[josephcsible]

Wow, what a nothingburger. First of all, the passwords are hashed on disk; this is just about their transmission over the network (where they can't be hashed without the hash being password-equivalent). Anyway, the headline is only true if you don't count TLS as encryption, which is absurd. Yes, we'd probably be better off using some sort of PAKE protocol, but SQL Server handles passwords the same way basically every other server of any sort handles them. If this were actually a vulnerability in SQL Server, then you could count on one hand the number of services today that accept passwords but weren't also vulnerable.

[maxtardiveau2]

I respectfully disagree. Like I said in the article, transmitting the password, rather than some form of token, means that this approach is exactly as insecure as HTTP basic auth over TLS, and how many serious enterprise apps use that?

>> we'd probably be better off using some sort of PAKE protocol

Yes, that's my whole point. This problem has been solved, there are tons of libraries, it's not that hard. Why have a weak link like this?

If everything is secured correctly, this is not a vulnerability, but how often are things 100% secured properly? TLS is fine, but many people use a self-signed certificate, which means a MITM attack is often possible. It's bad enough to have someone snoop on your connection, but to have your password compromised... And if your client is not Windows, it often has to use database authentication.

This just stinks. It's especially surprising in an enterprise-class system like SQL Server.

[josephcsible]

> transmitting the password, rather than some form of token, means that this approach is exactly as insecure as HTTP basic auth over TLS, and how many serious enterprise apps use that?

Isn't that also exactly as insecure as submitting an HTML form with <input type="password"> is? And I can't think of any enterprise apps that don't use either HTTP basic auth or that over TLS. Which ones are you thinking of?

[maxtardiveau2]

> Isn't that also exactly as insecure as submitting an HTML form with <input type="password"> is?

Absolutely, but I am arguing that this is not worthy of an enterprise system like SQL Server. This is especially true for a back-end system, because at least in a web browser, you'd get a warning if the certificate cannot be verified (and the connection is therefore vulnerable).

> Which ones are you thinking of?

Most of the enterprise apps I have worked with use something like OAuth or SAML. For sure, many have an option to use basic auth, but that's only used for testing and development, and would be a red flag in any security audit.

I'll just quote Microsoft's documentation <https://docs.microsoft.com/en-us/sql/relational-databases/se...>:

  Disadvantages of SQL Server Authentication

  The encrypted SQL Server Authentication login password, must be passed over the network at the time of the connection.

It's good that they acknowledge it, but it would be a lot better if they did something about it.

[josephcsible]

> Most of the enterprise apps I have worked with use something like OAuth or SAML.

If an app is a SAML SP or an OAuth client, then it's not really doing authentication itself, but rather delegating it to another system. When you go to log in to the SAML IdP or the OAuth authorization server, where the authentication actually happens, don't they let you use HTTP basic authentication or <input type="password">?

[maxtardiveau2]

You're right, I should not have gone on that tangent.

To get back to the main point, though, don't you think that the fact that Microsoft includes a disclaimer (kinda) in their docs lends some credence to the idea that they're not really proud of this?

[josephcsible]

Like I said, I agree things could be better. I just don't think things are currently quite as bad as your original article makes them sound, and I also think it's unfair to single out SQL Server as if it were worse than normal, when it's exactly the same as basically everything else today.