Myths of Password Security | Hacker News Reader

Myths of Password Security

stormpath.com

29 pointslightup8849 comments

With the recent flaw (though very small flaw) in 1Password, I decided to do a little research into the tools that password crackers use today. The first big tool is the cracker itself. John The Ripper and HashCat are 2 popular ones today, but I've focused my reading on hashcat.

Hashcat is a tool who's goal is to do various hashing functions as fast as possible in both CPU and GPU (depending on what the user has available). Where the tool gets interesting is the attack modes[1] it has. Each of the attack methods has different use cases, but I would say Rule-based and table-lookup are both fairly interesting (allow for smart generating of test passwords so things it's not a full character set brute-force).

Now, you have other groups that are going out and doing analysis on database password leaks[2]. This site tries to crack as many passwords as it can from publically leaked password databases (such as Gawker ane EHarmony), then do an analysis of the passwords they have cracked thus far. People can then use this data to write rules for Hashcat or other tools to possibly crack passwords a lot faster.

As rules for passwords change and people change their password habits, the crackers will adjust their methods. While this article does bring out some interesting points about passwords, it can still be good to know what tools the crackers have at their disposal so you can think about how to craft a password that won't be easily cracked.

[1] http://hashcat.net/wiki/#attack_modes

[2] http://www.adeptus-mechanicus.com/codex/hashpass/hashpass.ph...

A myth I'd like to dispel is that when login fails not telling whether username or password was wrong improves anything.

Please do tell clearly when username was wrong. It won't disclose anything new, because you're already disclosing this information by validating registration form (and reg form will even disclose existence of registered e-mails by forbidding duplicate registration), existence of profile URLs, all kind of "add a friend" APIs, etc.

If you're really concerned about somebody compiling list of usernames by brute force (and you've patched all other places where you disclose it) then put rate limiting on the login form (you should have it anyway to prevent brute forcing of passwords).

This comes from the days when you would have to call the IT desk, and have them reset your password.

I humbly disagree, if you say to the user that the username was correct but the password was wrong then you have leaked information to the user that they should not have access to. i.e. If email is used as a username and I go to a porn site and type in my wifes email address and a random email and it comes back 'username right, password wrong' then I now know she has an account there which she was trying to kepp secret.

It's not only bad from that point of view, but also the attacker now knows you have a account there and so could try social engineering techniques to gain access to your account, it basically provides a attack opportunity that it doesn't have to.

I think you're missing the point, which is that that information is already unavoidably leaked by the signup form. So leaking it from the log in form doesn't change anything.

Yep, the only way to really leak no information is to just always say that the action (sign up, forgot password) was sucessful, which is awful from a ux perspective. Don't forget to go through the motions of registration/recovery to stave off timing attacks as well.

I have actually worked on systems where the signup form allowed multiple users to have the same user name, and the log in process consisted of checking for the existence of a (username,password) pair. Needless to say, I regarded this as a major design flaw and fixed it (except for the existing non-unique users). But there is, for all the reasons it shouldn't be done, a situation where the signup form doesn't leak usernames.

That's a very binary way of looking at it. Depending on the sign up flow, it may be much more difficult to discover the existence of a username or not, than it is in the login form. (For instance, if the signup flow involves payment details before the username step.)

Are you going to enforce uniqueness of payment details? And if so, what part of those details must be unique? Will you require a unique credit card number?

If so, you will have to maintain either a bloom filter or a list of hashes for all eternity to ensure that nobody signs up twice with the same CC number. [Edit: actually, you could keep the scrambled CC data for a finite length of time if you're just trying to rate-limit, but still see below]

So suppose you've done all that. Are you going to do a test charge to see if the credit card number is valid before allowing the user to choose a username? If you don't, your efforts are wasted, since anyone can just use a CC generator (they're used for testing) to sign up for new accounts all day. If you do, you've probably driven away 99% of your users with your ridiculous signup process.

If you're using a third-party payment provider like Authorize.net or Braintree to store CC numbers, you'll be given a unique token from the provider when you add the CC number. When you add the CC you can also have the provider run and immediately void a dummy transaction to verify the card is real, too.

I was actually thinking of cases like MMO game accounts - often times the actual game account isn't created until after the game is purchased.

Leaking emails is bad. Leaking site-specific user-names seems like less of an issue, particularly when they're exposed anyway through simple use of the site.

