Network Monitoring with Dsniff By Duane Dunston,GSEC 5/29/2001 9:54 This is a practical step by step guide showing how to use Dsniff, MRTG, IP Flow Meter, Tcpdump, NTOP, and Ngrep, and others. It also provides a discussion of how and why we should monitor network traffic. In order to properly understand how your network operates and to debug any problems with network congestion, and other network issues, network monitoring is essential. It helps to quickly find out if your local network is having a problem, a particular host, or if some hosts are using up an excessive amount of bandwidth. It can also be used to just provide a historical analysis of how the network is being used. Below is a description of tools used to monitor "Our Network," hereby called ON-2. Also, the monitoring server will be called "Eve" and the gateway, "Trent". The Heart of the monitoring A freely available suite of utilities named dsniff ( http://www.monkey.org/~dugsong/dsniff/)is being used as the backbone to allow the monitoring of ON-2. Dsniff comes with tools to assist with network monitoring and auditing. It has received a lot of negative press (see dsniff's web page on "Recent press:" articles. because it is freely-available and can be run from anyone's computer on a network to sniff traffic, intercept SSL connections, hijack a network, etc.. In addition, it is available on two popular platforms, Windows and Unix. Despite the potential malicious use, this program has more positive (as seen below) than negative use. The programs that will be used from dsniff are arpspoof, tcpkill, and tcpnice. Why? This began as personal proof that Dsniff has good uses just as Dug Song mentions on his website "I wrote these tools with honest intentions - to audit my own network, and to demonstrate the insecurity of most network application protocols. Please do not abuse this software." (Song dsniff) Professionally, this was done to better see what is going on with our netowrk. We have a restrictive commercial firewall and we aren't allowed access to the console and it doesn't have the utilities needed to accurately monitor the network. After some experimenting under the cover of pretending to do real work, this paper is the result. Just kidding...do what is outlined below. Essential preparation 1. GET PERMISSION FROM YOUR IMMEDIATE SUPERVISOR AND ALERT OTHER SYSTEM ADMINISTRATORS IN YOUR DEPARTMENT SO THEY CAN MONITOR FOR ANY ABNORMALITIES THAT MAY RESULT FROM THE USE OF THE PROGRAMS MENTIONED IN THIS PAPER!!!!! THIS PAPER EXPLAINS HOW SENSITIVE INFORMATION CAN BE OBTAINED ON A NETWORK SO BE SURE YOU GET PERMISSION AND SECURE THE MONITORING SERVER !!! 2. ENABLE IP FORWARDING!!! 3. GET PERMISSION!!! 4. ENABLE IP FORWARDING!!! 5. GET PERMISSION!!! 6. ENABLE IP FORWARDING!!! 7. GET PERMISSION!!! 8. ENABLE IP FORWARDING!!! 9. GET PERMISSION!!! 10. GET PERMISSION!!!!!!!!!!! 11. ONLY RUN SSH ON THE MONITORING SERVER, WITH ALL THE LATEST PATCHES, AND READ MY OTHER PAPER ON CREATING SECURE TUNNELS USING MINDTERM AND SSH. 12. A webserver will be used on Eve for the utilities used in this tutorial. If you are using Apache as the webserver edit your httpd.conf file and change the "BindAddress *" option to "BindAddress 127.0.0.1" and the "Listen" directive to "Listen 127.0.0.1:80". Therefore, you can only be on the local machine itself or you can create secure tunnels to the machine and view it remotely over a secure connection. Restart the webserver and use SSH to create a tunnel to Eve so that nothing is transmitted in clear text across the network and only users with legitimate accounts can access Eve. Typing: /bin/netstat -anp --inet | grep httpd should show the httpd daemon listening on the local interface: tcp 0 0 127.0.0.1:80 0.0.0.0:* LISTEN 2676/httpd To access Eve over ssh with the above settings, the following command using ssh will create a secure tunnel: /usr/bin/ssh -C -l -L 2080:127.0.0.1:80 Eve's ip address or hostname The "-C" switch enables compression and "-L" means listen on the localhost port 2080 (2080:) and forward to Eve's local interface (127.0.0.1) on port 80 (:80). Then in a browser type: http://localhost:2080 You should then be able to access Eve over a secure connection. If you are running other webservers do the same with the respective settings. Also, use an initial login page to access the webserver. REMEMBER THAT THE INFORMATION THAT CAN BE OBTAINED FROM THE MONITORING SERVER CAN BE DISASTROUS IN THE WRONG HANDS. 