Hacker News Reader

My favorite bit: "Passwords may contain no more than two identical consecutive characters in any position from the previous password".

How would they know that unless they have your previous password in plaintext?

Security Fail

Not to mention they know the names of your pets :-)

They could store a hash of your old password, along with 40 hashes, one for each character, hashed against the hash of the old password. e.g.

'this_is_my_password" => hash1 'hash1_t' => hash2 'hash1_h' => hash3 etc

This way, when you enter the first character of a new password, they can hash it against your old password, see if it creates hash2 and refuse permission to use the new password. Hopefully this is the same way as websites can ask for only selected characters from a password when you log in.

That's functionally equivalent to storing the password in plaintext. The individual letter hashes would be trivial to reverse.

It would be _easier_ - you still need to know the hashing algorithm and salt.

I suppose since it's the TSA, they may use encryption and an HSM instead of hashes.

Kerckhoffs's principle: A cryptosystem should be secure even if everything about the system, except the key, is public knowledge.

So you can assume the algo is public and the salt too (assume someone has a dump of your source code and your database - how do you keep the salt secret? You can't. This is the case where all the effort to protect passwords and use strong hashes is aimed at.)

The solution of "enter your old password" when you enter the new one is simple and doesn't compromise security.

I agree, but "password is not secure if everything about the system, except the key, is public knowledge" is better than "password is saved in plain text".

Should we also be worried then about Facebook, which also stores previous passwords?

If your Facebook password and banking passwords are the same, or you believe your Facebook password is actually protecting anything you do there, then yes, you should be worried.

I'm curious what you mean by that. I work on security at Facebook (and actually help with the system that detects malicious logins), and think we have a pretty good / secure login system, even compared to a lot of banks.

The odds of a phisher who knows a bunch of valid Facebook credentials getting into any significant percent of the corresponding accounts are pretty low.

EDIT: If you want more protection than this, you can also turn on 2-factor authentication for your account, with "Login Approvals:" https://www.facebook.com/settings?tab=security&section=a.... I don't think I have the ability to do this at login time with my bank (though they do offer / require it when taking high-risk actions like initiating transfers).

All that being said, you're absolutely right that sharing passwords across sites is a bad idea.

Thanks for the two factor info and link. I wasn't aware that was available.

Glad it's helpful! If that's too hardcore for you, check out "Login Notifications" as well (on the same page). This sends you a text whenever someone logs in from an unrecognized browser (but allows the login).

Alright - It's nice to know that Facebook has an industrial strength login system. Nobody would know that unless, like here, someone like you chimed in to mention it. (And thanks for that)

My point was more that - as a user you should not expect a site like Facebook (or Twitter or FourSquare) to spend as much time firming up their login system as you would your bank.

The fact that you actually do so is a nice bonus. But users would be well-served by not expecting the same level of attention to security from social sites and treat them accordingly.

(And sorry if I came off as being unfair to Facebook. I meant my comment to apply to that genre of sites in general)

It's true that a lot of sites don't go to the trouble of implementing particularly secure login systems (it's a lot of work!). Sadly I think this includes many banks.

One way around this is using SSO with a site who does spend a lot of resources on making their login system secure. For example, using Facebook Connect is a great way to get all that security (plus all our fake account detection) and the added bonus of not having to store (and properly hash) passwords. If only RockYou, Gawker, Sony and others whose credential databases have been compromised had done something like that... Fortunately, we have an answer even to problems like that with 3rd parties: http://bucks.blogs.nytimes.com/2012/04/02/how-facebook-tries.... I've spent multiple nights up late trying to crack hashed passwords on a big dump before attackers can, so we can help the victims who shared credentials between the compromised site and Facebook secure their accounts.

I'd love to raise awareness around the stuff we do to protect users. We've done blog posts and people have written articles, and users often see this stuff when e.g. they're traveling (ever been asked to identify photos of your friends?). What else do you think we could do?

Here are some example articles about what we do around login security - if you're interested, check out http://www.facebook.com/security to see other related stuff. https://www.facebook.com/note.php?note_id=10150172618258920 https://blog.facebook.com/blog.php?post=389991097130 https://www.facebook.com/note.php?note_id=425136200765

For example, using Facebook Connect is a great way to get all that security (plus all our fake account detection) and the added bonus of not having to store (and properly hash) passwords.