This is true for some cases but not all. For example if you have a CAPTCHA on your registration form but not your login form.

Saying that using random words is better than using a string of unrelated characters assumes that password crackers don't know how to combine words.

There are about 180000 words in English (but don't assume everybody knows 180000 words). That means that for three words, you have at most 5.810^15 options.

For a password with mixed case and numbers, you have 62 different characters. With, 9 characters, you have 1.310^16 options.

I can understand that people would prefer common words to remember them, but they are not that much stronger.

The best password is the password I don't need to remember: generate long random passwords, and store them in your browser, in Keepass, LastPass or whatever system of your choice.

It depends on how many words you have in a string.

XKCD's oft-quoted comic uses 4 words. That would be 100000 times stronger than using just 3 words, and vastly superior to your example of a random 9-char password. The article's own example, "golf kangaroo crispy halitosis", also uses 4 words.

I agree with you that long random passwords are the way to go, but even in that case you need to remember at least one password: the master password to your password manager. It would be a good idea to make that a string of 4 or more words. My LastPass master password consists of 5 words with a bunch of symbols sprinkled in between, and my banking password is 4 words in a foreign language.

The problem is, if you actually choose your 4 words randomly out of the full dictionary, you won't get something like "golf kangaroo crispy halitosis" or "correct horse battery staple". It will sound closer to "capaciously endodermal remast amarantite". The set of words as familiar as "golf", "kangaroo", etc. is much smaller than 100k.

It doesn't have to be as memorable as kangaroo, it has to be more memorable than the equivalent (in entropy) number of random characters, which I think is basically any word.

$ wc -l /usr/share/dict/words

99171 /usr/share/dict/words

$ rl -c 4 /usr/share/dict/words | xargs -d\\n

contortionists mocking Alphard soling

$ rl -c 4 /usr/share/dict/words | xargs -d\\n

Toni's dish's mauled spillages

$ rl -c 4 /usr/share/dict/words | xargs -d\\n

expedited tireless interneships tranquiller

$ rl -c 4 /usr/share/dict/words | xargs -d\\n

bohemian rogering unkindliest ayes

How about mixing languages in the words chosen. English, French, Italian, Spanish, German, etc. Are they using multi-language dictionaries? Also, are they looking for typos? Voluntarily misspelling words would not make it harder to remember yet it might be harder to figure out.

I'd say jjjjjjjjjjjjjjjjjjjjjjjjjjjj, from a human-interpetation standpoint, has very low entropy, since there is very little uncertainty as to what will come next... This would be evident if password security checkers tried to compress passwords and compare them to the original message.

i.e. "jjjjjjjjjjjjjjjjjjjjjjjjjjjj" is 224 bits as a string. Huffman-encoded, you only need 0=j as a code to convert it into 28 bits, a compression of 87.5%. So this is a good way of actually checking the security of a password, if it seems to have high entropy.

>Simply put, adding length increases entropy more efficiently than replacing letters with symbols.

I don't like this meme. Because it's not necessarily "more efficient." Like most things, it depends. If I have a password made of 10 lowercase characters, it's much better to replace with an upper case character rather than add another lower case one.

Option 1) Add a lower case character. Password is 26 times stronger.

Option 2) Replace 1 upper with 1 lower case character. Password is 1024 times stronger.

What you say is true if attackers know the composition of your password: If the system provided them with some sort of hints telling them how many upper, lower, digits and special characters existed in the password.

Hopefully such systems don't exist.

Instead most attackers have to run through the obvious, then through brute forcing against the known character set.

But don't do what someone I know did and replace every single e with 3, every single a with 4, every single o with 0.

Identical entropy, extra theatre.

Not to imply that what you describe is a great strategy, but I wouldn't describe it as pointless. The password "12345" has the same entropy as "b0g4p" but it would be a mistake to think they are equally secure.

>The password "12345" has the same entropy as "b0g4p" but it would be a mistake to think they are equally secure.

Those do not have remotely equal entropy.

At the byte level it does, which is presumably what the OP was talking about when saying that replacing characters with digits (uniformly) does not affect entropy.

Maybe, but I would hope that's not what s/he meant, because that kind of entropy is basically irrelevant to password strength.

Is bogap a dictionary word?

What I'm describing are passwords like

T1g3rF33t Cam3lT03

etc.

