The idea of a SQL Firewall came to me a number of years ago, probably sometime around 2008, however I have not given it much thought since. It has come up a number of times in discussions with colleagues, managers, and clients however the discussions did not move further.
The topic has recently come up again, and I was presented with an actual use-case, so I decided to delve into this further and gauge the feasibility.
The solution would have to be designed and developed, then tested to ensure that it did not have any impact on performance, and of course that it does “what it says on the tin”.
The scenario was suggested by someone I know who works as an IT Auditor. They wanted to know whether, when accessing a SQL Server instance, Logins (SQL or Windows) can be restricted to a specific list of Hosts (or IP addresses). Those working with Firewalls and Access Control Lists (ACLs) are familiar with this concept.
There are a number of ways that a technical implementation can be done. All of them revolve around a Logon Trigger working around a Table, created in the MASTER database, and which would store the mappings between Logins and the Hosts, or rather the IP addresses since the Host Name can be defined as part of the connection string - see Workstation ID (or WSID) in the SqlConnection.ConnectionString Property reference document or the documentation for the HOST_NAME function.
An Access Control List (ACL) would contain a list of Logins and the IP addresses from where they are allowed to or blocked from accessing the SQL Server instance. The ACL would provide flexibility in that it would allow an Administrator to control who can or cannot connect. This also leads to increased overheads since the Logins would have to be maintained (i.e. every time a Login is created/deleted) and Login-Host mappings would have to exist for all possible combinations (e.g. an Organisation where Users do not have a fixed desk/location).
This approach would also allow Admins to create restrictions, that is, from where a Login cannot connect. Of course, a larger number of entries could lead to increased performance overheads.
On the other hand, an Allow List (AL) would be a simpler approach, requiring the least number of records since we are only storing who can connect and from where. This was the selected approach.
As mentioned above, the AL Table contains the Long-Host mapping, hence when the Login Trigger fires the code would:
Check the source IP Address against the one stored
Originally this was being extracted from the
client_net_addresscolumn of the sys.dm_exec_connections Dynamic Management View. This however required the
VIEW SERVER STATEpermission, which might not be granted to all Logins due to security restrictions. An alternative, and quite possibly a better one because it does not require accessing a DMV, is using the EVENTDATA() function. In a nutshell, this returns an XML structure which would have to be “shred” in order to extract the “ClientHost” property value. The intermediate results of the
EVENTDATA()function are being stored in a XML variable, then querying that variable for the “ClientHost” property.
That’s it. So if the first condition is not met this means that the Login does not have any restrictions and the connection is allowed.
If on the other hand the first condition is met and the second is not, then the connection is dropped.
If any of the rules need to be “disabled”, we can simply delete the record from the Table, thus ensuring that the data remains valid and trim.
The solution works best with SQL Logins. When someone connects using a Windows Logins it is relatively easy to capture who initiated the connection. In the case os a SQL Login, which I personally advocate to avoid using unless there is no other technical solution, the credentials can be shared between multiple individuals and used from unauthorised machines. Of course if the User has been granted permissions to log on to an “allowed” Host, and uses the SQL Login to initiate the connection, there isn’t much a DBA can do. If this scenario exists in your organisation, I would suggest reviewing your security policies.
When a Login attempt fails, due to the above conditions being met, the End User will get the following error message:
Logon failed for login '<LOGIN_NAME>' due to trigger execution.
This is a standard message and cannot be replaced by a friendlier one, such as to inform the End User that they are not allowed to connect from that specific Workstation/Host for example.
The Logon Trigger will however write the
RAISERROR message to the SQL Server ERRORLOG, which can then be used for analysis and troubleshooting. A sample of the message is shown below:
Failed connection attempt for <Login_Name> from <Host_Name>
Error: 3609, Severity: 16, State: 2.
The transaction ended in the trigger. The batch has been aborted.
Error: 17892, Severity: 20, State: 1.
Logon failed for login '<Login_Name>' due to trigger execution. [CLIENT: <IP_Address>]
The solution will not work with Active Directory Groups since the Group Name cannot be retrieved when a connection is attempted. Implementing such a feature would necessitate granting additional permissions, and would also incur increased performance overheads.
Finally, it is highly recommended that the Login-Host mappings for every SQL Server instance are are stored in an alternate location - a source-control system might just do the trick, with the audit trail providing evidence should this be required.
The code to implement this SQL Firewall can be downloaded from my SQL Firewall GitHub repo. This has been tested with SQL Server 2012, 2014, 2016, 2017, and 2019, however you are urged to carry out your own testing (never trust code off the internet!).