5 ms·
Couldn't password databases be implemented as hardened "appliances?" This wouldn't have to be sold as hardware, it could just be an install. The machine would
by stcredzero 14y ago
Couldn't password databases be implemented as hardened "appliances?" This wouldn't have to be sold as hardware, it could just be an install.
The machine would only have the function of storing and verifying passwords, and secure communication with authorized clients. All apis could use fixed-length fields.
Passwords themselves could be stored using a modified salting technique using a white-box version of block cipher. This would make it much harder for attackers to crack the password database.
The block cipher used for the modified salt could also be implemented by separate hardware, which would make it much harder to crack the password database.
- eli 14y agoThat seems like it would be just as much work (and just as prone to errors) as building it right in the first place.
- stcredzero 14y agoNah, you just a) have to make the call to an authentication function twice (first call sends the old hash hashed again into to the new system, then if that doesn't auth, call the old system), b) set up a migration that inserts the old password hashes into the new system. The new system hashes that data for you.
- lawnchair_larry 14y agoI think if you care enough about password security to know these appliances exist, you already know how to use a proper password storage scheme. The ones that we are seeing dumped were not even trying.
- derefr 14y agoWhat the parent is talking about is basically a solution to the http://en.wikipedia.org/wiki/Principal%E2%80%93agent_problem http://en.wikipedia.org/wiki/Principal%E2%80%93agent_problem. I don't think these password compromises were likely failures of management to say "let's build a secure system"; they were failures of the engineers told to "build a secure system" to know and care enough about building secure systems to learn about things like bcrypt/scrypt in the first place. The solution to this, of course, is to make password security into its own industry, with its own principals who are incentivized to understand every facet of secure systems. This would resolve as the parent explains--third-party vendors selling these companies multi-million-dollar "password appliances" with installation, support contracts, and all of that brouhaha--and then bringing in their support engineers to teach the company's own people how to securely call into the appliance.
- stcredzero 14y agoI don't think these password compromises were likely failures of management to say "let's build a secure system";... The solution to this...would resolve as the parent explains--third-party vendors selling these companies multi-million-dollar "password appliances" with installation, support contracts, and all of that brouhaha--and then bringing in their support engineers to teach the company's own people how to securely call into the appliance. Screw that. Just use a specific API in BSON structured with fixed-length fields to make the call. Build the appliance as proprietary software to install on top of an existing secure Linux distro. Sell to small sites for $1000 a pop or so, with different licensing for virtual server and cloud hosting companies. If Big Co. needs serious bandwidth, then they can do a project. Hardware component: that we'd need VC funding for. Hardware is just harder.
- lawnchair_larry 14y agoMy position is that the engineers were probably told to "build a system". It's still uncommon for security to even be a thought, sadly.
- viraptor 14y agoHow many services use ldap with per-user controlled access, kerberos or radius for authenticating their users? Yeah... If we're not using what's available now, why would people start using specific appliances? I know the list here has a wide range of possibilities and doesn't map exactly to what you describe. What I mean is that we can do things 10 times better than we do them now with off-the-shelf (or even, off-the-github) products right now. Yet there's still a lot of services authenticating by `SELECT ... WHERE user='" + $user + "' AND pass='" + $pass + "'`.
- stcredzero 14y agoYeah... If we're not using what's available now, why would people start using specific appliances? Convenience? If it was possible to just plop down $1000, redirect some calls, setup a migration, and be done with it in an afternoon, more people would do it.
- Lukeas14 14y agoI don't think the problem is that storing passwords is too hard, at least not for the majority of startups not dealing with PCI compliance. Setting up a user system with bcrypt/scrypt is pretty trivial. The problem is you have several apps that were built before "just use bcrypt" became a meme and switching to bcrypt or some kind of hardened "appliance" just isn't seen as a priority, especially if it would require changes in several legacy systems. It's like telling people to backup their databases. Everyone knows they should do it. It's not hard. But it doesn't directly bring in revenue so therefore some people will inevitably just never get around to it (until they get get hacked and it becomes a PR issue).
- pbreit 14y ago_Beginning_ to use bcrypt/etc may not be that hard but migrating existing users from some other scheme to bcrypt/etc takes a bit more thought and planning. And the "use bcrypt" crowd _never_ provides any guidance on that.
- jleader 14y agoI've heard 2 ways to do it: 1. Add a flag column to your password database; whenever anyone signs in without the flag set, encrypt their password the old way (for checking against their old entry in the database) and the new way (for storage in the database) and set their flag. The only trouble is, users who never log in don't get migrated. 2. It's possible to treat the old encrypted form of the password as if it's plaintext, salt that, encrypt with bcrypt, and store the result in the database. Then password checking becomes: encrypt password with old algorithm, salt, encrypt with bcrypt, and check against the database. It lets you convert everyone at once, but makes the login code a little more complicated, forever. If you can't convert everyone's database entry at once, you might still want a flag column, so you can migrate more gradually. There are still all the usual headaches of QA, release management, etc., but presumably you already know how to handle them. What else am I missing?
- stcredzero 14y agoThis would be a good selling point for companies once they have a little funding. Save yourself some embarrassment and plop down $1000 for a properly secured password db. There's also a strong selling point for the sites that have gotten hacked.