13. Finally, GET PERMISSION!!!!!! Software Used * RedHat (dsniff works on other platforms. See the dsniff website for further information) * Dsniff * Tcpdump * Multi Router Traffic Grapher (MRTG) * IPTRAF * IP Flow Meter (IPFM) * NTOP * Ngrep 'To spoof or not to spoof, that is the packet' First, let's look at how normal traffic works. Here is an illustration: 1. Node A transmits a frame to Node C. 2. The switch will examine this frame and determine what the intended host is. It will then set up a connection between Node A and Node C so that they have a 'private' connection. 3. Node C will receive the frame and will examine the address. After determining that it is the intended host, it will process the frame further. Please note that the hosts still perform the examination of the destination address even though the switch 'guarantees' that they are the intended host...In general, when Node A wants to communicate with Node C on the network, it sends an ARP request. Node C will send an ARP reply which will include the MAC address. Even in a switched environment, this initial ARP request is sent in a broadcast manner. It is possible for Node B to craft and send an unsolicited, fake ARP reply to Node A. This fake ARP reply will specify that Node B has the MAC address of Node C. Node A will unwittingly send the traffic to Node B since it professes to have the intended MAC address. (Sipes Why your switched network isn't secure). This technique of Node B intercepting the frames destined for Node C is called, "arp spoofing"; hence, the name of the utility that is being used from the dsniff package, "arpspoof". For more detailed information on arp spoofing read Stepen Whalen's paper Intro to Arp Spoofing. What's being done, using Sipes's example above and in this paper, is that the monitoring server is intercepting all traffic on ON-2 and then sending it to Trent. Therefore, we are able to get an accurate analysis of what is going on with our network. You should know how to defend against possible malicious use of dsniff and related programs than have to read this paper and wait to the end. The above example, only works for networks that are sharing a given gateway. If multiple networks are sharing the same gateway and proper filtering rules were in place then this would only work on the network where Eve is located. Also, hard-coding the mac address of the gateway on the switch would help prevent arp spoofing. That is a temporary fix because a "mac flood" attack can be performed on the switch. A "mac flood" is when a bunch of bogus mac addresses fills up the memory of the switch and could possibly cause it to "fail-open" (yes dsniff has that utility as well called, "macof"). This is essentially the state of a non-switched netowrk where packets are broadcasted to every machine on the network until it finds the intended host. In this state, and on a non-switched network, users only need to put their interface in promiscuous node to sniff traffic. Also, the tedious task of hard-coding the mac addresses of the network card on each machine can be done. For example, on linux machines, adding the mac address for each machine in the "/etc/ethers" file will prevent arp request and replies from being sent and received. A utility named "arpwatch" can be used to email the administrator if mac address mismatches are detected on the network. Read Carefully! The Dsniff utilities below can affect the entire network. Please be careful when using them!!!!! The most important thing to be done is enable IP Forwarding else the entire network, will be incapacitated. Again, ENABLE IP FORWARDING BEFORE RUNNING ARPSPOOF. The utilities used with dsniff can be used to intercept passwords, email, instant message conversations, and other potentially critical information. These tools should be used with the utmost care and only authorized users should have access to the server that is doing the monitoring. This is the time to pull out that dusty security book and implement strict access controls and read up more on security. And away we spoof!!! First, enable IP Forwarding!! By default IP Forwarding is disabled on Linux. Before this can start forwarding traffic to Trent, IP Forwarding must be enabled on the computer running arpspoof, hereby called Eve. This is done with the command: echo 1 > /proc/sys/net/ipv4/ip_forward However, this is enabled in a file named /etc/sysctl.conf: net.ipv4.ip_forward = 1 so IP Forwarding will always be enabled if the network interfaces are reset on Eve or if the machine is restarted. (NOTE: If IP Forwarding is not enabled and arpspoof is running the network will come to a stand still. Be sure you have IP Forwarding enabled!!!) "Arpspoof" at the core of the monitoring can be run by using the following command: /usr/sbin/arspoof 192.168.0.1 2>/dev/null 1>/dev/null & "2>/dev/null 1>/dev/null" is used to keep the output