All dictionary stuff along a theme, but made "secure" by applying a zero-entropy substitution of all occurrences of (L,O,A,E) with (1,0,4,3).

Given that crackers know people do this, they add them to their dictionary attack routines so this is no more secure.

Given I'm still fighting with them over, "don't store passwords plain text" I've not even begun to attack them over this practice yet.

Ha no...I just randomly typed in symbols. Guess it does look kind of "wordish" though.

I don't think that's identical entropy, since the usage of that strategy varies, and the non-substitution case is more common. But it probably adds less than one bit, and last I checked, 2 * instant = instant.

This is a bit of a fluff piece, isn't it? All five points are "true", to some degree, but not when you consider the details.

Take point #4:

> "4. MYTH: Two-Factor Auth Means Any Password Will Do" > > This is absolutely not a dismissal of multi-factor auth. Use it when it’s available! Two factor is on the Stormpath roadmap, and we use it for many backend systems. Just don’t consider it a substitute for strong passwords.

The example that the OP blithely cites is Cloudflare's incident in 2012, when CloudFlare's CEO explained how his GMail account got compromised despite activating 2-factor authentication. However, if you actually click through the post, you see that CloudFlare's problem had nothing to do with a weak password, but a flaw in AT&T's authentication procedures which allowed a hacker to bypass GMail's password/auth requirement with a little social engineering. It had nothing to do with the CF CEO's password, which he says was strong and 20-characters long. In his case, it wouldn't matter if the password was 8 characters long and a dictionary word, if 2-factor auth worked as it was supposed to.

http://blog.cloudflare.com/post-mortem-todays-attack-apparen...

And that's kind of the whole dilemma of security, isn't it? That certain maxims ("Your password should be x characters long and contain y different kinds of characters") do not universally apply, and that the implementation of security protocols is extremely important into understanding your potential to be hacked. The OP's mythbusting is somewhat counter-productive here.

Fist myth is that passwords provide security. The human brain is crappy rng so it cannot create proper high entropy passwords. No way getting around that - whatever your password is if it is "derived" from something - someone else is using it too.

The only viable security I can think of is if we have distributed public key infrastructure. Bitcoin style wallets. And securing the "PK wallets" is up to the end user. Persona is close I think - so if the app fad/insanity finally dies we will be on a good track.

>The human brain is crappy rng so it cannot create proper high entropy passwords. No way getting around that...

Yes there is: you use a real RNG to generate your password. A good option is http://passphra.se. Just remember that you can't go clicking "Generate Another!" until you find something particularly memorable, or you'll be sacrificing an unknown amount of entropy by narrowing the space to memorable passwords. Better to keep the phrase somewhere safe (encrypted) until you have managed to drill it into your head. I like to fit the phrase into a sentence, which adds both memorability and entropy.

For example, it just gave me: "song surprise calm task", which I might turn into "Sing a song of surprise, and resume your calm task.", which Dropbox's zxcvbn tool estimates at 141 bits of entropy. If a password form has a character limit, you can initialize the sentence, yielding "Sasos&ryct", which zxcvbn rates at a not-too-shabby 52 bits.

If we assume that English has 65535 words a four word password gives you 64 bits of entropy. Of course the real number of words that you can use for mass generation is 2k (or even less) if we search words that everyone that is standard deviation or two below the mean on the vocabulary graph can use and spell correctly. Spelling is going after the ability to memorize telephone numbers too with everyone using the crutches of autocorrect and other spell checking tools. The moment everyone uses that scheme the people interesting in reversing passwords will move in that direction too.

This is fake security trough obscurity. You can get pretty secure passwords nowadays by just concatenating your username and site domain and md5-ing them and taking the first half. It generates awesome passwords ... until you are the only one using this way of deriving passwords.

The human brain is not suited for remembering long complex passwords. Even less suited for unique for every service.

Four words chosen from 2048 possibilities gives you 44 bits of entropy (this is what passphra.se does). If you're being specifically targeted in an offline attack, that won't hold up. But in that very rare scenario, you are already in trouble. A 44-bit password will protect you from attackers trying to harvest from a big pool of stolen hashes, or from attackers bruteforcing the login form on a website. That is not the same thing as "fake security through obscurity".

