3 ms·

Just by itself, long passwords will result in people using "franklymydearIdontgiveadamn" and other things susceptible to a dictionary attack. Long passwords pl

Just by itself, long passwords will result in people using "franklymydearIdontgiveadamn" and other things susceptible to a dictionary attack. Long passwords plus a bad password check (https://www.troyhunt.com/ive-just-launched-pwned-passwords-version-2/ https://www.troyhunt.com/ive-just-launched-pwned-passwords-v...) might suffice. However, without that check, I suspect requiring special characters marginally improves entropy in practice.

guitarbill 8y ago
Pretty much everybody I know and asked in the office just sticks some punctuation at the end, e.g. "Passw0rd$". IDK how much of an "improvement" that is, compared to e.g. 4 extra characters.
- hyperpape 8y ago
  That's definitely a thing, and you're right, it doesn't add much entropy (people choose from a few numbers, a few special characters, not that many substitutions). Four extra characters that are truly random (even just out of 26 letters) is way better. However, that's not the alternative I'm worried about. I'm worried about people using phrases that will be in the first few thousand entries in a phrase dictionary.
beatgammit 8y ago
And that's still better than a crappy, short password. I use lyrics, quotes, etc for my important passwords (password manager, email, etc), and my passwords are >20 characters (usually >30) and really easy to remember (and I can secure it by placing one extra symbol in there). The attacker doesn't know what I'm using for my password, so they have to attack all valid combinations up to the max length the server supports.
- hyperpape 8y ago
  Song lyrics are not safe https://arstechnica.com/information-technology/2013/08/thereisnofatebutwhatwemake-turbo-charged-cracking-comes-to-long-passwords/ https://arstechnica.com/information-technology/2013/08/there... I don’t know exactly where the breakeven point is, but there is some point where a song lyric or other phrase provides less entropy than a relatively short password. Bottom line: use a lot of random ascii characters(at least 14 or 16, or 4 random words) and you’re probably set. Use less, or a found phrase and it’s not good.
  - acdha 8y ago
    You can't talk about safe without a threat-model, especially distinguishing between whether you're trying to prevent offline attacks against stored data or brute-forcing a running login system. If you can rate-limit many things are fine, especially if you have some confirmations at key points (“Your account was just accessed from …”) with restrictions which make it hard for an attacker to permanently wrest control of the account or incur large costs immediately after gaining access — think about how Amazon.com doesn't require 30 character passwords because adding a new shipping address requires you to re-confirm the credit card information first.
    - hyperpape 8y ago
      That's definitely true, though you can simplify matters by saying "I'm always going to try to have passwords that are safe against offline attacks".
      - acdha 8y ago
        I would ask what’s the cost benefit ratio for defending against offline attacks with increasingly unpleasant password requirements versus things like increased KDF rounds, general server security improvements, and the defense in depth measures which increase the costs of actually profiting from a user account compromise. I think we’ve historically been shaped by classic Unix crypt() memories where something couldn’t easily be fixed without breaking compatibility even though that era is largely in the past.
        hyperpape 8y ago
        I see how this wasn’t obvious, but I never actually said anyone should implement any particular password requirements, just that certain requirements might not increase entropy as much as you might think. My above comment about offline attacks was my personal stance/ideal: I use a password manager for just about everything, not something I’d mandate. As a developer, in many contexts I might implement min 8 chars with no special char requirements. I think the bad password check is also a good idea.

guitarbill 8y ago

Pretty much everybody I know and asked in the office just sticks some punctuation at the end, e.g. "Passw0rd$". IDK how much of an "improvement" that is, compared to e.g. 4 extra characters.

hyperpape 8y ago
That's definitely a thing, and you're right, it doesn't add much entropy (people choose from a few numbers, a few special characters, not that many substitutions). Four extra characters that are truly random (even just out of 26 letters) is way better. However, that's not the alternative I'm worried about. I'm worried about people using phrases that will be in the first few thousand entries in a phrase dictionary.

beatgammit 8y ago

And that's still better than a crappy, short password. I use lyrics, quotes, etc for my important passwords (password manager, email, etc), and my passwords are >20 characters (usually >30) and really easy to remember (and I can secure it by placing one extra symbol in there). The attacker doesn't know what I'm using for my password, so they have to attack all valid combinations up to the max length the server supports.

hyperpape 8y ago
Song lyrics are not safe https://arstechnica.com/information-technology/2013/08/thereisnofatebutwhatwemake-turbo-charged-cracking-comes-to-long-passwords/ https://arstechnica.com/information-technology/2013/08/there... I don’t know exactly where the breakeven point is, but there is some point where a song lyric or other phrase provides less entropy than a relatively short password. Bottom line: use a lot of random ascii characters(at least 14 or 16, or 4 random words) and you’re probably set. Use less, or a found phrase and it’s not good.
- acdha 8y ago
  You can't talk about safe without a threat-model, especially distinguishing between whether you're trying to prevent offline attacks against stored data or brute-forcing a running login system. If you can rate-limit many things are fine, especially if you have some confirmations at key points (“Your account was just accessed from …”) with restrictions which make it hard for an attacker to permanently wrest control of the account or incur large costs immediately after gaining access — think about how Amazon.com doesn't require 30 character passwords because adding a new shipping address requires you to re-confirm the credit card information first.
  - hyperpape 8y ago
    That's definitely true, though you can simplify matters by saying "I'm always going to try to have passwords that are safe against offline attacks".
    - acdha 8y ago
      I would ask what’s the cost benefit ratio for defending against offline attacks with increasingly unpleasant password requirements versus things like increased KDF rounds, general server security improvements, and the defense in depth measures which increase the costs of actually profiting from a user account compromise. I think we’ve historically been shaped by classic Unix crypt() memories where something couldn’t easily be fixed without breaking compatibility even though that era is largely in the past.
      - hyperpape 8y ago
        I see how this wasn’t obvious, but I never actually said anyone should implement any particular password requirements, just that certain requirements might not increase entropy as much as you might think. My above comment about offline attacks was my personal stance/ideal: I use a password manager for just about everything, not something I’d mandate. As a developer, in many contexts I might implement min 8 chars with no special char requirements. I think the bad password check is also a good idea.