With the downside of compromising your users integrity and open your users up to even more tracking as well as relying on a 3rd party for a system-critical component and force your users to sign up for a facebook account.

Storing and properly hash passwords is not hard to do.

While I appreciate the work you do, facebook connect (as well as most, if not all, similar solutions) just sickens me and any service that exclusively rely on it is just pathetic (for the reasons above).

Todo: Create multiple fake facebook accounts for such services, can't believe I haven't done this already.

If you would just care to acknowledge that there might be potential issues with the things you mention your posts wouldn't seem like pure PR-statements.

> With the downside of compromising your users integrity and open your users up to even more tracking as well as relying on a 3rd party for a system-critical component and force your users to sign up for a facebook account.

Obviously using Connect means that there's a 3rd party with some data about your users. IMO that's worth the tradeoff, but if you feel differently, then don't bother with Connect (though I'm curious what concrete concerns you have, more specific than just "opening them up to even more tracking").

As far as reliability goes, I agree that minimizing external dependencies is a good idea. Again, this is a tradeoff - you're saying "I want to work on things that are specific to my app or site, not this boring login stuff that has already been solved." I suspect that in reality, our uptime is good enough that sites are more likely to notice issues from their own (or their hosting provider's) instability than from ours. But you're right - this does introduce the possibility of login being broken for your site if Facebook is broken. Hopefully the benefits Connect offers outweigh this risk, but again, you're fee to not use it.

> Storing and properly hash passwords is not hard to do.

Perhaps it's not that hard, but many still people fail to do it properly. And it's still not a silver bullet - even properly hashed passwords, when exposed from a compromised DB, can be reversed. Regardless, this is only a small part of the problem - dealing with stolen (e.g. phished) credentials is another, arguably more important, problem that's much more difficult.

> Todo: Create multiple fake facebook accounts for such services, can't believe I haven't done this already.

Please don't do that. We're pretty aggressive about proactively seeking out fake accounts and getting rid of them, and as a result I imagine this would end up being a pain in the ass for you. I'm curious what threat this mitigates?

Obviously using Connect means that there's a 3rd party with some data about your users. IMO that's worth the tradeoff, but if you feel differently, then don't bother with Connect (though I'm curious what concrete concerns you have, more specific than just "opening them up to even more tracking").

As a user: My integrity. I don't want to link accounts on otherwise completely different services. You have no business in knowing what other services I use and I see no, absolutely no, advantages to being forced to use facebook connect. As an alternative? Sure! But exclusively? No. You fail to mention the advantages with using facebook connect so I would really hesitate to call it a tradeoff when you have nothing to gain...

And as a user I'm not free not to use it... And facebook haven't exactly had a solid track uptime record, it has even deserved a poor reputation in that regard - sure it isn't that bad but thing with relying on external parties is that it can only get worse than handling it yourself.

http://www.pcworld.com/article/193423/facebooks_sneaky_apps_...

Perhaps it's not that hard, but many still people fail to do it properly. And it's still not a silver bullet - even properly hashed passwords, when exposed from a compromised DB, can be reversed. Regardless, this is only a small part of the problem - dealing with stolen (e.g. phished) credentials is another, arguably more important, problem that's much more difficult.

The problem isn't, and has never been, that it is hard. The issue at hand is incompetence in its purest form and nothing else. And such incompetence will bite you in the ass regardless. Sure, I'd prefer my password not leaking but just deriving your password from the domain name or something like that is simple enough to combat that. In any case that is a tradeoff that isn't hard to make.

External dependencies is not only something a developer has to take into consideration, as a user you have to keep track of permissions, ever changing security settings - but for what? If I don't want to link my music usage to my facebook account what possible can I gain from using facebook connect? Seriously? Have you guys even considered that scenario? Of course you have, you just couldn't give a shit about your users privacy or concerns. Rather make an easy buck than creating a good platform, just don't pretend you try to do both.

You can't seriously argue that having to deal with all that shit is easier than signing up for a new account for every service.

Please don't do that. We're pretty aggressive about proactively seeking out fake accounts and getting rid of them, and as a result I imagine this would end up being a pain in the ass for you. I'm curious what threat this mitigates?