As for the problem of strong, memorable passwords that can stand up to offline attacks, there is still a gap (a 253 bit password made of 11 bit words would be 23 words long), but I don't think the problem is unsolvable. A project I am interested in working on when I have time is one that produces very long, but memorable high-entropy passwords with rigorously quantified entropy by generating evocative, grammatically structured sentences, using a carefully selected word list and a set of mad-libs style templates.

So to get a ~256 bit pass phrase, you might be left to type out a 50 word poem. It would take effort to memorize, but it would be reasonably doable, whereas memorizing a random 39 character password is extremely difficult for most people.

Moreover, since you'd be working with a limited set of words and grammatical constructs, the password entry form could have robust auto-correction so that typos and spelling errors wouldn't be a significant problem.

Complexity Trumps Length

Question for the smarter people out there. If everyone started using passwords like "donkey computer watch" would this mean that this format would then be much easier to crack as opposed to xy7*hdkSD

Security is a bit of game theory right?

It actually wouldn't, because the entropy doesn't rely on the attacker not knowing the general format - it relies on the number of words available as choices.

English has roughly 100,000 words, so that password scheme has 10^15 combinations. Suppose the latter scheme incorporates the 96 printable Ascii characters. log96(10^15) is 7.5, so yes the latter scheme gets ahead on entropy at 8 characters or longer.

But it's not a question of maximizing entropy. It's a question of maximizing entropy relative to the human brain's ability to remember it. The brain handles words as single units, so it's a lot easier to remember three of them than eight or more arbitrary characters.

I go by the xkcd standard of 4 words, though that doesn't change the math, just the numbers.

Doesn't the entropy just measure the difficulty of brute forcing a password though? From an entropy perspective, "password" may be more secure than "r&E2pX@", but I doubt anyone attempting to crack passwords in a practical manner is going to adopt a direct brute force approach.

With advice like this seemingly contradicting other seemingly sound advice and the debate taking place here at HN it's no wonder regular people (and geeks) have trouble deciding what is and isn't a secure password.

Another myth: 20% of root passwords are now exactly "CorrectHorseBatteryStaple"

I like how Myth 1 tells us long passwords are better than complex passwords then Myth 3 complains that Microsoft rate the long but non-complex password "jjjjjjjjjjjjjjjjjjjjjjjjjjj" highly

They aren't contradictory. While this is, as someone else stated, a "fluff piece", neither b4#L or aaaaaaaaaaaaaaaaaaaaaaaaaaa are necessarily secure passwords. It's a gradient.

Having said that, does any cracker even try for something like jjjjjjjjjjjjjjjjjjjjjjjjjjj? If I were trying to brute force password, I would think that such a password would be relatively safe simply because it's such an irregular pattern.

>Having said that, does any cracker even try for something like jjjjjjjjjjjjjjjjjjjjjjjjjjj? If I were trying to brute force password, I would think that such a password would be relatively safe simply because it's such an irregular pattern.

Here's my response to that: https://news.ycombinator.com/item?id=5546741

tldr: I could try all repeating characters so fricking fast it would be worth it to build into a cracker.

tldr: I could try all repeating characters so fricking fast it would be worth it to build into a cracker.

In the context of this discussion, of course we'll all say that. Does any actual cracker program try for that, though? Not to my knowledge.

Sorry for the back of forth here, but I'll try to make this the last comment:

Some people are using all repeating characters if they can get away with it. ESPECIALLY because articles like this keep hinting that they are decent.

It doesn't matter how small this percentage of people is. The time it takes to check against it is infinitesimal and the reward is nonzero. If I were building a cracker, I would do it and not just because it's the topic of discussion.

Clearly I'm not making the argument that jjjjjjjjjjjjjjjjjjjjjjjjjjj is a recommended password. Of course it isn't.

Use a long, complex password. Period.

However practically we know the history of password crackers, and while it's easy to look at specific examples given and say "Oh I would totally crack for that", in most general cases they wouldn't be cracked. On the flip side, passwords that are dictionary words following by numbers and or punctuation are cracked because they are exactly what people migrated to once complexity rules were implemented.

> If I were trying to brute force password

People tend not to bruteforce passwords. People tend to use dictionary attacks. I don't know if any current wordlists include 'jjjjjjjjjjjjjjjjjjjjjjjjjjj', but it'd be easy enough to create a wordlist that is just repeated characters.

> It is also important for users to use random text in passwords, [...] repeating characters [...] are expected – the opposite of random.

Reply on news.ycombinator.com