The Snort intrusion detection system can identify suspicious and malicious activity by inspecting network traffic. Snort makes a judgment based on its analytical capabilities and notifies the operator of its decision by generating an alert. I call the output of this collect-inspect-report process "alert data."
While this is a good and necessary methodology, it has one important flaw. In most configurations, Snort is not told to report on what it sees if the traffic in question is deemed to be "normal." One might consider this aspect of Snort to be a benefit. Why generate an alert if the traffic is "normal" and not suspicious or malicious?
No alerting system can perfectly identify all suspicious or malicious activity. In many cases it's simply not possible -- especially on a packet-by-packet basis -- to identify a packet or stream as being worthy of an operator's attention. In those cases it makes sense to keep a log of the traffic. Recording traffic or characteristics of traffic for later analysis has recently been labeled retrospective network analysis (RNA), not to be confused with Sourcefire's Real-time Network Awareness. Others call recording traffic in this manner "network forensics," but that implies a degree of care and evidence handling that exceeds the methodology I present here.
When you collect data about traffic that Snort didn't consider to be suspicious or malicious, you have the opportunity to look back (hence the term "retrospective") to see what happened during an incident. How do you know to look back? Perhaps you receive a tip from law enforcement. Maybe a client reports odd activity. Or you perform a manual investigation and realize you'd like to know as much as possible about the network traffic of a certain host. In all of these situations, Snort might not have provided any clue that something was amiss.
Despite my attention to Snort in this series, I never deploy Snort as a stand-alone tool. I always supplement Snort with additional data sources. One of the most important supplementary data sources I collect is session data.
What is session data?
As I wrote in my first book, The Tao of Network Security Monitoring (Addison-Wesley, 2005): "Session data represents a summary of a conversation between two parties. [...] A session, also known as a flow, a stream or a conversation, is a summary of a packet exchange between two systems." I also defined the seven minimum elements of a session, namely:
- Timestamp (generally when the session began, preferably including the end of the session, too).
- Source IP address.
- Source port (for protocols that offer ports, like TCP, SCTP and UDP).
- Destination IP address.
- Destination port.
Measure of the amount of information exchanged during the session (to include bytes sent by the source, bytes sent by the destination, packets sent by the source and packets sent by the destination).
I don't consider records that lack these seven basic elements to be session records. If you have a record showing only a source IP talking to a destination IP, it isn't completely worthless. However, a real session record includes the other five elements listed in addition to source IP and destination IP.
What does a session look like?
The following is a session record. I'll explain how I generated this record later.
StartTime Proto SrcAddr Sport DstAddr Dport SrcPkts SAppBytes DstPkts DAppBytes 1204514703.744796 tcp 220.127.116.11.32903 18.104.22.168.443 10 1953 9 4049
The columns contain the following:
1204514703.744796: date in Unix time format. Using the date command we can convert to something more obvious:
p># date -r 1204514703
Sun Mar 2 22:25:03 EST 2008
tcp: the protocol of the session
22.214.171.124.32903: the source IP and source port
126.96.36.199.443: the destination IP and destination port
10: number of packets sent by the source
1953: number of bytes of application (above layer 4) data sent by the source
9: number of packets sent by the destination
4049: number of bytes of application (above layer 4) data sent by the destination
This is a bidirectional session record. In other words, it displays data sent by the source and destination in a single record. Other types of session records, such as NetFlow (not shown here) are unidirectional. That means two records are needed to describe a conversation. The first lists data from source to destination, while the second lists data from destination to source.
Getting started with Argus
Let's look at how you can collect session data using Argus 3.0, developed by Carter Bullard. Argus dates back to 1993, the days before Snort and Wireshark/Ethereal, when Tcpdump was one of the few tools a network security analyst had. For the last two years, Carter has been coding Argus 3.0, and this article uses code posted to ftp.qosient.com/dev for introductory purposes.
Argus 3.0 is a complicated and powerful tool. Those who frequent the Argus Development Mailing List are not going to learn anything new here! This article is for those who are unfamiliar with Argus but would like to try it for network session data analysis.
The following examples are run on FreeBSD 6.2:
Argus is currently packaged as a server component (argus) and client programs (argus-clients). I retrieve both using "fetch" and then extract them.
hacom:/usr/local/src# fetch ftp://ftp.qosient.com/dev/argus-3.0/argus-3.0.0.tar.gz
argus-3.0.0.tar.gz 100% of 391 kB 833 kBps
hacom:/usr/local/src# fetch ftp://ftp.qosient.com/dev/argus-3.0/argus-clients-3.0.0.rc.69.tar.gz
argus-clients-3.0.0.rc.69.tar.gz 100% of 1422 kB 1058 kBps
hacom:/usr/local/src# tar -xzf argus-3.0.0.tar.gz
hacom:/usr/local/src# tar -xzf argus-clients-3.0.0.rc.69.tar.gz
Because I like to experiment and keep multiple versions of programs like Argus, I install each program into its own directory in /usr/local. (Yes, I could use Stow.)
