Why not let the administrator decide? [LWN.net]

Why not let the administrator decide?

Posted Aug 11, 2026 15:52 UTC (Tue) by farnz (subscriber, #17727)
In reply to: Why not let the administrator decide? by mb
Parent article: Changes in shadow-utils password-expiration features

And while we are at it get rid of stupid character rules as well. A password does not get noticeably better by forcing the user to add numbers, because that will just make them put a 1 at the end.

It feels like the end point we're reaching towards is a human-friendly entropy estimator for passwords: roughly how long would it take a sophisticated password cracking tool to guess this password?

The entropy of "SzEp1r03G" (9 random characters, with 62 choices for each character, chosen by my computer's random number generator, so just under 54 bits of entropy) is similar to the entropy of a 4 Diceware word password such as "WrongnessPopulatePastorEclipse" (7776 words in a Diceware dictionary, 4 words, so just under 52 bits of entropy), and both passwords should thus be as acceptable (or unacceptable) to a systems administrator; a cracker who knows the exact algorithm used in both cases will take about the same amount of time to guess both passwords, whereas "lwn.net1998" is very easy to crack if you know the algorithm (domain name and date of founding for the site - extends nicely as an algorithm across many sites, but takes 1 guess per site).

The hard question is how to implement such an entropy estimator. I can estimate entropy for passwords I create algorithmically (since I can do the log2(possible choices) * number of choices made calculation) very easily, but shadow-utils won't know how the user created the password, and will need to make a guess about how hard it'll be for a password cracker to brute force the password.

to post comments

Why not let the administrator decide?

Posted Aug 11, 2026 16:02 UTC (Tue) by alx.manpages (subscriber, #145117) [Link] (5 responses)

> The hard question is how to implement such an entropy estimator.

I don't think password validators should concern about entropy. Users should use password generators, and those know the entropy of what they generate. The validators should just accept that users know what they're doing (or that at least, they should know).

If the admin wants their users to have strong passwords, it should point them to password generators, or at most, require a minimum length. More than that is experimentally known to be useless, or even detrimental.

Why not let the administrator decide?

Posted Aug 11, 2026 16:39 UTC (Tue) by farnz (subscriber, #17727) [Link] (2 responses)

If you're trying to make passwords secure, you end up needing an entropy estimate in order to be able to detect users who are lying to you - they claim to use a password generator, but do things like "tweak" the generated password to be more memorable (so they change "swxjzdap", which is generated by an 8 character, 26 choices per character ⇒ 37 bits entropy, to "sexizdad" for memorability).

Note that a minimum length is a weak entropy estimate - you know that written text has an entropy around 1 bit per character in most alphabetical languages, so you know that a minimum length of 10 characters is going to be 10 or more bits of entropy.

Ultimately, though, while it's fun to think about how to make passwords secure, we already know how to do far better - key based authentication (like SSH keys) means that you've got as many bits as you want for the authenticate to a system approach (a 256 bit ECDSA SSH key is equivalent to a 10 word Diceware passphrase, or 22 characters of random case-sensitive letters and numbers), and you can have a master key backed by a hardware token to give you 2FA for unlocking access to your keys (with, again, the large number of bits a key can have).

Why not let the administrator decide?

Posted Aug 11, 2026 16:51 UTC (Tue) by mb (subscriber, #50428) [Link] (1 responses)

we already know how to do far better - key based authentication (like SSH keys)

These only solve part of the problem. If done properly then they still need a (local) password. Otherwise they are as insecure as storing your password database as passwords.txt on the local disk.

Why not let the administrator decide?

Posted Aug 11, 2026 18:38 UTC (Tue) by farnz (subscriber, #17727) [Link]

Even a strong password that's stored in ~/passwords.txt on my local disk is a significant security upgrade on a weak password on a networked system. Attacking a system that deliberately lets you log in over the network is significantly simpler than getting a copy of a file on my local disk.

And the admin can't guarantee that I'm not doing something stupid - I did point out that you can have the keys stored in a key manager that requires a hardware token to unlock (modern password managers can act as SSH agents, for example, requiring a hardware token to unlock the password manager). But making the attacker attack my system to get my account, rather than opening up attacks like "crack the password someone uses on MySpace, transform it and use it to log into your system" is a worthwhile endeavour.

Why not let the administrator decide?

Posted Aug 11, 2026 16:45 UTC (Tue) by mb (subscriber, #50428) [Link] (1 responses)

If the admin wants their users to have strong passwords, it should point them to password generators

Well, yes and no.

If the generator generates truly random passwords then: No. This only works, if the password does never have to be remembered. This does not work for many passwords that matter like main login passwords or password manager main passwords.
If the generator generates these readable random passwords (like pwgen (1)) then: Maybe. They are still kind of hard to remember and therefore risk being short or written down on Post-Its.
If the generator generates long xkcd-936 style passwords then: Yes. Is there a generator with a big enough word database that can do this?

The most important thing and at the same time the hardest thing is to get the old and counterproductive rules out of people's heads. There are rules like "8 chars are enough", "passwords must be as complicated as possible" or "no words from the dictionary" that truly hurt security, but are buried deep in peoples heads. I have no idea how to do that easily, though. Unlearning is harder than learning. :)

Why not let the administrator decide?

Posted Aug 11, 2026 21:08 UTC (Tue) by alx.manpages (subscriber, #145117) [Link]

If the generator generates long xkcd-936 style passwords then: Yes. Is there a generator with a big enough word database that can do this?

There is. :)

$ whatis xkcdpass
xkcdpass (1)         - generate memorable, secure multiword passphrases

https://manpages.debian.org/buster/xkcdpass/xkcdpass.1.en.html

I have no idea how to do that easily, though.

We can have a mention in the manual page of passwd(1). It isn't great, but that's as much as I can do.