With 3 attempts allowed I only need 230k usernames, even better!
The "secret" stored in the browser doesn't protect you from this, since I'm not stealing someone's code; I'm logging in from my own browser. (Of course I'm doing this via a botnet, so you won't notice it by IP address)
By regenerating longer token each time a 6-digit token is required and only allowing something like 3 guesses per longer token (Before regenerating and sending a new 6-digit code) you have a secure system.
But sure, let me explain it:
Step 1: acquire a list of 230 000 usernames/emails.
Step 2: for each username/email, launch a browser and request a login code. The login code is sent to the users' email address, and the super-ultra-secret token is stored in the browser.
Step 3: for each username/email, do 3 random guesses from their respective browsers.
Each guess is a Bernoulli trial with P = 10^-6. Thus the probability of at least one success is 1 - (1 - 10^-6)^(230000 * 3) = 49.8%
I hope this made it clear for you. I think the fallacy in your reasoning was assuming that only legitimate users can get the "secret" token, which is false. It is true for things like debit cards, where the token is stored on the card itself, hence the 4-digit PIN with 3 tries allowed is sufficient. In this case however a "secret" token is handed out to anyone who requests a code. It protects you from someone peeking into your mailbox and stealing your code, but it fails against brute forcing codes on a large number of users.
1. Someone (anyone) initiates the login flow
2. The user record in a DB gets enriched with
- the 6-digit OTP (sent via email)
- the "secret" (shared with the initiating browser)
- the current session failed attempts counter (default: 0, max: 3)
- the total failed attempts counter since last successful login (default: 0, max: 20)
Let's assume that you've initiated the session (hence own the "secret"), but have no access to the OTP - you will always have 3 attempts to guess the OTP before the session gets reset (a new "secret" + new OTP). So after the 3rd unsuccessful attempt you start from scratch. And at that point other means of protection might come into play based on the second counter (throttling the auth. requests on the WAF/fronting server, blacklisting IP addresses, etc.)
The number of accounts you're targeting does not matter - in fact it may help other systems mentioned catch your malicious attempts sooner.
You didn’t explain how this gives you any protection akin to a physical chip on a debit card…
Let me illuminate the problem with a real-world analogy: this system is like me walking into a bank with a list of account names, and the bank issuing me debit cards for those accounts with random PIN codes, no questions asked. I’d have 3 attempts to guess the PIN, then the ATM swallows the card. Same as in OPs login system!
Still, don’t you think banks would be silly to do this?
>So after the 3rd unsuccessful attempt you start from scratch.
You don’t start from scratch. You do 3 attempts per user, period.
>The number of accounts you're targeting does not matter
It does matter. The entropy of their six-digit OTP is small enough that pwning at least one account from a pool of users is feasible. As I’ve shown above, the number of users necessary for pwning at least one account with 50% probability is log(0.5)/log((1 - 10^-[digits])^[attempts per user]). In our case digits = 6, and attempts per user = 3, which gives us ~231000 users. Which isn’t too many for a website.
In both cases you (an admin) control how many attempts you wish to accept before sacrificing the usability over security (temporarily disabling the card and forcing you to call the bank and disabling 6-digits OTP in favour of ie. 12-digits OTP).
Your real-world analogy is, this time, too simplified :) Let me counter-argument with my example:
You are in a a possession of 1 million user emails and request OTPs for all of them (let's assume no other security measures are in place). If you hit all of the 1M records with 3 attempts of OTP - according to my simple experiment - you might hit (on avg) 3.18 accounts.
Of course, in real life, your attack would be mitigated on other (non-application) level.
Please find my ugly code for that experiment: https://pastebin.com/S2ufabhU
PS. When OTP size is increased to 7 digits - the avg account hit drops to 0.33. For 8 digits - it drops to 0.02. (please mind I'm using lousy RNG :)
PS2. What I'm trying to emphasize is that there's nothing wrong with OTP-auth like this if properly implemented with other basic anti-bruteforce techniques. You could get "better" results with simple password spraying.
1. I don’t have any problem with OTP-based magic links, if they’re implemented carefully. As you’ve just demonstrated, a 6-digit OTP with 3 retries allowed is not sufficient at scale.
2. I’ve also demonstrated that OP’s proposal to augment the 6-digit code with a “secret” token gives you no additional security. It can only protect against someone peeking over your shoulder to snatch your OTP.
3. As you’ve just demonstrated, increasing the OTP length will mitigate the problem, as it increases entropy. For peace of mind, I’d recommend 64 bits of entropy, which can be achieved with a 13-letter alphanumeric code (case insensitive), or with 5 words from a dictionary of 10 000.
ad 1 - IMO, it's still sufficient at scale with some basic infra. hardening
ad 2 - AFAIK the "secret" was never meant to protect from brute-force, but rather mitigate threats from actors controlling the email part
ad 3 - again, let's be more pragmatic - no one would use it if it required typing 13-letters OTP :) there are other ways to mitigate the potential attack you've described.
Best.
1. This def needs hardening. I currently only use it for really low stakes stuff. I need to make that clearer in the README.
2. The next level of hardening I was planning on implementing is rate limiting code requests per email address. I figure I can step up the time-outs between code requests pretty aggressively to slow down brute forcing the code. The ATM that eats the card is a great analogy.
3. Maybe I missed it in the thread, but nobody mentioned that this from of auth makes it evident to the persons being attacked that something is happening if they have a bajillion code requests in their email inbox. There’s a lot of other problems that creates for these auth scheme, like filling a users inbox beyond its quota, triggering high volume breakers for SMTP relay services like AWS SES, etc. that wouldn’t be good.
For those watching, if you’re a security researcher feel free to open a GitHub issue on that repo and I’ll address it.
Try making password creation too difficult and now password reset will be the default authentication for a lot of users anyway.
I have nothing against password-reset-as-authentication aka "magic links", when done properly. And by properly I mean not using an ultra-low entropy auth code, like the parent is proposing (6 decimal digits, aka 20 bits of entropy).