from being redirected to the console (It is sent into nothingness, a little Eastern Philosophy humor). "&" is used to put the process in the background. That is really all there is to it to take over a network, see why it has the negative publicity. All traffic intended to go to Trent will first be redirected to Eve and then to Trent. Two other utilities that come with dsniff of use can be effective for bandwidth control, "tcpkill" and "tcpnice". There are other bandwidth control programs out there. Tomasz Chmielewski Bandwidth-Limiting-HOWTO would probably work just as well. There are other utilities with dsniff but they don't have any immediate use for this paper. They can either be deleted or made non-executable: /bin/chmod 000 /usr/sbin/urlsnarf (Just repeat for each unwanted binary) Tcpkill can be used to kill connections to or from a particular host, network, port, or combination of all. These programs take standard bpf (Berkeley Packet Filters) filters so read the tcpdump man pages for examples on how to create those filters. Also, there are other sites on the Internet that have example filters. For example, to prevent any connections to the host www.playboy.com use this command command: /usr/sbin/tcpkill -9 host www.playboy.com The computer that is attempting to go to that site will be blocked from that site only, but can surf any other site. It is a good idea to either redirect the output into nothingness ( > 2>/dev/null 1>/dev/null) or into a file for later analysis (> file.tcpkill ). By default, it will redirect output to the console. More hosts can be specified with the command: /usr/sbin/tcpkill -9 host www.playboy.com and host www.real.com or /usr/sbin/tcpkill -9 host www.playboy.com or \( www.real.com or www.penthouse.com \) to block well-known ports eg., napster (port 8888 and port 6699) or gnutella (port 6346), the command: /usr/sbin/tcpkill -9 port 8888 and port 6699 or /usr/sbin/tcpkill -9 port 6346 will do the trick. "Tcpnice" is similar except that it only slows down the connection. The range is 1 through 20. 20 being the slowest. For example, to slow down napster and gnutella traffic to a crawl this command with do that: /usr/sbin/tcpnice -n 20 port 8888 and port 6699 or usr/sbin/tcpnice -n 20 port 6346 If a particular subnet (192.168.1.0/24) was using up a lot of bandwidth this command will slow down that subnet: /usr/sbin/tcpnice -n 20 net 192.168.1 This command is a bit drastic because there connection will be crawling. The default is "-n 10". Substituting the tcpkill command can be used to block that entire subnet for a while: /usr/sbin/tcpkill -9 net 192.168.1 These examples should get you started. Bandwidth usage The programs being used to monitor the network are MRTG, IPFM, IPTraf, and NTOP. MRTG MRTG is used to monitor bandwidth throughout the day and it graphically archives the data over a period of a day, week, month, and year. This data can be backed up each week and stored to either removable media or on a remote server. The main config file is /etc/mrtg.cfg. The auto-configuration command (/usr/local/mrtg-2/bin/cfgmaker) used to generate the config file is: /usr/local/mrtg-2/bin/cfgmaker --global "WorkDir: \ /usr/local/apache/htdocs/stats/mrtg" --global "options[_]: bits" \ --noreversedns --output /etc/mrtg.cfg public@Eve This command puts the graphs and logs used to create the graphs in "/usr/local/apache/htdocs/stats/mrtg". The root The root directory of the web server is /usr/local/apache/htdocs/ so the directories /stats/mrtg will have to be created: /bin/mkdir -p /usr/local/apache/htdocs/stats/mrtg The -p switch creates subdirectories within a new directory without having to create each directory separately like this: /bin/mkdir /usr/local/apache/htdocs/stats/ ; /bin/mkdir /usr/local/apache/htdocs/stats/mrtg The date is displayed in bits(--global "options[_]: bits. No name lookups are performed (--noreversedns), the autoconfigured file is saved to /etc/mrtg.cfg (--output)and the snmp community name (public) and server name to monitor is specified "public@Eve". This command only had to be run once. If the server is destroyed or the hard drive dies then this is the command to run on a fresh install of MRTG. A cron entry runs the statistics gathering every 5 (*/5) minutes; thus, the graph is updated every 5 minutes. Here is the cron entry: */5 * * * * * /usr/local/mrtg-2/bin/mrtg /etc/mrtg.cfg Graphs can then be viewed over the web by browsing to the location specified in the MRTG command above ( /usr/local/apache/htdocs/stats/mrtg). Since the root directory of the web server is /usr/local/apache/htdocs/, the last two directories "/stats/mrtg/" is what is specified in