It’s been a year long process now the book is finally been released. There are a few things I would have written different and a few other subjects I would have liked to cover. Perhaps that will come in my next book or future posts.
In this book we cover Secure Coding, setting up Encryption, and audit. We also dive deep into performing privilege analysis.
When evaluating the security of a database, and/or the security of an application, there are a few things that will make me start pounding my head against my desk. What prompted this? A customer asked me help evaluate a system that is about to go through an IG Audit and they want to be ahead of the curve. The good part is I was able to come up with a lockdown procedure that is specific to this customer and their needs. The bad part is, I keep seeing these same mistakes over and over again. So, lets sit down and go through five of the mistakes that I keep seeing over and over again.
A COTS application required any of the following privileges to get installed or operating. I’m sorry guys, if you have not figured out what permissions your application really needs, then it’s not ready to be put on the market. I was able to do a privilege analysis and find out what this application really needed.
DBA – I have a hard time justifying granting DBA to an application to be installed or operate.
SELECT / CREATE / UPDATE / DROP ANY <FILL IN YOUR TYPE>. Hey, this is another privilege that just makes me shake my head. Because the ANY any privilege allows the grantee to operate on ANY object in the database, this just tells me you have not thought it through. Many databases have multiple applications running in an instance. Why would your application need to access schemas that are not part of your applications?
Oh and as a side note, why did the application need both DBA and all these ANY privileges?
Okay there are a whole bunch of privileges that should never be granted to an application, these are just a few.
Not cleaning up users that have left. Please tell me why you need to keep a user in the system who left two years ago. If someone can give me a good excuse, I’m willing to listen, but I got to tell you, in thirty years in this business, I have not heard any good reason to keep a user in the system.
Go ahead, drop those users, including the test, sit, and dev instances. They don’t need to be there.
Connecting as the application to do your job. I see this so often that it really makes me wonder about the technical ability of the people doing the job. Please tell me why you need to connect to the application to do your work? If there are permissions that you need that are missing, we can resolve that.
Mixing Data and Code in the same schema. Why do you need to do this? You really should separate your data and code. All you need is one little sql injection bug for a bad guy to find, then she owns your database. Oh, and it’s highly likely you have a sql injection bug, you just don’t know it yet. We have a plan to separate the code from the data, it’ll take a while, but this should keep the auditors happy.
Writing queries in production. I get it, it happens; you are supporting your users and the user needs something from the database quick. There is not a report available that answers the users question. You connect to the production instance through sql plus, sqlcl, sql developer or some other tool. You then type your query, hit enter and production slams on the breaks. Yea’ I have lost count of the number of times I’ve seen this, you forgot to join those two really big tables and just got a cartesian product. A junior developer just walked up to me and asked the question, why is production so slow. My gut reaction was to chew him out. In the end, I did not chew him out, and I gave him an explanation of what happened (forgetting to put in a where clause) and gave him a stern warning about writing code in production. You have a test instance, please use it before running in production.
August 13, 2018: NOTE UPDATE TO POST THIS IS SPECIFIC TO Oracle 12.1 and bellow. Oracle 12.2 and above, you can change an unencrypted tablespace to an encrypted tablespace.
1) When we start talking about securing information, the first thing that always seems to come up is encryption. Everyone has heard about it, but some don’t really understand just what encryption is protecting. When we are discussing Transparent Data Encryption (TDE) we are discussing data at rest. The attack vectors we are protecting from is a bad actor gaining access to the physical hardware.
1a) Now, the easiest and fastest way to implement TDE is to encrypt tablespaces and move the sensitive data into the encrypted tablespace. You need to be careful here, just because you identified the tables that are sensitive, what about objects that are dependent on the table? (Indexes, Materialized Views, etc). Each of these sensitive objects need to be moved into encrypted tablespaces.
Find dependent objects.
set pagesize 1000 set linesize 132 col owner format a30 col name format a30 select d.owner, d.name, s.tablespace_name, t.encrypted from dba_dependencies d, dba_segments s, dba_tablespaces t where d.owner = s.owner and d.name = s.segment_name and s.tablespace_name = t.tablespace_name and referenced_name IN ( SELECT segment_name FROM dba_segments WHERE tablespace_name IN (SELECT tablespace_name FROM dba_tablespaces WHERE tablespace_name = upper('&&tbs'))) UNION SELECT i.owner, i.index_name, i.tablespace_name, dd.ENCRYPTED FROM dba_indexes i, dba_tablespaces dd WHERE i.tablespace_name = dd.tablespace_name AND table_name IN ( SELECT segment_name FROM dba_segments WHERE tablespace_name IN (SELECT tablespace_name FROM dba_tablespaces WHERE tablespace_name = upper('&&tbs')));
You can not change an unencrypted tablespace into an encrypted tablespace, so you are going to need to first create the encrypted tablespace. UPDATE, in Oracle 12.2 and above you can change an unencrypted tablespace into an encrypted tablespace.
Now that we have an encrypted tablespace, we need to start moving all the sensitive data into it. It’s important to know, that to prevent ghost data you we are going to need to move everything out of the tablespace and into a new tablespace. I normally use alter table move, but you can also use dbms_redefination and create table as select. Use the report from dependent objects to make sure you have everything out of the tablespace. Once you have everything out, drop the tablespace then use a utility like shred to over right the data file(s) with random data. Once you have done that, you can safely delete the data file(s).
