In 1999 we wrote that forensic computing was "gathering and analyzing data in a manner as free from distortion or bias as possible to reconstruct data or what has happened in the past on a system." Trusting your tools and data once you have them is problematic enough (Chapter 5 - "Systems and subversion" - talks about this at length), but there is an even greater problem. Due to the Heisenberg principle of data gathering and system analysis (see section 1.4) even with the right and trusted tools you cannot get all the data on a computer, so where should you start? This appendix gives an overview of how to gather data on a computer as well as the problems caused - mainly by the Order of Volatility, or OOV.
As we have demonstrated throughout the book, computers store a great amount of information in a significant number of locations and layers. Disk storage and RAM are the two most commonly thought data repositories, but there are a great number of places - even outside the system if it is connected to a network - that useful data can hide.
All data is volatile, however. As time passes the veracity of the information goes down, and the ability to recall or validate the data also decreases. When looking at stored information it is extremely difficult to verify that it has not been subverted or changed.
That said, certain types of data are generally more persistent, or long-lasting, than others. Backup tapes, for instance, can typically be counted upon to remain unchanged longer than things in RAM - it is less volatile. We call this hierarchy the order of volatility. At the top you have some pieces of information becoming virtually impossible to recover within a very short time - sometimes nanoseconds (or less) of their inception date - such as data in CPU registers, frame buffers, etc. At the bottom of the order you have things that are very persistent and hard to change, such as stone tablets, printouts, and other ways of imprinting data into a semi-permanent medium.
So in most cases you simply try to capture data while keeping this order in mind - the more rapidly changing information should almost always be preserved first. Table B.1, also seen in the first chapter, gives a rough guide to the life expectancy of data.
Table B.1: The expected lifespan of data.
Registers, peripheral memory, caches, etc. nanoseconds Main Memory nanoseconds Network state milliseconds Running processes seconds Disk minutes Floppies, backup media, etc. years CD-ROMs, printouts, etc. tens of years
Information lost from one layer may still be found in a lower layer (see sections 1.5, "Layers and illusions" and 1.7, "Fossilization of deleted information" for more about this.) But the point of OOV is the opposite: doing something in one layer destroys information in all layers above it. Simply executing a command to retrieve information destroys the contents of registers, MMUs, physical memory and time stamps in the file system.
And if the act of starting up a program to read or capture the memory of a system has the potential to cause a loss of existing data in memory as the kernel allocates memory to run the program that performs the examination in the first place. So what can you do?
With our additional work on forensics we have decided to put less emphasis on the "in a manner as free from distortion or bias as possible" from our definition in the introduction to this appendix. We believe that by what can be seen as risking digital evidence you are more likely to get additional data and a better chance of addressing and understanding the problem at hand.
This goes against traditional forensic computing wisdom, which uses very conservative methods - rarely more than turning off a computer and making a copy of a system's disk [US DOJ, 2004]. Certainly if the goal is to ensure that the data being collected is optimized for its admissibility in a court of law and you've only got one shot at capturing it, a very cautious methodology can be the best thing in some cases.
Unfortunately this misses a wealth of potentially useful information about the situation, such as running processes and kernels, network state, data in RAM, and more. Only a limited understanding can arise from looking at a dead disk. And while perhaps dynamic information is a bit more volatile and therefore suspect, any convictions based of a single set of data readings are suspect as well. Certainly, as we've seen throughout the book, no single data point should be trusted anyway. Only by correlating data from many points can you begin to get a good understanding of what happened, and there is a lot of data to look at out there. It would be a pity to throw it away.
Gathering data according to the OOV helps in general preserve rather than destroy, but unless computer architectures changes significantly there is no single best way to capture digital evidence. For instance RAM might be the first thing you'd like to save, but if you're in a remote site and you have no local disk it could take hours to transfer it to a safe disk somewhere else, and by the time you're done much of the anonymous memory (the most ephemeral part - see chapter 8) could be long gone.
Certainly the current set of software tools to capture evidence is not terribly compelling. Our own TCT, while at times useful, could be much improved upon. Other packages, most notably SleuthKit [Carrier 2004], and Encase [Encase 2004] are worthy efforts but still have far to go. It's too bad that we have not progressed far beyond the erstwhile dd copying program, but automated capture and analysis is very difficult.
The spirit of Darryl Zero (see section 1.1) infuses our mindset - if you're looking for anything in particular, you are lost. But if you keep your mind and eyes open you can go far.
The reproducibility and provability of results is difficult when dealing with the capture of very complex systems that are constantly in motion. Starting states of computers will always be different, often with significant changes in operating system, kernel, and software versions that are too complex for anyone to fully understand.
So ideally you would want both raw and cooked (processed) data. Having the process table from a FreeBSD computer is of limited worth if you don't have native programs for analysis, so the output from ps is still important. You you will also want to gather data on the system while it is still running as well as at rest, in case they return different results. Volume also becomes problematic - where do you store all this data? It's one thing storing a personal workstation, but what happens when trying to analyze a Petabyte or Exabyte class server?