the url location bar. Open a browser and type the location to the graphs: http://localhost/stats/mrtg/ Click on the "Eve.html" link of the name of the server and the graphs should then be displayed. At this point that page can be bookmarked or an initial Monitoring webpage can be created with the absolute path to the "eve.html" file. Interpreting MRTG MRTG graphs show the max output for the time periods displayed. Time is in military time. As a quick reminder, if the time is 13 to 23 subtract 12 from that number to get the standard time. So 16:00 would be 4:00 because 16-12 is 4. Each line on the y-axis represents one hour. High spikes indicate a large amount of traffic being sent or received. The example above shows a high traffic spike. Notice how it distorts the normal looking traffic just above the x-axis. Though accurate it is difficult to see those results. That is why we are using multiple monitoring utilities. The daily chart on MRTG shows a two-day period so the graph will return back to normal after that time. IP Flow Meter (ipfm) Ipfm collects the amount of data used for each host on the network. It is run in the background with the following command: /usr/sbin/ipfm -c /etc/ipfm.conf (-c) specifies the config file to use. Here's the config file /etc/ipfm.conf: # Global variables DEVICE eth0 # local variables ##### FIRST LOGGING CONFIGURATION ##### #log our subnets 192.168. LOG 192.168.0.0/255.255.0.0 #Path and name of log file. FILENAME /var/log/ipfm/ipfm-%A-%d-%B-%Y-%H.%M.%S #Time to output collected statistics. TIME 30 minute #Data is sorted by the host that has sent the most traffic SORT OUT #Whether or not to resolve hostnames RESOLVE ##### SECOND LOGGING CONFIGURATION ##### #Create a new log file. NEWLOG #Log our subnets LOG 192.168.0.0/255.255.0.0 #Path and name of log file. FILENAME /var/log/ipfm/ipfm-week-%A-%d-%B-%Y-%H.%M.%S # Log on a period of one week TIME 7 day SORT OUT RESOLVE The congfiguration "192.168.0.0/255.255.0.0" is used because 192.168.0.1 is our gateway and it is assumed all of our networks, 192.168.0.0/16, are using it as the gateway as well. The output of each log file includes the IP (or Hostname), In, Out, and Total traffic for each host. Traffic is in bytes. The example below, in the section Interpreting ipfm output is the format of the ipfm-* log file, without the color and bold. Data is saved in the /var/log/ipfm directory so a symbolic link was created to the web directory /usr/local/apache/htdocs/ipfm so data is readily available. The command to do that was: /bin/ln -s /var/log/ipfm /usr/local/apache/htdocs/ipfm (Important Security Note: Be very careful when you create symbolic links to your web directory. It is possible that improper server configurations could lead to a malicious person reading very sensitive files by exploiting this link. Read your webserver documentation carefully about creating symbolic links. All this data can be backed up to the "/var/log/ipfm/backup" directory and viewed from the webserver. First create a new directory under the "/var/log/ipfm" directory named "backup". /bin/mkdir /var/log/ipfm/backup Here is a shell script to perform the backup: #!/bin/sh #Create a new directory under /var/log/ipfm/backup with "ipfm-" and the current date as the name. /bin/mkdir /var/log/ipfm/backup/ipfm-`date '+%A-%d-%B-%Y'` #mv the ipfm daily stats into /var/log/ipfm/backup/ipfm- /bin/mv /var/log/ipfm/ipfm-* /var/log/ipfm/backup/ipfm-`date '+%A-%d-%B-%Y'`; Add the contents to a file under the /etc/ directory named "ipfm-backup.sh". Make it executable only by root: /bin/chmod 700 /etc/ipfm-backup.sh. Backups are kept in the /var/log/ipfm/backup/ipfm- directory, where is the day of the ipfm log statistics, respectively. The cronjob is run at 11:56 each night. Here is the cron entry: 56 11 * * * /bin/sh /etc/ipfm-backup.sh The second part of the /etc/ipfm.conf file specifies that a log is created once a week with the overall stats for each host. Interpreting ipfm output Data output is self-explanatory, except to know that the data is sorted by the host that has sent the most data. That is specified in the /etc/ipfm.conf file with the option: SORT OUT Looking at the graph above we can see that the time period the spike occurred was between 9:00 and 10:00am in the morning. All that needs to be done now is to browse to the /var/log/ipfm/ directory and look for the output files with the appropriate time stamp. In this case the files ipfm-Thursday-10-May-09.01.53 and ipfm-Thursday-10-May-09.31.53 will show what host(s) caused the spike. (Note the file name in the previous sentence. The filename comes from the /etc/ipfm.conf config file from above. The FILENMAME option is "ipfm-%A-%d-%B-%Y-%H.