioned a hash. The crypto keys used for U2F are indeed domain-specific, if that's what you're trying to ask.
> It could be by having those two talk to each other.
Who's "those two"? And what's "it"? I'm very confused.
I mentioned a hash. The secret is hashed together with the time. _That_ hash.
> The crypto keys used for U2F are indeed domain-specific, if that's what you're trying to ask.
I know the secret is domain-specific. What I was describing is taking the secret, and the time AND THE DOMAIN and use them to produce the hash. This would break MITM. One of the comments above me mentioned this and I run with it. But you're talking to me like you didn't read anything above....
> Who's "those two"?
Those two are the yubikey and the browser.
I think you're confused. You have not mentioned the word "hash" even once in this thread prior to the previous comment I replied to.
Anyway, I think you're confusing U2F with TOTP. U2F does not rely on the time at all AFAIK; it uses public key cryptography, and authenticates by signing a data structure containing the domain name of the site and a server-provided nonce (among other things).
> What I was describing is taking the secret, and the time AND THE DOMAIN and use them to produce the hash.
I think there's still some sort of disconnect here, because up until this this comment you've described nothing of the sort in this thread. Could you link the comment you're referring to where you explained all this?
> One of the comments above me mentioned this and I run with it.
If you're referring to acdha's comment about U2F, as acdha and others in this thread have explained, U2F (aka Universal 2nd Factor) is an entirely different protocol from TOTP (aka Time-based One Time Password). U2F does not use hashing or the system time in the way you seem to be envisioning, but it is also not vulnerable to phishing like TOTP is.
U2F interfaces with your browser, and uses a set of public and private keys (that is stored on the U2F device, not in your browser) to authenticate to sites in a way which can't be phished. It's not theoretical; it exists and can be used today with many popular sites, including Google, GitHub, Dropbox, and more. You just need a USB device which supports U2F (YubiKey is one, but there are many others).
TOTP barely protects against phishing. What you want is an U2F key as the second factor. It's not like they are expensive anyway (usually 7-15 Euro) and quite some large services support U2F tokens already (Google, Dropbox, GitHub, Fastmail, etc.).
Is the 1 minute window always the case? In the authenticator app, it seems like codes expire after ~30 seconds. If I wait till the last few seconds before using the code, does that make me any safer?
It is common but not universal for sites to accept, at a given time, 1) the current U2F token, 2) the U2F token from the previous window, 3) the U2F token for the next window. This is done as a partial mitigation for potential clock skew issues on the client that's generating the TOTP codes (e.g. your phone). In practice this means every code is valid for 1m30s, although sites may customize this (with or without changing the window size, which is typically not done because that parameter must be consistent system-wide).
> If I wait till the last few seconds before using the code, does that make me any safer?
Maybe, but this is not practicable security advice. The latency of a MITM attack on a 2-factor TOTP login depends on the attack infrastructure and design, but can easily be made to be on the order of tens or hundreds or milliseconds. Reducing the window seems like it might help your security, but it can never be perfect and there is a direct tradeoff with usability because users need time to look up the codes on one device and enter them on another.
Folks often say "enable 2FA" in response to news of new and sophisticated phishing attack campaigns, but it's critical to note that most commonly deployed 2FA (TOTP, HOTP, SMS) is trivially phish-able. 2FA is not an automatic defense against phishing, although some newer designs achieve this and were created specifically with this goal in mind: U2F is a good example.
The validation system should compare OTPs not only with the receiving timestamp but also the past timestamps that are within the transmission delay. A larger acceptable delay window would expose a larger window for attacks. We RECOMMEND that at most one time step is allowed as the network delay.
We RECOMMEND a default time-step size of 30 seconds. This default value of 30 seconds is selected as a balance between security and usability.
Since the client's clock could be in the behind or ahead of the server's clock, I have to correct myself and the window would be 90 seconds.
One could be a bit strict and e.g. the previous time step only until 1/2-way the current time step, which would bring the window make to 60 seconds.
At any rate, all these timeframes are far to large to avoid real-time phishing attacks.
One of the tweets points out that something like lastpass would help with this as it wouldn't allow you to autofill your password (as it's not on the google the domain), but then you could get it manually from there anyway.
Normally this would alert you to the fact that someone is logging in to your account, and would stop the attacker since they lack the 2FA one time pass. In this case though, since you've already fallen for the "I'm trying to log in to Google again", the attacker will probably fake the 2FA screen as well, and you'll merrily type it in.
2. The attacker immediately tries them, triggering an SMS to you and an 'enter SMS code' page for them.
3. The attacker shows the 'enter SMS code' page to you, and you enter the code from the SMS you just received, giving it to the attacker.
4. The attacker completes their login using the SMS code.
5. The attacker shows the user some believable error message (implying an error on Google's end, or a typo in the SMS code) then forwards the user to the legitimate Google login page.
Obviously this isn't perfect because it depends on people actually paying attention to that, and on not having too many false positives due to GeoIP failures, but it seems like a nice improvement.
Apple has a nice UI on it (no surprise, I'm sure) where they show a map centered on the location in question, but even SMS-based solutions could include a quick "Login attempt from City" along with the code.
It's enough to concern me on the odd occasion that someone is trying a MITM attack.
I am guessing it is because in Australia, quite often the central server allocating IP addresses for our major ISPs can be in a completely different city?!?
Someone else mentioned U2F would work though as that's tied to the domain, but I don't really know much about that.
Not perfect but atleast they're not blindly typing in passwords.
U2F basically signs the current URI and HTTPS key and sends it back. If there is a man-in-middle then the signatures will not match and the auth will fail.