Even spotify recommends you to create an empty account ;) There is nothing you can do to stop me and the huge relief for my ass would be to not being afraid of spamming my friends with my spotify usage, not having to even care how spotify/whatever uses my facebook account (and how facebook/services will handle this in the future, facebook has a terrible reputation of adding new options with questionable defaults). Is it so hard to grasp that these are valid concerns? Valid concerns that should never, in anyones wildest dreams, ever have existed.

There is nothing of value you can offer me but there are countless ways you can annoy the shit out of me. Solution? Create dummy facebook accounts. It is the only solution to the problem that you have created - suit yourselves.

I don't trust facebook and I don't trust anyone else having any form of access to my facebook account (also, I'm so old fashioned that I don't even run executables that I get in the mail of an unknown sender). Anyway, thanks for making the web a worse place!

If you ever find yourself working for someone other than Zuckerberg, please note that integrity issues is part of security as well.

We only store hashes of your previous passwords - we never store them in plaintext. We don't have the ability to enforce a silly requirement like "none of the n grams in a new password can be the same as those in an old password."

Why not? Doesn't the user have to enter both the old and the new when they're changing passwords?

Sorry I don't understand, but why not what?

Often times users change their password because they've forgotten the old one, so we cannot ask for their last password in that case.

In other cases, users may have a number of previous passwords (some of which we know to be compromised). We want to check against these lists as well, and asking the user to enter all their previous passwords is obviously unhelpful. They also are unlikely to remember which ones are compromised (because e.g. we detected an unauthorized login from them and the user confirmed that it wasn't him; or because we found them in a publicly-available list of credentials from a phishing site or 3rd party database breach).

That requirement actually isn't as rigorous (or as difficult to implement) as it sounds. The keyword there is "password" instead of passwords. Based upon similar systems, I feel comfortable that this is not a typo.

Whenever the user changes a password, the user has to provide a (single) previous password. This password is hash-checked against the current user password. Then the password duplication rules are applied.

(Note, none of the above should be construed as an indication that I agree with this approach to password policy :) )

Very good point. If they make you enter your previous password as you change it, my criticism is invalid.

Whenever the user changes a password, the user has to provide a (single) previous password.

What about if you forgot your password? (which I feel is the scenario I usually change a password in) Do they just ignore that rule in that case? If so, whats the point of having the rule?

The fact that you know your password is what identifies you as the legitimate owner of your account. If you forget your password, you can't identify yourself as the legitimate owner of your account and thus, to the eyes of the system, you aren't. And about the only person who should be able to change the password of an account is its legitimate owner.

So, to answer your question:

Do they just ignore that rule in that case?

That case should not even happen in the first place.

That's ridiculous. It is extremely easy to forget your password, that situation does happen and resetting passwords should be catered for in the system.

Your premise is basically false. A system administrator should be able to reset a password.

I've just intentionally ignored that case in the past. :-) Ie, sending a password reset email will bypass the check, just as you suggested. There are some other things that people can (and I'm sure do) do to try to meet the requirement, but there isn't really a solution. For example, a lot of people who hit this restriction are just trying their old password with one or two characters appended, so you can try the hash of a few substrings of the new password.

But in the end, none of that really keeps people from inventing new, easy to remember, and insecure passwords. Eventually people do learn to just alternate between about three different forms of passwords. IMO, most of these draconian policies are invented to make things look secure more than to provide real security. That seems appropriate for the TSA, right? :-)

Don't be so quick to judge.

Believe it or not, it is not so uncommon for LargeCorps to store passwords in a way they can be retrieved again at least in one place - which is usually their central identity management system. This helps them enforce password policies and expirations across a wide range of systems and gives them better control and overview of all the many user accounts lurking around on all their systems. And obviously being able to make sure all accounts of a user are properly locked or deleted is more important to them than giggling at centrally storing potentially tens of thousands of passwords.

The system I know and maintain stores your current password and history in an encrypted form and protects them through various means, so if you got a full DB dump you would still have to brute-force. Works pretty well for us and the advantages seem to greatly outweigh this downside of having passwords stored somewhere in a retrievable form. Plus in typical LargeCorp you have (more or less) sophisticated logging and monitoring tools on top of that so a lot of suspicious activities are detected pretty well. And there are a LOT of other LargeCorps I know of which are using that same product and on a global scale.

So, such a system is definitely overkill in ye-old startup webshop, if only for the insane bucks it costs, but for LargeCorps with way too many systems piling up it can be very valuable.