%M.%S". In this example, the file name is ipfm-Thursday-10-May-09.01.53. All files will be created in that format. Snipped for brevity. Host IN OUT TOTAL Eve 5006296 306296058 311302354 192.168.0.5 305264283 5483840 310748123 Trent 1538863 11686 1550549 Host 192.168.0.5 is responsible for the spike in traffic, from the "MRTG Example" illustration above. It received over 300 megs of data (IN). Which is accurate considering I transferred over 300 megs of data from Eve (OUT) to test it out. IPTraf Iptraf is a superb utility for network monitoring. It allows you to view and log all incoming and outgoing requests to the Internet and on the local LAN. It is a very flexible and lightweight program. It has a very easy to understand interface. IPTraf's website has superb documentation, and screenshots on the website so details will be spared and only mention what we are currently using. Also, it comes with a precompiled ready to run binary. If you want to compile it yourself then you can do that as well. It is run in interactive mode from the console by issuing the command: /usr/local/bin/iptraf IPTraf gives detailed information and statistical data on each host, port, protocol, etc., on ON-2. The option to log data will always be prompted before the data analysis is begun while in interactive mode. All data is stored in the directory /var/log/iptraf/ and has a corresponding file for what is being monitored. IPTraf runs in the background (-B) with the following command: /usr/local/bin/iptraf -d eth0 -B This gives detailed (-d) information on the protocols used and an overall rate of how much bandwidth is used since the first instance of IPTraf was run. The data is updated every 5 minutes and redirected to a file in the web directory on Eve. Currently it runs continuously, however that can be changed by adding a cron entry to stop and start it at a certain time of day. The data is continuously updated and averages are kept since iptraf first ran so stopping it each day and starting it again gives a better view of what's going on during that day. The documentation located on the website has more extensive information on what IPTraf can do. However, experimenting with it without the documentation has no affect at all on the network. Only the utilities used with dsniff can alter the network. A cron entry redirects the output of the 5 minute statistics logging to a file in the web directory: /usr/bin/tail -30 /var/log/iptraf/iface_stats_detailed-eth0.log \ > /usr/local/apache/htdocs/iptraf/fiveminutes.txt tail -30 is used because the last 30 lines of the output gives all the information for the last five minutes. And the entire iface_stats_detailed-eth0.log file is redirected to the web directory each hour for analysis over a period of time: /bin/cat /var/log/iptraf/iface_stats_detailed-eth0.log > /usr/local/apache/htdocs/iptraf/all.txt Tcpdump Tcpdump (WinDump is available for Windows platforms) is used to capture traffic on an interface with the specific purpose of helping to debug network problems. It is also great for capturing data to send to an ISP if problems appear to be occuring on their network, eg. their servers killing connections. We have used tcpdump to debug a lot of problems like that. The tcpdump output can be stored in any directory, provided the filesystem where the directory is located has enough space. Space is crucial as tcpdump files can grow very large. For this example, we'll use the /log directory. Here is the command used to capture data: /usr/sbin/tcpdump -w /log/dump_`date '+%A-%d-%B-%Y'` \ ! broadcast and ! multicast The data collected from the tcpdump -w command is stored as a binary file. This format type is good because other programs such as traffic-vis, tcptrace, tcprelay, and other programs use the binary format of tcpdump to analyze data. Normal viewing yields a bunch of garbage. The `date '+%A-%d-%B-%Y'` command grabs the current Month, Day, and numerical day and Year, respectively, and uses that as the file name. ! broadcast and ! multicast is used to not log broadcast and multicast as that could cause the file to grow insanely large. Tcpdump can be started from a cron job. Here is a shell script to do that, named tcpdump.sh #!/bin/sh /usr/sbin/tcpdump -w /log/dump_`date '+%A-%d-%B-%Y'` ! broadcast and ! multicast Add that file to your /etc directory and make it executable: /bin/mv tcpdump.sh /etc ; /bin/chmod 700 /etc/tcpdump.sh. Here is the cron entry to start it at 24:00: 0 0 * * * * /bin/sh /etc/tcpdump.sh Tcpdump is stopped at 23:59, moved to a backup directory, and tarred and gzipped with the following shell script named "tcpdump-backup.sh": #!