Managing Network Security

Watching the World

by Fred Cohen

Series Introduction

Computing operates in an almost universally networked environment, but the technical aspects of information protection have not kept up. As a result, the success of information security programs has increasingly become a function of our ability to make prudent management decisions about organizational activities. Managing Network Security takes a management view of protection and seeks to reconcile the need for security with the limitations of technology.

Speed Kills

I do a fair amount of watching when I run a firewall or similar network protection device, and one of the more frequent questions I get is how I can deal with the rate of information. To give you a sense of this, on the network in my house, we have a maximum of 9Mb/s inbound traffic and about 500Kb/s outbound traffic. At work, we run at more like 48Mb/s, and on some special links, even faster. When I tell people that we watch the traffic, they find it more or less beyond belief. After all, 48Mb/s corresponds roughly to a 48 hefty books per second, and nobody can read and understand that fast.

Of course I am known as an avid reader, and I do read a lot of things and do so rather quickly, but still, they are right. I cannot read that fast and likely never will be able to. So the question is: How, how much, and to what extent, can I watch the traffic?

A Question of Meaning

The first thing to understand about watching traffic is what the traffic is made up of. For example most IP traffic is made up of packet headers and not of packet content, and most of the packet headers can be summarized pretty well by a very limited amount of information. For example, here is the summary of a relatively innocuous packet given by TCPdump:

05:43:01.870000 > . ack 13280 win 32120 (DF)

This packet is simply an acknowledgment of a packet that originally came from on port 23 and went to on port 13104. It was sent at 05:43:01.87 (5:43 and 1.87 seconds). It's not very interesting, but it occurred, so we observe. If you watch all of these packets, you will quickly come to the conclusion that there is very little information content to be gleaned from these headers on an individual basis. Rather, the headers provide information on the flow of traffic. The information is mostly useful to see how much of what is going where, and its absence often means more than its presence. While I sometimes watch this in detail, I usually watch either the flow of it, or select out a particular part of the flow for observation - for example, I might look at a particular exchange between IP addresses - or count sessions and the total amount of information communicated and number of packets involved.

Another thing I sometimes do is watch the content of the packets going by. For example, on my firewall machine, I always track the content of packets and sometimes look at it to see if I see anything strange. Of course, in order to know what's strange, you have to know what's normal, and you learn that by watching the traffic and digging into what every part of it is until you think you understand it. Here's a higher level log of session content: -> over TCP -> over TCP        +OK TTFN! (clean). -> over TCP -> over TCP -> over TCP -> over TCP -> over TCP -> over TCP -> over TCP
        +OK QPOP (version 2.2) at starting.  <>. -> over TCP        USER fred. -> over TCP -> over TCP        +OK Password required for fred.. -> over TCP        PASS mypassword. -> over TCP -> over TCP        +OK fred has 0 messages (0 octets).. -> over TCP        STAT. -> over TCP        +OK 0 0. -> over TCP        QUIT -> over TCP -> over TCP        .

Now in this case, I have removed the extraneous information (relative to the discussion) and we see a user logging into a post office protocol (pop3) server to check their email. Notice the user ID (USER fred) and password (PASS mypassword) are easily readable. This is what most 'sniffers' do to get user IDs and passwords. You can also find out a lot of user IDs and passwords in this way. As an example, I seem to remember that in the early 1990s, the CMU CERT reported one attack that grabbed about 100,000 user IDs and passwords. Cases involving 10,000 or more stolen user IDs, passwords, or credit card numbers are not that rare. At this level of detail, there is a lot to see, and a lot of it is meaningful. For example, if you watch a single IP address in this level of detail, you can tell what sites the user(s) on that system visit on the Web, what information their Web browser gives to the web server on the other side, what they are searching for with a search engine, how well they do their searches, and so forth. If there are many users on a system, you can often tell how many users are using the Internet and characteristics of their use.

Here's an example of a part of a Web session I recorded from a computer at a site. This tells me what that user was telling remote sites when visiting them: -> over TCP
        GET / HTTP/1.0.
        If-Modified-Since: Friday, 08-Jan-99 22:42:26 GMT; length=11936.
        Connection: Keep-Alive.
        User-Agent: Mozilla/3.04 [en] (X11; I; Linux 2.0.13 i486).
        Accept: image/gif, image/x-xbitmap, image/jpeg, image/pjpeg, image/png, */*.
        Accept-Language: en.
        Accept-Charset: iso-8859-1,*,utf-8.
        . -> over TCP

The web site now knows, if they care to, that I am using Netscape version 3.04 in English running under X11 on a Linux operating system version 2.0.13 on an Intel 486 computer. They could also find out my email address (if I included it), other places I have visited, what kinds of files my browser automatically processes, and so forth. One of the most interesting lines is the list of what is accepted by the browser. For attackers, this lists which attack scripts might work, and when used in combination with the other information, can form the precise profile information required to attack a site. It's good to sniff this sort of traffic from your site to determine just what your systems are telling potential attackers. This is also very handy if you suspect someone is sending something they should not be sending, or to detect the transmission of encrypted files.