hacom:/usr/local/src# mkdir /usr/local/argus-3.0.0
hacom:/usr/local/src# mkdir /usr/local/argus-clients-3.0.0.rc.69
Next, I compile the Argus server package. Only an "argus" binary is created in the "bin" directory:
hacom:/usr/local/src/argus-3.0.0# ./configure --prefix=/usr/local/argus-3.0.0
hacom:/usr/local/src/argus-3.0.0# make install
hacom:/usr/local/src/argus-3.0.0# ls /usr/local/argus-3.0.0/sbin/
Finally, I compile the Argus clients package. A variety of tools are created in the "bin" directory:
hacom:/usr/local/src/argus-3.0.0# cd ../argus-clients-3.0.0.rc.69
hacom:/usr/local/src/argus-clients-3.0.0.rc.69# ./configure --prefix=/usr/local/argus-clients-3.0.0.rc.69
hacom:/usr/local/src/argus-clients-3.0.0.rc.69# make install
hacom:/usr/local/src/argus-clients-3.0.0.rc.69# ls /usr/local/argus-clients-3.0.0.rc.69/bin/
argusbug radark rahisto rapolicy rastrip
ra radump rahosts raports ratemplate
rabins rafilteraddr ralabel rasort ratimerange
racluster ragraph ranonymize rasplit ratop
racount ragrep rapath rastream ratree
Now I'm ready to collect session data with Argus.
Starting the Argus server
Let's look at Argus 3.0's more easily demonstrated features. The help screen summarizes command-line options; I don't create a config file here:
hacom:/root# /usr/local/argus-3.0.0/sbin/argus -h
Argus Version 3.0.0
usage: argus [options] [-i interface] [filter-expression]
usage: argus [options] -r packetfile [filter-expression]
options: -A Generate application byte metrics.
-b dump filter compiler output.
-B specify bind interface address.
-c <dir> daemon chroot directory.
-d run Argus in daemon mode.
-e specify Argus Identifier .
-h print help.
-F read configuration from .
-J generate packet performance data.
-M set MAR Status Report Time Interval (300s).
-m turn on MAC Layer Reporting.
-O turn off filter optimizer.
-p don't go into promiscuous mode.
-P enable remote access on (561).
Note that the help page is missing the -i option to specify a sniffing interface.
Now I start the Argus server.
hacom:/root# /usr/local/argus-3.0.0/sbin/argus -A -B 10.1.13.4 -P 651 -R -u sguil -g sguil -U 256 -w /nsm/argus/argus3.test.arg -Z -i bridge0 - ip
ArgusWarning: argus: 02 Mar 08 22:10:18.647475 started
ArgusWarning: argus: 02 Mar 08 22:10:18.648282 ArgusGetInterfaceStatus: interface bridge0 is up
I tell Argus to generate application byte metrics, bind its server to IP 10.1.13.4 and port 651 TCP, to generate response time data, to run as user sguil group sguil, to capture 256 bytes of user data, where to write its output, to generate packet size data, to sniff on interface bridge0 and to watch only IP traffic.
The sockstat command shows Argus has bound a server to port 651 TCP. We'll see what that offers soon.
hacom:/root# sockstat -4 | grep argus
sguil argus 41996 4 tcp4 10.1.13.4:651 *:*
sguil argus 41996 6 udp4 *:* *:*
As Argus collects data it appears in the specified directory:
hacom:/root# ls -lh /nsm/argus/
drwxr-xr-x 2 root sguil 512B Mar 2 22:04 .old
-rw-r--r-- 1 sguil sguil 3.5M Mar 2 22:04 argus3.test.arg
With Argus saving data, I need to use one of its client programs to read it.
Using the Argus ra client
One of the fundamental client programs packaged with Argus is ra. The help options in the 3.0 ra client are extensive:
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./ra -h
Ra Version 3.0.0.rc.69
usage: ra [options] -S remoteServer [- filter-expression]
usage: ra [options] -r argusDataFile [- filter-expression]
options: -A print record summaries on termination.
-b dump packet-matching code.
-c specify a delimiter for output columns.
-C <[host]:port> specify Cisco Netflow source.
-e convert user data using method.
Supported types are and .
-E write records that are rejected by the filter
-F read configuration from .
-h print help.
-M rmon convert bi-directional flow data to RMON in/out stats
poll attach to remote server to get MAR and then disconnect
-n don't convert numbers to names.
-p print fractional time with precision.
-q quiet mode. don't print record outputs.
-r read argus data . '-' denotes stdin.