/bin/sh #Grab the pid for the current running tcpdump process tcpdump_pid=`/sbin/pidof tcpdump` #kill current tcpdump process /bin/kill -9 $tcpdump_pid ; #Move to the backup directory /bin/mv /log/dump_`date '+%A-%d-%B-%Y'` /log/backup/ ; #Tar and gzip the days dump file. /bin/tar -czf /log/backup/dump_`date '+%A-%d-%B-%Y'` Make the file "tcpdump-backup.sh executable and place it in the /etc directory. /bin/mv tcpdump-backup.sh /etc ; /bin/chmod 700 /etc/tcpdump-backup.sh. Here is the cron entry: 59 11 * * * /bin/sh /etc/tcpdump-backup.sh Interpreting tcpdump traffic For a more thorough explanation of how to interpret tcpdump output read Karen Frederick's excellent three-part articles. She explains the output using WinDump but it applies to Unix's tcpdump, as well. Richard Bejtlich has an excellent article on interpreting tcpdump output. Also, Stephen Nortcutt's books "Network Intrustion Detection: An Analysts Handbook" Edition 1 and Edition 2 are highly recommended. NTOP NTOP is no longer being used because it was very CPU intensive, especially the way we were using it, but it gives superb statistics in a web-based format. It is an excellent program without a doubt that is why it is included here. If there is a way to use ntop only to collect data from a pcap_dump file without collecting ongoing statistics please let me know. Ntop has feautures similar to MRTG and IPTraf combined. Ntop gives graphical statistics in a web-based display. It gives the overall network usage as well as per host usage, the number of packets sent and received by each host, and also statistics from each remote hosts to name just a few features. It also keeps a history of what websites each host has visited. Ntop was stopped every hour for 15 seconds (it had a habit of crashing sometimes) and started up again at which point it reads the tcpdump output file to continue gathering statistics. Ntop has its own database files that gathers its own statistics. Ntop was running with the command: /usr/sbin/ntop -i eth0 -f /log/dump_`date '+%A-%d-%B-%Y'` \ -F "OurNetwork='net 192'" -r 400 -l 300 2>/dev/null 1>/dev/null Ntop reads the tcpdump output file (-f /log/dump_`date '+%A-%d-%B-%Y'`) to gather statistics and other info. This is a great feature because it gathers everything that a each hosts has done, how much bandwidth was used, the number of packets each host sent and received, each hosts history, pretty graphs, protocols used, it can create matrixes between networks using the filters commands, etc., but it used too much CPU because of the large log file. The filter (-F) OurNetwork is created to provide statistics about communication on ON-2 (net 192). Other filters can be created as needed (see the ntop man pages). The information collected by ntop is updated every 5 minutes (-l 300) and the web based screen is updated every 6 minutes (-r 400) All the information gathered is self-explanatory as the links on the left side of the screen explain what data is to be shown. Ripped from the Headlines There are a number of utilities that can be used to extract information from what is collected over the network while using the dsniff utilities, like tcpdump (or WinDump for windows), and also to capture information on the fly. Also, problems can be debugged by using these same utilities. Below you can see how data can be extracted from was explained above using tcpdump. Normal viewing of tcpdump with the -w switch yields a bunch of garbage. However the command: /usr/sbin/tcpdump -n -r /log/dump_ will output the contents of the file in human-readable format, well so to speak. The "-n" switch will output the data without doing a dns lookup so it is extracted much, much, much, much faster. Or a particular host can be grepped from the file like this: tcpdump -n -r /log/dump_ | grep '' If the file is large then a fast system should be used else system utilization will degrade until the process finishes. Luckily we have two hard core machines to do just that. To view the ascii output of a particular host then tcpdump, tcpdump2ascii (http://www.bogus.net/~codex/), and sed will help. This command, though wicked, will do the job nicely: /usr/sbin/tcpdump -n -x -r /log/dump_ | tcpdump2ascii \ | sed -e '/./{H;$!d;}' -e 'x;/AAA/!d;' (where "AAA" is the name of the host you want to lookup) It is a good idea to redirect the output into a file for analysis add "> filename.tcpdump" to the end of the command above to do that. /usr/sbin/tcpdump -n -x -r /log/dump_ | tcpdump2ascii \ | sed -e '/./{H;$!d;}' -e 'x;/AAA/!d;' > .tcpdump First the data is outputted into hexadecimal (-x)format from the tcpdump output file (/usr/sbin/tcpdump -n -x -r /log/dump_) then