A thorough discussion on how to gather and store digital evidence would perhaps warrant a book unto itself, but we shall try to give some basic guidelines.
Richard Saferstein [Saferstein, 2003] writes that when processing a crime scene a basic methodology should be followed, one that we espouse when dealing with a computer incident as well. Here are his first steps to an investigation:
It doesn't take much imagination to see how all of these apply to computers.
You should first consider how much time you would like to spend on analyzing the data, as it is a time-consuming processes to both collect and to process all the information. As an only slightly tongue-in-cheek guide we offer table B.2 as a possible guide.
ACTION EXPERTISE REQUIRED TIME CONSUMED Go back to work None Almost none Minimal effort Installing system software 1/2 - 1 day Minimum Recommended Jr. System Administrator 1-2 days+ Serious effort Sr. SA 2+ days - weeks Fanaticism Expert SA days - months+
Table B.2: Rough estimate of the cost of an investigation.
A tremendous amount of time can be consumed taking care of the problem at hand, but as a rule of thumb if you don't expend at least a day or two you're probably doing yourself and your system a disservice. One of the more difficult things to judge is how much effort to put into the analysis. Oftimes the more analytical sweat you emit the more clarity and understanding you have of the situation. That said, some situations are harder and at times an intruder is more careful and skilled than others, and unfortunately you never know prior to the break-in what to expect. The truth is you'll never absolutely, positively know that you've found all that you can. Experience will be your only guide here.
You'll next need, at least, a pad of paper, something to write with, and a secure, preferably off-line location to store information that you collect. A second computer that could talk to the network would be a welcome, but not necessary, aide. A laptop may be used to store results, notes, and data, but be cautious about who has access to the second system.
Even though installing or downloading programs will damage evidence it is sometimes a necessity in order to process a computer. The negative effects of this can be mitigated by having a personal collection of forensic tools at the ready. But automation should be used at all costs, unless it is completely unworkable.
If commercial packages are not an option such open source projects as FIRE [FIRE, 2004], PLAC [PLAC 2004], and others based on the impressive Knoppix Linux distribution [Knoppix, 2004] may be placed on CDs or other portable media and used when an emergency occurs.
We apologize for our UNIX-centric advice, but the same roughly holds for any operating system. The computer(s) in question should be taken offline. There are some potential problems with this, as the system might expect to be online, and this could destroy evidence by generating errors, repeatedly retrying connections, or in general causing the system to change. Alternately you might try cutting it off at router and keep it on a LAN, but DNS and network services as well as other systems in area can still cause problems.
As you proceed along you need to keep track of everything you type or do. In general it's a "grab first, analyze later" situation, however. Note the hardware, software, system, and network configurations that are in place.
If you're serious about collecting the data, however, we might suggest you capture data in the following order, which mirrors the OOV:
Forensic computing does not often yield black and white decisions or clear and unambiguous understandings of a situation in the past. Like its counterpart in the physical realm of criminal investigations the quality and integrity of the investigators and laboratories involved are the most important thing. And the tools and methodologies used will significantly changing for the better over time.
It's a field that perhaps should be taken with more gravity than most of computer science. The FBI said that "Fifty percent of the cases the FBI now opens involve a computer" [Williams, 2002] and the number is surely rising. Programs and people involved in the gathering and analysis of evidence must take special care, for people's freedoms, lives, jobs, and more can and are seriously impacted by the results.
Nothing is certain, but while probabilistic forensics does have a negative sound to it, it's what we now have. However, much has been written and achieved in the short time this field has been around, and we fervently hope that progress continues, because it will benefit us all. With best luck to the future,
[Carrier, 2004] Brian Carrier, TASK (Sleuth Kit):
[Coffman, 2002] K. G. Coffman and A. M. Odlyzko 'Internet growth: Is there a "Moore's Law" for data traffic?' in: "Handbook of Massive Data Sets," J. Abello, P. M. Pardalos, and M. G. C. Resende, eds., Kluwer, 2002, pp. 47-93.
[Encase, 2004] The Encase forensic tool:
FIRE [FIRE, 2004] The Forensic and Incident Response Environment (or
F.I.R.E.) project is hosted at:
[Knoppix, 2004] Knoppix Linux distribution website:
PLAC [PLAC 2004] The Portable Linux Auditing CD project's web site is: http://sourceforge.net/projects/plac
[Saferstein, 2003] Richard Saferstein, "Criminalistics: An Introduction to Forensic Science", Prentice Hall, 2003.
[US DOJ, 2004] Forensic Examination of Digital Evidence: A Guide for
Law Enforcement. U.S. Department of Justice, Office of Justice Programs.
[Williams, 2002] Scott C. Williams, supervisory special agent for the FBI's computer analysis and response team in Kansas City, as quoted by David Hayes in The Kansas City Star, 4/26/2002.