-R <dir> recursively process files in directory
-s [-][+[#]]field[:w] specify fields to print.
fields: srcid, stime, ltime, sstime, dstime, sltime, dltime,
trans, seq, flgs, dur, avgdur, stddev, mindur, maxdur,
saddr, daddr, proto, sport, dport, stos, dtos, sdsb, ddsb
sco, dco, sttl, dttl, sipid, dipid, smpls, dmpls, svlan, dvlan
svid, dvid, svpri, dvpri, [s|d]pkts, [s|d]bytes,
[s||d]appbytes, [s|d]load, [s|d]loss, [s|d]ploss, [s|d]rate,
smac, dmac, dir, [s|d]intpkt, [s|d]jit, state, suser, duser,
swin, dwin, trans, srng, erng, stcpb, dtcpb, tcprtt, inode,
offset, smaxsz, dmaxsz, sminsz, dminsz
The following is one way to invoke ra. I tell ra to show time in Unix format, with TCP flags, column headers, no resolution of IP addresses, while reading the specified file. The format of the remainder of the output is the same as demonstrated earlier. I conclude with a filter "- tcp" to only display TCP records, and I use the head command to limit the output to 10 lines.
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./ra -u -Z b -L0 -n -r /nsm/argus/argus3.test.arg -s stime proto saddr sport daddr dport spkts sappbytes dpkts dappbytes state - tcp | head
StartTime Proto SrcAddr Sport DstAddr Dport SrcPkts SAppBytes DstPkts DAppBytes State
1204513818.770060 tcp 188.8.131.52.2377 184.108.40.206.5223 1 0 1 0 A_A
1204513828.429709 tcp 220.127.116.11.443 18.104.22.168.40237 1 0 1 0 FA_A
1204513828.858757 tcp 22.214.171.124.2377 126.96.36.199.5223 1 0 1 0 A_A
1204513830.041338 tcp 188.8.131.52.3192 184.108.40.206.995 15 421 17 3974 FSPA_
This sort of raw output is helpful if you want to inspect specific records. For a more top-down approach I turn to racluster. Note the first and third records describe the same session. This must be a persistent connection. We can use the racluster client to aggregate these, assuming Argus has seen most or all of the session from the beginning.
Using the Argus racluster client
The Argus client racluster is useful for getting a higher-level view of session data. The client merges records from the same flows. First I show the help options for racluster:
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./racluster -h
Racluster Version 3.0.0.rc.69
usage: racluster [-f racluster.conf]
usage: racluster [-f racluster.conf] [ra-options] [- filter-expression]
options: -f read aggregation rules from .
-m flow key fields specify fields to be used as flow keys.
-M modes modify mode of operation.
ind aggregate multiple files independently
norep do not report aggregation statistics
rmon convert bi-directional data into rmon in/out data
replace replace input files with aggregation output
-V verbose mode.
In the following example I tell racluster to collect all records involving host 220.127.116.11 and present the first four results:
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./racluster -L0 -M norep -u -n -r /nsm/argus/argus3.test.arg - host 18.104.22.168 and tcp | head -n 5
StartTime Flgs Proto SrcAddr Sport Dir DstAddr Dport TotPkts TotBytes State
1204513828.429709 e tcp 22.214.171.124.443 126.96.36.199.40237 6 421 RST
1204513862.958283 e tcp 188.8.131.52.40238 -> 184.108.40.206.443 25 7741 RST
1204513878.329589 e tcp 220.127.116.11.40239 -> 18.104.22.168.443 23 5684 RST
1204514058.412265 e tcp 22.214.171.124.57330 -> 126.96.36.199.443 23 5684 RST
The following shows aggregation based on unique source and destination IP address pairs:
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./racluster -L0 -M norep -m saddr daddr -u -n -r /nsm/argus/argus3.test.arg - tcp | head -n 5
StartTime Flgs Proto SrcAddr Sport Dir DstAddr Dport TotPkts TotBytes State
1204516393.988996 e ip 188.8.131.52 <-> 184.108.40.206 4 276 CON
1204513828.429709 e ip 220.127.116.11 <-> 18.104.22.168 1513 377809 CON
1204513831.418396 e ip 22.214.171.124 <-> 126.96.36.199 1679 446483 CON
1204516648.005118 e ip 188.8.131.52 <-> 184.108.40.206 8 504 CON
To change the order of the output, call rasort. Although I won't show the output here, you could use the following command to sort the racluster output by bytes.
hacom:/usr/local/argus-clients-3.0.0.rc.69/bin# ./racluster -M norep -m saddr daddr –u
-n -r /nsm/argus/argus3.test.arg -w - - tcp | ./rasort -m bytes -L0 | head -5
This article has only scratched the surface of Argus 3.0. If you're interested in using Argus for the collection and analysis of network session data, I recommend starting to record some data using the Argus server. Once you have data to inspect you can begin trying the client programs.
About the author
Richard Bejtlich is the founder of TaoSecurity, author of several books on network security monitoring, including Extrusion Detection: Security Monitoring for Internal Intrusions, and operator of the TaoSecurity blog.