MitM-ing Postgres | Hacker News Reader

MitM-ing Postgres

thusoy.com

119 pointsthusoy14 comments

Excellent post. The post doesn't detail a problem with Postgres per se, rather the author uses plain language to explain Postgres' TLS connection mechanisms and how they are applied in AWS and Heroku RDBMS services.

When quickly iterating on software, I must admit I don't always consider configuring a root cert for the db. This nuanced look at Postgres' security options gives me much to consider for real production work.

tl;dr: configure a root cert for your Postgres db and use verify-full in your connection string.

Thanks, glad you liked it!

> configure a root cert and use verify-full

Edit: The interpretation below is incorrect, see the answers to understand why.

I don't understand why the post insists on using self signed certificates and ?sslmode=verify-*. It's MitM-prone by design.

Just use ?sslmode=require and a CA-issued certificate. It's even easier than with a webserver and clients can identify that you are the domain you pretend to be. Obviously Postgres did it wrong by being to lazy to deploy normal certificates for all dbs.

"Verify" is only with client certs, which are difficult to issue and install, and allows the server to identify the client.

Am I correct?

Yes, self-signed certificates without a CA is MitM-prone, but that's the only thing Heroku enables. You're only given a username and password and have to do with whatever configuration they've done, thus the problems.

> Just use ?sslmode=require and a CA-issued certificate.

This is not secure. See Table 31-1 here:

https://www.postgresql.org/docs/9.5/static/libpq-ssl.html#LI...

As you can see, "require" provides no MitM protection. The only option among the six options that provides both eavesdropping and MitM protection under all scenarios is "verify-full".

Your misunderstanding is reasonable, and shows what can happen when software is insecure by default and has too many different security-sensitive options. I'll bet that very few Postgres users fully understand the nuances between the various sslmode options.

he is not incorrect. if you are on a secure network you can stick with the default i.e. a aws network could be secure.

tajen advocated a CA-issued certificate along with the "require" option, which makes sense under no common scenario. If your network really is secure, you don't need TLS at all, so you can use "disable" and not bother with a cert. If your network isn't secure, then you need "verify-full" to actually have security.

actually i just wanted to point that out. actually on aws you actually connect with ssl when you have an rds instance and using psql on a ec2 instance. however it will run with "prefer". which isn't really needed there.

If you consider the Internet a secure network you can. Note that unless you connect to IP addresses you still need to do host name resolution which typically involves the Internet and if the Internet is considered insecure, may return answers controlled by an attacker..

I'd suggest avoiding shortcuts and using certificates.

Correct me if I am wrong, but this doesn't sound like a problem with Postgres. It seems like a problem with the author's specific implementation (self-signed certificates) specifically on Heroku.

Anyone else take something else away from this piece?

OP here. That's entirely correct, Postgres deployed correctly is secure against a MitM, but since Heroku has not issued a trust root, they are not.

Edited to add: As a user of Heroku Postgres you can't configure Postgres correctly, Heroku is supposed to do that for you.

The blog post raises several distinct issues. Heroku's poor configuration is the most significant issue, but it's also concerning that Postgres is insecure by default, and supports several different options to sslmode with confusingly-similar names, only one of which is secure under normal circumstances. That's a recipe for developers and database administrators to shoot themselves in the foot. If Postgres had been secure by default, it might have saved Heroku from making this mistake.

See OPs response.

I went down a similar road in 2013 and contacted heroku via a support ticket. They said they "do not offer up [their] certs for verification" and wouldn't answer any other questions. Doesn't seem like much has changed :/

Reply on news.ycombinator.com