I think the author is missing the real problem. Why is there a single "admin" account at all? Why don't users log in with their "normal" user accounts, and then use some kind of authenticated, audited privilege escalation (like sudo, for example) to perform tasks that require administrative privileges?
Seriously. I almost stopped reading after this bit...
'specially for infrastructure accounts (if your company uses SSH, chances are you have one Unix Login that all your admins/employees share). Which makes non-repudiation harder.'
Chances are???? What credible sys admin would ever do something like that...?
Oftentimes, the decisions admins make are determined by the needs of management of fellow employees.
I have on frequent occasions mouthed words along these lines: "It's a bad idea to do it that way, but I can see how it would be much cheaper, efficient or easier to teach employees, so I will tell you the best way to do this bad thing."
You can carefully set up the best security possible, but the instant that a client or bigwig is waiting on something because of it, credentials will be shared and you or other employees may be ordered to share them.
A lovely anecdote: When I worked for Bellsouth.net, the ISP, our router credentials started with a shared admin/(password) pair. The change happened, according to speculation, because of a breach that couldn't be tracked back to a specific user.
If a company's smart, it'll only take one such complication to change that behavior. That behavior barely made sense in 1999 though and any company using shared super-user privileges in 2013 is just asking for a world of hurt...
The idea that a serious compromise will present a clear path back to a specific ssh key that got used by the attackers and that you'll possibly be able to stop it just by turning off that key is pretty laughable. But then again, so is protecting your core infrastructure with 1.5 factor android soft tokens. Google isn't even willing to make it sound like especially strong protection for your gmail account. How much for a CAC style pki infrastructure? Hard to believe it's more than $50-$100/seat for a small organization. If you're worried about figuring out which employee got his phone dropped after your whole backend got molested perhaps an actual security posture would be more suitable.
What percentage of those android phones would you say are upgraded to a level where they don't have any publicly announced cve's against them that allow for rce or close enough? Like 5 or 10 percent? I agree that it's better than a single secret, but how does a soft toekn count as "something you have" if it can be stolen from your phone and not end up "missing"? My google auth secret continued to work without a hiccup after apple repaired and wiped my phone and i restored from their cloud backup service. That's not too bad for keeping my voice mail private, but it's a pretty weak protection for sudoers on boxes that are pretty much critical to your company existing.
> Which brings us to shared accounts. Having worked with hundreds of companies over the past couple of years, we've learned that shared accounts are ubiquitous, specially for infrastructure accounts (if your company uses SSH, chances are you have one Unix Login that all your admins/employees share). Which makes non-repudiation harder.
I've seen this first hand at a number of large companies for both unix accounts and (even more so) with database accounts. After some period of time (varies from 1 month to 1 year) everybody who worked on an application would have the database credentials of the application and would use them to diagnose app issues. New members of the team for an app would be given all the credentials too as "it's the only way to get things done...".
It wouldn't start out that way as things would be "locked down" at first but every app ended up like that. Add to this that folks tend not to ever rotate their application database passwords for fear of breaking something and you've got a ton of people with the same access credentials and no way to distinguish them. People would join/leave the group (or even the company) that managed an app and would forever have access to all the databases for the app[1].
This is one of the problems that we're solving with JackDB[2](disclosure: I'm the founder). It allows you to share data sources without sharing passwords. User's authenticate as themselves and the server proxies the database connection for them. All executed commands (SQL, etc) are added to an audit trail and since you can tie actions to actual users you know who made changes, who queried customer records etc. Similarly since access is done per user you can selectively grant/revoke access to user's without asking them to "forget" old credentials.
[1]: Yes firewalls can block things as well but if you're still an employee somewhere then you can probably get passed the internal firewall by tunneling through some other server you have access to.
This is the role of configuration management software. Applications should not have hardcoded passwords and all decent configuration management systems allow for controlling access.
I think the core issue behind this is the complexity of managing SSH keys across your infrastructure. Have a look at https://bastio.com
JackDB: I like that the service keeps an audit trail on database transactions transparently, it is a big thing for companies that have to be compliant to a set of standards (i.e., health sector, financial sector, etc.) Good job!
Thanks. I've previously worked in technology in the financial services industry so a lot of the ideas for auditing/compliance come from meeting the expectations of that market. Our logic is that if it's seamless enough for them to use it and satisfies their compliance requirements, it should be good for most other industries as well.
