Timothy Smith

Securing access for SQL Server auditing

May 21, 2018 by

Our organization must restrict permissions and prove to an independent party that we investigate access and restrict permissions. We restrict permissions to objects using least permissions and give full access to an exceptional few based on an organizational design that follows best practices. What can we do additionally that will help us prevent unauthorized access or catch when someone who shouldn’t have access is able to infiltrate our systems, especially in the context of showing a third party that we track this.

Overview

In this post we’ll look at two methods we can use for providing information to a third party auditor. The first we’ll be using some techniques by monitoring access, which can be on the level of procedure or T-SQL execution to tracking logins running queries. In addition, we can separate our server by data access – high priority or PII data being placed in a node where few have access to, while lower priority data exist in nodes where more have access to.

Monitoring access and execution

We can monitor access outside of running a consistent trace – though the trace is more comprehensive if we must provide a third party with evidence that we’re tracking user access. We’ll look at three methods of doing this and we have more than these three options, they provide us with some best practices to consider. Using a derivative of our restricting object access, we’ll limit our environment to using procedures or queries that always log who accesses data. In an application layer, this may be a query that returns a result and also inserts data, or we could use the procedure method, which we see below this.


In addition to encrypting our procedure, we also log the object called along with the user and date.

The downside is that all our procedures must log the user, objects and time, the upside is that even if a user has access, we can spot inconsistencies with access, such as an employee looking at a report 12 times in 1 day for a monthly report, which they generally only look at once or twice during that month. This helps us possibly catch a scenario where UserA comprises UserB’s access and looks at data without approval using UserB’s credentials.

We’ll also observe that in the above procedure example, we’re using the option WITH ENCRYPTION, which prevents users from scripting a stored procedure (with restricted access they shouldn’t have this anyway) to discover what the definition of the procedure is. This adds another layer of security to our objects, though I will caution developers who user procedures like this that they must know the definitions in order to alter them correctly – the cost of inconvenience adds a layer of security.

What about invalid logins, where a user attempts to log into a server when they shouldn’t, or when a valid user tries to log in multiple times in a row as if they lost their password? One way we can track this on the server level is to retain all invalid login information from the error log.


We save the failed the login information so that we can later run aggregates against it.

The downside to saving this information is that it can become noisy in some situations, so I would approach alerting with this information carefully, so that people don’t learn to ignore it. This can be a very accurate way of catching a hack attempt, but too much alerting with it can cause developers to ignore the attempts.

The upside is that we can spot irregularities by using aggregate queries and track how this changes over time. In an example scenario, we may get 120 bad logins a month. Suppose that we migrate some priority data to our server without adding new user access and where the priority data only have access a few times a month, but we suddenly see a spike of 7000 bad logins per month. This is an example scenario where we have a possible compromise attempt and we can architect a new design for our priority data, or track who leaked information.

In some situations, we may want to take a snapshot of the current logins running queries in addition to the other ideas. In the below query, we get the users running transactions (the filter on status for running) along with the date and time information of the login. Depending on our requirements, we may want to run this on a schedule or we may want to run this randomly if we want to prevent an infiltrator from identifying a consistent query. In some situations, an infiltrator doesn’t want to be on a server consistently logged in, otherwise, they’re easier to identify. An occasional login may get caught in these types of audits.


Saved information from our random or scheduled audits of currently logged in users running transactions.

Since we need to submit evidence that we’re tracking information to a third party, these types of audits also make this process easier, since we’re logging this information with a timestamp for the random audit or scheduled audit, depending on the requirements.

Delineating data storage by access

For a database with priority and (or) PII data, we may consider placing the database on a separate server with a few restricted accounts, firewall restrictions, and other security enhancements like two-factor authentication. Even if an infiltrator managed to compromise an account and had access to a username and password, without the firewall being open, or without a two-factor code, this would mean nothing. Even though this doesn’t guarantee security, we can provide this as a security design we’re using if we’re answering to an audit from a third party involving our security practices.

Part of the reason why demarcating data can be so effective, even if costly, is that it requires companies consider who should have what level of access. Unfortunately, some security breaches occur because these discussions never happened, so everyone has full access to all data. In addition, because our database servers are separate, hacks trying to increase permissions become more difficult, as it may not be a service account that an infiltrator must get access to, but also passing through a firewall and other obstacles. Finally, because fewer people will have access to this server, this makes tracking the behavior of the users even easier – how do two people tend to query versus hundreds?

The price tag for this option does come with higher costs, as demarcating data on separate servers for enhanced security means that we have multiple servers and these servers do not directly communicate with each other. This means our methods for transferring data will leave us with less automation and convenience, even if the security is more robust.

Conclusion

In addition to retaining this information, we want to make sure that where we store the data is restricted. If we keep a copy of the audit data local to the database where all users have read access, this defeats the purpose. We can retain copies in other restricted databases and migrate the data, if needed, since the audit data itself will become priority data.

In addition, all security requires a combination of techniques and no technique alone is sufficient to protect against compromise, especially in a complex technical environment. While these will add some costs, for some environments, the cost for any compromise is much greater than the cost for monitoring and preventing possible infiltrations.

References


Timothy Smith
Auditing, Security

About Timothy Smith

Tim manages hundreds of SQL Server and MongoDB instances, and focuses primarily on designing the appropriate architecture for the business model. He has spent a decade working in FinTech, along with a few years in BioTech and Energy Tech. He hosts the West Texas SQL Server Users' Group, as well as teaches courses and writes articles on SQL Server, ETL, and PowerShell. In his free time, he is a contributor to the decentralized financial industry. View all posts by Timothy Smith

168 Views