zxcvbn VS pashword

Lightweight is an understatement here.

A client's project (with not necessarily technical customers) has had pretty reasonable success using the Dropbox originated library[1] for this, `zxcvbn`[2], on both frontend via js (for "instant" feedback) and on the backend via php (to enforce the requirements when writing password hashes to the database)

1: https://dropbox.tech/security/zxcvbn-realistic-password-stre...

2: https://github.com/dropbox/zxcvbn

use zxcvbn to check your password strength more thoroughly

source if anyone wants the whole list https://github.com/dropbox/zxcvbn/blob/master/data/passwords.txt

Password strength is evaluated based on the zxcvbn library.

In contrast, let's consider the password "zxcvbn214". How might we assign an entropy to this password? Is it 369? Or 266 * 103? Anyone familiar with a QWERTY keyboard or Dropbox's password strength estimator knows that "zxcvbn" is hardly a random sequence of letters. This same principle applies to "l33t" speak, e.g. replacing all "e"s with 3s and "a"s with 4s. These strategies may "trick" simple entropy calculations into estimating a high entropy, but it won't trick sophisticated attackers. This leads to strength over-estimation, which is, I argue, the worst thing we can do in this context.

> except for ZXCVBN

You mean the Low-Budget Password Strength Estimator?

https://github.com/dropbox/zxcvbn

Yeah, that name is totally legit.

For any part of the password that the zxcvbn cannot match to a known pattern, it uses a brute-force cardinality of 10, i.e., it estimates that the number of guesses required to crack a password or password segment of length N is equal to 10N (equivalent to the number of guesses required to exhaust all possibilities if your password consisted only of numbers).

We took a similar approach to passphrase stretching in EnvKey[1] v1 (EnvKey is a secrets manager, not a passwords manager, but uses end-to-end encryption in a similar way). We used PBKDF2 with iterations set a bit higher than the currently recommended levels, as well as Dropbox's zxcvbn lib to try to identify and block weak passphrases.

Ultimately, I think it's just not good enough. Even if you're updating iteration counts automatically (which is clearly not a safe assumption, and to be fair not something we did in EnvKey v1 either), and even with safeguards against weak passphrases, using human-generated passphrases as a single line of defense is just fundamentally weak.

That's why in EnvKey v2, we switched to primarily using high entropy device-based keys--a lot like SSH private keys, except that on Mac and Windows the keys get stored in the OS keychain rather than in the file system. Also like SSH, a passphrases can optionally be added on top.

The downside (or upside, depending how you look at it) is that new devices must be specifically granted access. You can't just log in and decrypt on a new device with only your passphrase. But the security is much stronger, and you also avoid all this song and dance around key stretching iterations.

1 - https://github.com/envkey/envkey

2 - https://github.com/dropbox/zxcvbn

This is beautifully done, great design work.

Scrypt for password stretching seems good. I see you're using CPU cost of 2^15. When storing a password hash you'd want to use 2^17 (with agility to change algorithm or increase cost in the future) [1]. Since you're not storing the result, I suspect the lower number is reasonable.

I don't like simple concatenation when building a salt from two variable length fields. You'll get the same salt for `"foo" + "bar"` and `"foob" + "ar"`, but the salt should be unique. Although I don't think that's an issue for this project since the first is a website.

Using the website in the salt has some issues when there are multiple domains that use the same password. Do I use mail.google.com, auth.google.com, or google.com? trello.com or atlassian.net? What if the website it bought and the new owner changes the domain name? With a password manager, I can just look in my vault to figure out the old domain name.

Phishing is a major way passwords are stolen and this project doesn't seem to do anything to protect against that. A browser extension (and mobile app), that checks the domain name before showing/filling the password could help.

The secret key field let me use `1234` as the key, although the color of the field was red. I think this should either prevent obviously weak passphrases or show a much more obvious warning if when one is used. Using a password found in a breach is also a bad idea (even it the password looks strong). You don't have a way to check HIBP, so users will be vulnerable if they make that mistake. It's too easy to make a critical mistake with the current design.

A bug: I filled out the form but forgot to enable JavaScript. The form posted my passphrase back to the server (https://pashword.app/?website=google.com&username=me&passphr...). I'd recommend changing the form so the submit button doesn't do anything when JS isn't loaded, otherwise the server will learn users passphrases. This is also a good place to remember that the user fully trusts that you wont steal their info (I'm not sure why anyone should trust that).

Also check out other similar projects, lots of discussion which likely applies here as well. I believe one of these supports uses a counter to support password rotation. You'd just need to remember the counter value for each site.

* LessPass - https://news.ycombinator.com/item?id=12889807

Just take a look here: https://github.com/pashword/pashword/blob/main/pages/index.tsx

I'm looking for something close to https://pashword.app

Password managers can be a bit hard to manage, people don't even bother using them. There's https://pashword.app that solves this but not many people know about it.