Accept: image/gif, image/x-xbitmap, image/jpeg, image/pjpeg, application/msword, application/, */*.
Referer: /.
Accept-Language: en-us.
Accept-Encoding: gzip, deflate.
User-Agent: Mozilla/4.0 (compatible; MSIE 4.01; Windows 95).

This example is a bit more onerous. It indicates that version 4.0 of Netscape is running on a Windows 95 box with Microsoft Explorer present and that the system will automatically run msword and ms-excel programs if I send files with the proper extension. This could be exploited by providing a Trojan horse in a spreadsheet or word document. It also tells me that the last site this browser went to was - always a good place to visit.

We can also watch the traffic of remote terminal sessions. Here's a partial example (I have removed the traffic details to leave only the text of the sessions this time). In this case, I can tell the user is using a Unix system (it looks like Linux because of the device names returned from the df command), how much space is available, the name of the user. their directory, the name of the machine, and that they have a very big tar file.

        /dev/sda1             132185   18633   106726     15%   /.
        /dev/sda5            1119301  205134   856335     19%   /usr.
ls -l
        total 1126094.
        drwxr-xr-x   2 fc       users        1024 Jan 26 16:48 bin.
        drwx------   9 fc       users        1024 Jan 13 06:56 clisp.
        drwxr-xr-x   4 fc       users        2048 Jan 22 07:28 emacs.
        -rw-r--r--   1 fc       users         693 Jan 25 19:48 setup.
        drwx------   3 fc       users        1024 Jan 22 07:33 src.
        -rw-r--r--   1 fc       users       56496 Jan 26 22:50 tcpshow.c.
        -rw-r--r--   1 fc       users    1148549120 Jan 27 06:00 u.tar.

Here's a different kind of terminal session. It has the same content as the last session - or very nearly the same content, but it is encrypted - in this case by ssh - the secure shell. -> over TCP -> over TCP
        ..\.K.k.=.d!....R..a.r..,.... over TCP -> over TCP
        ...-5..SD.C.5..Z ..@Of.]...I.&..3.....*.g0V...B. -> over TCP -> over TCP
        .)Y....KF.l...3........?.Q.i`..n.......K -> over TCP
        .Gd..D....."Fn........:.5.E..d.T9....... -> over TCP -> over TCP
        .....>1..P.]...Xt........n.....0....r.xv -> over TCP
        ...+.\.Mp..*.y.h.....r.>.@...........-.. -> over TCP -> over TCP
        ..(.;M. \.[E&#hI...+w.U.m....Q5...wk!... -> over TCP
        ..r<..v.F.~%6bk -> over TCP -> over TCP
        ..L.    E"<.Aa.0Q.Y...bx...lOV.....D.......J............._,...3.9."
        ].W     ...|.._J.@.O24Y6.N.R.]..\.0.`.P.1.'.A
        .U....6aP.Bt..D....B.{S.;.M..&.,.u .hfJ,1.v......eZ.2.p~..~(t/.A:...z...^

There are a couple of important things to note here. First, and most obviously, the content of the session is not immediately readable. This is important to observe because not all cryptography works all of the time. In some cases, you may think information is encrypted, but until you verify it by observation, you cannot be sure. Another important thing to note is that the size of the input and output from the two sessions are actually pretty similar. While the content may be obscured by cryptography, in this case, the source, destination, and size of exchanged information were not changed. Similarly, other traffic characteristics remain largely the same. As a result, a large file transfer can clearly be differentiated from telnet sessions and Web sessions, even when encrypted.

Moving On Up

This detailed information is very helpful, but in truth, it is only good for looking at details and there are too many details to look at for this to be very useful for a high bandwidth connection. As the speed goes up, the amount of detail we can watch goes down. The issue then is to figure out what is most important to watch at any given moment and to find ways to watch it.