Regarding SSH, it has the advantage that you can (and should!) have multiple authorized keys for the same account. For example for the bastion host to connect to our VPC of servers we use individual SSH keys tied to the same proxy account. Databases generally don't allow that type of access.
Some allow using client certificates to login (which nobody uses). Some allow integration with external LDAP/AD (which very few use). The vast majority uses standard username/password based authentication that hopefully is going over SSL (seriously go and check ... for most DBs by default it's sent over plaintext).
Some places try to have per user database accounts (possible tied to AD/LDAP) but that usually devolves to the same app password sharing as the privileges aren't kept in sync or it's just too much of a pain to provision vs just giving someone the app password.
I couldn't agree more; client certs are too much work and come at a non-trivial maintenance cost, same goes for AD/LDAP. So JackDB also acts as authentication provisioning front-end for a DBMS so users don't have to do that themselves and suffer the maintenance cost/overhead. Interesting.
I'm curious, your VPC bastion; how do you make managing SSH keys scalable given that setup? And why not separate accounts which in fact help with keeping sane logs of who did what, and distribute the risk of compromise across several accounts?
> So JackDB also acts as authentication provisioning front-end for a DBMS so users don't have to do that themselves and suffer the maintenance cost/overhead.
Yep. Couple of mouse clicks and a user no longer has access to one or more data sources (grants are role based). Similar effort for provisioning new users.
If the level of effort for a secure system is on par with emailing a password to a co-worker then people will actually use it. Plus with JackDB nobody other than the system admin even has the DB passwords so a normal user can't even send an email out if they wanted to.
> I'm curious, your VPC bastion; how do you make managing SSH keys scalable given that setup? And why not separate accounts which in fact help with keeping sane logs of who did what, and distribute the risk of compromise across several accounts?
At the moment things work for us as we're a very small team. Obviously something more scalable would be needed once you have a large number of people involved. The single proxy account is for convenience. That way the SSH keys to log into the other servers (accessible only through the bastion) can be shared. All servers (bastion included) log the public keys used to connect so we do have a trail of who/what/when. It would just be a bit more involved to filter it. Given our relatively small size it feels like a good tradeoff of security and convenience.
As we grow it's definitely something to (repeatedly) revisit, as is true about any security procedures.
Why do so many people need access to this 'admin' account. Is this an infrastructure problem? I'm struck by the idea the author is discussing how to properly share access to a single account on an OS that was conceived as multiuser.
We released Authy-SSH last september and this was by far the biggest feature request. You will be shockingly surprised if you knew the names of the companies that not only share unix accounts (which is very common) but were actually sharing Two-Factor Authentication keys.
Does obtaining the original Google authenticator QR code actually let you impersonate the authenticator? I was hoping it was a one-time shared secret only used for the authenticator to identify itself to the server for bootstrapping purposes and that a used QR code would be worthless.
> Send it to the email account that will be protected by two-factor authentication using the key, and it won't really matter.
Unless, of course, the email is intercepted in transit.
If your email is being sent in the clear, it doesn't matter whether or not you use SSL and/or two-factor authentication to connect to your server - the email has been compromised long before.
>> Send it to the email account that will be protected by two-factor authentication using the key, and it won't really matter.
> Unless, of course, the email is intercepted in transit.
> If your email is being sent in the clear, it doesn't matter whether or not you use SSL and/or two-factor authentication to connect to your server - the email has been compromised long before.
This isn't really as big of an issue as people make it out to be. TLS is fairly prevalent for MTAs.
Anyway, I assumed this was describing a scenario where a sysadmin/IT guy sends a QR code to another employee, in which case it is all internal.
I like the setup since it doesn't require my phone or other device to be connected to the internet. Also, "bootstrapping" would be problematic since I have 2 or 3 devices that I keep my codes on. I switch between them often and I don't always carry the same device with me all the time.
Doesn't non-repudiation mean the opposite of what he is saying? I.e. isn't "being able to proof where something comes from or who did what" just repudiation?
> isn't "being able to proof where something comes from or who did what" just repudiation?
No, "repudiation" is (in the relevant sense) denying responsibility for sending data or some other action. Non-repudiation is the property of a system that prevents (or, at least, makes more difficult) repudiation.
How should we store passwords? IAM is clearly a big industry for big companies, but lots of people have the problem. There has to be a simple solution?