Here is a link to my demo on moving data to an encrypted tablespace. This demo assumes the base table is already in the encrypted tablespace, now we need to move indexes and materialized views. https://www.youtub
TDE also offers column encryption, the analysis required to properly implement column based encrypted is time consuming. So for now we are going to pass over column encryption.
1b) SQLNet encryption. Information that moves through the network is subject to various attacks including man in the middle, replay and modification attacks. With these data can be leaked, corrupted, or even replayed. So we use sqlnet encryption and integrity to protect our data from leaking, replays and modification. You are going to user net manager to setup. Make an encryption or integrity method either Accepted, Requested, Rejected or Required. You can read more on these in the Oracle Documentation.
$ORACLE_HOME/bin/netmgr Open local –> profile then select network security and click on the encryption tab. Select the encryption algorithms you need and then enter 256 characters in the encryption seed block.
Select an integrity method. Remember MD5 has several weaknesses. SHA has become the defacto standard.
2) Audit your users and environment. I’ve have heard this one time and time again, “It’s not my job. It’s the auditors job.” The fact remains many breaches exists for weeks, months and even years before they are discovered. Than when a breach is discovered, the auditor request audit logs. We need to do better!. I get audit reports every morning and review them before I do anything else. So what do you want to audit?
2a) Audit login failures. Login failures can be a sign someone is trying to gain access to your system. If you start seeing login failures, investigate. Is the issue user training or is there something else going on.
2b) Audit logins from yesterday. Why you are going to be looking at os_username / username / userhost to see if people are logging in from multiple workstations. This could be an indicator of a username/password being shared. Another reason I do this audit is to check os_username / username. Is the user using their proper account. I have issues in the past where a user was using the application login to do their normal work. This audit showed this and allowed us to correct the situation.
2c) Audit logins for the past 31 days. This gives you a 30,000 foot picture on how often users are connecting and are then disconnecting at the end of the day.
set heading off set pagesize 1000 set linesize 132 set serveroutput on col object_name format a24 col object_type format a24 col doctype format a10 col userhost format a40 col os_username format a15 col username format a15 col terminal format a15 spool $HOME/session_audit.txt select 'login failures' from dual; select os_username, username, userhost, terminal, to_char(timestamp, 'dd-mon-rr hh24:mi') from dba_audit_session where returncode != 0 and trunc(timestamp) >= trunc(sysdate-1) and username != 'DUMMY' order by timestamp /
select 'logins yesterday' from dual; select os_username, username, userhost, count(*) from dba_audit_session where trunc(TIMESTAMP) >= trunc(sysdate-1) and username != 'DUMMY' and action_name != 'LOGOFF BY CLEANUP' group by os_username, username, userhost order by os_username, username /
select 'logins last 31 days' from dual; select os_username, username, userhost, count(*) from dba_audit_session where trunc(TIMESTAMP) >= trunc(sysdate-31) and username != 'DUMMY' and action_name != 'LOGOFF BY CLEANUP' group by os_username, username, userhost order by os_username, username /
2d) Audit changes to any database objects. This is a simple query you can start with. You are checking objects based on last_ddl_time and created.
select 'changed / created objects last 24 hours' from dual; select owner, object_type, object_name from dba_objects where (trunc(created) >= trunc(sysdate-1) or trunc(last_ddl_time) >= trunc(sysdate-1)) order by owner, object_type, object_name /
2e) Use a product like tripwire to check for any changes to ORACLE_HOME. You can also roll your own but getting a checksum of all the files in ORACLE_HOME, than doing the check everyday to see if a file has changed. (there are some files you will want to filter because they change in the normal course of operations)
3) Identify the sensitive data in your database. How can you know what to protect if you don’t know what is sensitive? When you identify what is sensitive, it’s easier to track that data through your enterprise to limit access to the data.
Now there is something most people know but don’t realize they know. You can have a simple piece of data, that by itself is not sensitive, but when combine it with other data that’s not sensitive and now you have sensitive data. IE: I’m Robert, that by itself is not vary valuable. Combine that with my zip code that is not very sensitive and you have narrowed down the universe of Roberts’. Next add that I drive a Ford F150 and a BMW R1150RS, now you have uniquely identified me.
3b) If you have not already done so, you can reverse engineer your application scheme into Oracle SQL Developer Data Modeler. SQL Developer Data Modeler has the ability to mark and report columns that are sensitive.
Heli made a blog entry on how to mark and report on sensitive data in SQL Developer Data Modeler. Sensitive Data From Heli
4) Create trusted paths to your sensitive data. Or at the very least, limit the number of paths to get to your sensitive data. Now that you have a list of sensitive data, document where that data is getting accessed by. You can use unified_audit_trail to get hosts names and users accessing the data. Once you have validated how the data is getting accessed and from where and by who you can setup Virtual Private Database and redaction to limit the paths to get to the sensitive data.
Here is a simple example, if people who are authorized to access the data are in a specific subnet, then you can check the subnet and use the VPD policy to append where 1=2 onto the where clause to anyone querying the data from outside that subnet. You can also use authentication method in this check. Say the data is so sensitive you only want people to access it if they connected using RADIUS. If someone connected using anything else, again you would append where 1=2 onto the where clause to return nothing. This important thing to remember is, Your VPD policy can be anything you can code in PL/SQL.
Lets start this with setting up some support objects in the security schema.
Setup a table of ip_addresses and if the user is granted access to credit card number, social security number and if they user has customer access.