The collection of exchanges associate with a telnet session, or the visit to a Web site can often be rolled up as a single entry of the form:

Jan 27 05:38:22[32136]: twist to /u/fc/bin/

In this case, we see that on the specified date and time (according to the computer we are looking at), a program called with process ID 32136 ran on behalf of IP address, which appears to be a web proxy server in The Netherlands. A more detailed log of the specific transaction is kept:

1999/01/27 05:38:23 Allow GET /game/HackMove?Q0=L1%2FUnixSendmail HTTP/1.0

This log indicates that the URL they requested was:


This has particular meaning on the particular system and is not a sign of abuse in this case, however, some log files do indicate attempted entries. For example, here is a similar level log file entry produced by deception toolkit - an intrusion detection and reporting system based on deceptions:

'', '23', '23', '1999/01/25 10:56:01', '30877', '30877', '1', '', 'S0', 'R-Peace', ''
'', '23', '23', '1999/01/25 10:56:03', '30877', '30877', '1', '', 'S1', 'RPeace-Peace', 'guest'
'', '23', '23', '1999/01/25 10:56:05', '30877', '30877', '1', '', 'S2', 'RAttack-Warnings', 'guewst'
'', '23', '23', '1999/01/25 10:56:08', '30877', '30877', '1', '', 'S1', 'RWarnings-Warnings', 'guest'
'', '23', '23', '1999/01/25 10:56:10', '30877', '30877', '1', '', 'S2', 'RAttack-Warnings', 'guest'
'', '23', '23', '1999/01/25 10:56:14', '30877', '30877', '1', '', 'S3', 'RExploit-', 'ls'
'', '23', '23', '1999/01/25 10:56:19', '30877', '30877', '1', '', 'S3', 'RExploit-', 'cd etc'  

This log shows an attempt to telnet into a computer and login as the user guest. It is clear that it was a real human being doing this and not a computer because they made a typo on the password the first time and did it again. They then tried a few commands and the program eventually kicked them out.

Hopefully, intentional attacks happen far less often than access to Web services. This means, among other things, that we can watch every incident involving such an attack far more easily than we can watch normal user traffic. It also means that, while examining normal user traffic is infeasible, examining details of these attacks is quite easy even for a large network. Similarly, retention of full details of these sessions is an easy matter.

Returning briefly to the notion of time differences, the notion of relative time is very important in trying to understand sequences of events. It is very common for systems to have time differences ranging from seconds to years, and when times are important to establishing the sequence of events, tracking system time vs. actual time is important. In addition, changing the system time to correct it during an investigation is particularly problematic because it makes the historical data no longer relatable to the current time frame. For those of us who operate systems from all around the world, we always have a few systems that are running in tomorrow or yesterday and at all hours of the day. Relative time is the only way to track these systems, and the global standard against which relativity is measured is Grenich Mean Time (GMT).

Higher and Higher

At the next level of logging and analysis, we typically see roll-ups of services over time. For example, a typical corporate telephone bill or similar statement shows summaries of calls by area code, time of day, and so forth. By putting this into a database, a wide range of analyses can be done to detect patterns of usage, find anomalous periods of operation, do trends analysis, and so forth. These get rolled up into various forms depending on what is desired by the information consumer. As we move farther from the details, the ability to detect information related to specific events that are security related tends to become harder and harder.

How High Do You Go?

Some might put forth the contention that events that do not tend to have large scale effects also tend to be relatively unimportant to the organization, and that therefore, examining higher level information would be more relevant. My view is that we need to look at issues from many perspectives if we are going to detect the full range of security events that might occur.

In examining telephone billing records, we may not be able to detect every unauthorized phone call by looking at weekly billing summaries, but on a call by call basis, we probably can't tell very much either. At the low level, you can't see the forest for the trees, while at the top level, you can see that there is a forest, but may not know what trees are in it. A single call to Indonesia, likely means little to a major corporation, but a change in telephone bills of $40,000 for one weekend compared to the previous weekend would seem to be a clear indicator of a toll fraud, or at least a good basis for an investigation. A few years ago, toll frauds had this characteristic signature.

On the other hand, the value of information is often highly disproportionate to volume. A very small amount of information relative to the overall Internet traffic might contain a great deal of value. For example, a recent Trojan Horse in a Word document sends your private cryptographic keys to a site in Europe where they can be exploited to forge digital signatures, read encrypted communications, and perhaps gain unauthorized access to major financial systems. A few hundred bytes can easily be worth millions of dollars in this context, and the only way to detect this sort of activity is by looking at a very detailed log of activities. It will not show up on the monthly summary report, and by the time the financial impact appears, it may well be too late to even determine how the information was released.

It Cuts Both Ways

While you may be more aware now than you were before about the ways in which you can watch the world, you may rest assured that the world has known how to watch you for a long time. If you don't think they are watching, let me recall a few widely publicized recent incidents.


My view is that information at all levels is useful to protection management as well as to technical protection, and logs such as these must be understood and regularly viewed and reviewed in order to get those jobs done.

And please remember - as you are watching the world - the world is watching you.

About The Author:

Fred Cohen is a Principal Member of Technical Staff at Sandia National Laboratories and a Managing Director of Fred Cohen and Associates in Livermore California, an executive consulting and education group specializing information protection. He can be reached by sending email to fred at or visiting /