5. Passwords shall not contain any dictionary word.

Bear with me as I close my eyes and hammer wantonly on my keyboard:

kjhjks;dfijwefasdkfj

Aha! First try, and I've already generated an insecure password containing five dictionary words. Until vowels are banned as well, I fear our national security is in peril.

did you actually check this against the unix dictionary? I only found: ['as', 'fa', 'fi', 'we']

No, just sort of boggle-eyed it. I was counting "fas" as well, as the plural of "fa".

What about 'i' and 'a'? So that's six (or seven including 'fas')

Oh-- the unix dictionary includes every letter as it's own word for some reason. So I required words to be length of >= 2. Probably my mistake.

This looks exactly like the Dept of Energy password guidlines when I worked there.

For item #5, you should be aware that they check for dictionary words of 3 or more letters, both forwards and backwards. I'm not sure what dictionaries they were using, but I found that most of the Japanese, Russian and Finnish words/phrases that I frequently pick from were forbidden.

What, precisely, is bad about that list?

There's a problem that people when faced with a dialogue box asking them for a password will be unable to think of a strong password, but that's solved by giving them some form of auto generation.

Writing passwords down is not a security problem, if the slip of paper is kept safely.

The link I provided gave some pretty good reasons, I thought.

The linked article addresses what, precisely, is bad about that list on a point by point basis. It's good reading!

Most web browsers now have built-in password managers. Here's to hoping that loop is soon closed with password generation!

That's what LastPass or OnePassword or KeyPass is for.

None of them come with the browser.

Because when it comes to changing passwords, it would take me about 45 minutes to come up with a new password that met the rules and had a reasonable chance of me remembering it. I took to writing them down as I was trying to come up with something, and keeping the paper handy for a few days because I'd frequently forget the passwords by lunchtime and lock myself out of the system.

One thing to remember about these rules is that there are some other ones not publicized. All you know when it refuses to take your password is that it says "password doesn't match" instead of "password doesn't meet the rules." I got some different error message when I tried using Russian obscenities, but I forget that error message.

If nothing else, the list itself is long and intimidating.

Wow ... their online security rivals their airport security!!

With all these restrictions, I wonder how far they are reducing the possible key space? Some of these suggestions actually remove a number of permutations - like for instance you can't put in two digit years - this must remove a HUGE number of combinations!

Then you remove ALL the dictionary words, all proper names, all the TSAs serial numbers, all reversed words - well, that's gotta make cracking an 8 letter password easier :-)

But most of those will be common passwords and already included on password lists.

password72 is not more secure than password. username84 is not more secure than username.

Not quote my point. What I'm saying is that you make an assumption that 8 letters are used due to the ridiculous complexity requirements. Then you get all the possible combinations of letters and numbers for 8 character passwords. Then you remove all dictionary words and reversed words, all dictionary words (and reversed words) ending with a year, all combinations with repeats, all serial numbers and known patterns (like ABC). Then you further remove from this list all the combinations that don't have mixed case, a number and a special character.

I'm interested in knowing how many possible passwords are left?

Getting your hands on the 1300000 common passwords and not allowing them would be a good start.

Do you not see the irony of talking about 1.3 million individual common passwords?

No.

Using 94 possible characters for an 8 character password means there are 6.09568939 × 10^15 possible passwords. Removing even 20million means you still have 6.09568937 × 10^15.

Yes, and seeing the message that the password is already in a database used by cracking software should give users pause.

I actually did an analysis of that once but I can't seem to find it now. There actually is a pretty big reduction of keyspace (in the billions) but most of it is keyspace you don't want anyway. That is actually why the best policy is to have minimum length requirement of 12-14 characters and nothing else.

Yep, they won't even let you use "fingernail clippers" as a password.

> My first thought here is that if you are going to have such an extreme password policy, you could at least set the minimum password length to ten or even twelve characters

Well, and then in reality you have a couple of those old 40 ton host systems lurking in the basement and all its former tamers are long gone, everyone is just glad it is still running and somehow it only allows exactly 8-char passwords with characters and digits only... and somehow it still is one of the most fundamental systems of what you are doing on a day-to-day basis so if you are introducing a central password policy and maybe even a central identity management (which I assume those huge departments are using) then you have to make the bitter decision of accommodating for the "old timer" systems as well or leaving out the proverbial 20% of your systems that see 80% use each day...