piped into tcpdump2ascii ( | tcpdump2ascii). "tcpdump2ascii" converts the hexadecimal to ascii and each packet and its contents are separated by a paragraph, so sed is used to print each paragraph matching the keyword (sed -e '/./{H;$!d;}' -e 'x;/AAA/!d;') (in this sed command substitute AAA for the host name or anything for that matter). NOTE: The data is still kind of difficult to read because of the html format so another sed command will strip out most of the html code: sed -e :a -e 's/<[^>]*>//g;/ The sed commands are taken from the "Sed One-liners" documentation found on various websites. Programs like tcpslice or pcapmerge can be used to extract particular data from tcpdump output to help narrow down the file size. Here is a sample tcpdump file to test out the commands above. (Note: The first few packets is not a SYN scan. The frames and ads on securityfocus.com probably caused those multiple SYN connections to be created.) Ngrep Another excellent utility to use is ngrep (http://www.packetfactory.net/Projects/ngrep/). With ngrep certain texts can be sniffed from the network and displayed when a match is found. For example, to dump packets that contain the words, "playboy, bunnies, bunny, Hefner", this filter would output all packets that match including the local host host name and remote server name: ngrep -d eth0 'playboy|bunnies|bunny|Hefner' or ngrep -d eth0 'playboy|bunn*|Hefner' The pipes (|) delimit each key word. -d specifies the interface to sniff. In the second command the words"bunnies" and "bunny" was substituted with the wildcard match (*). Filters can be added to the end of the text expressions like the above examples with the dsniff utilities. To sniff a particular host on the network this command will work: ngrep -d eth0 '*' host Again redirecting the output to a file will make the output more readable and using the sed command above to strip out the html code will help as well. The documentation contains more examples of how to specify filters using ngrep. Instead of using the tcpdump extraction commands above, ngrep will read tcpdump output as well. For example, this command will do the same as the tcpdump output above but quite a bit prettier: /usr/bin/ngrep -I /log/dump_ The "-I" switch implies data will be read from a file instead of from the network. This command will only print packets from the file that contain data. For more verbose output, incluing empty packets, and flags add a few extra switches: /usr/bin/ngrep -qte -I /log/dump_ This command means be quiet "-q", print the timestamps (t), and show empty packets (e). If there were multiple interfaces in the file the "-d" switch means listen on that interface. /usr/bin/ngrep -qted eth0 -I /log/dump_ Using the bpf filters, specific data can be extracted from a tcpdump file as well. Test out ngrep using the sample tcpdump file above. Again, these tool should be used with care as very sensitive information can be obtained from it. Conclusion Hopefully, this paper gives some general ideas on what can be done using dsniff as an aid in network monitoring. There are many tools out there that can be used to assist with other aspects of auditing a network too. A search on Google, Securityfocus, Freshmeat (Linux newbies, it is not what the name implies), LinuxApps, and various other websites will yield many other programs that will help you with monitoring. There are other ways to monitor a network but because of our situation this way was inexpensive. The only financial cost was an available Dell Pentium III 600, 6 gig, 256mb RAM server. All the utilities in this paper are free. Also, the floppy distribution, Trinux, contains all the programs mentioned in this paper, except MRTG. Trinux is the floppy Linux distribution for Network Administrators and Security Administrators. It has many programs that comes with it and the most basic distribution requires only 3 floppies. Everything mentioned in this paper was also done on a 486sx, 16mb, computer running only in RAM with no performance degradation. Trinux is designed to be a portable distro to help Administators audit and monitor their network. It can be installed on a hard drive as well. We ran Trinux in RAM and mounted the hard drive and redirected all output on the hard drive, then used the SSH utility scp to send it to a more powerful machine for analysis. Dsniff is here and will be around for a while. It has some great uses and that is the sole intent of this paper. It is hoped you find this article useful and I would like to hear how you are using it. TODO Try out Tomasz Chmielewski Bandwidth-Limiting-HOWTO on Eve as a more effective and free bandwidth-limiting usage alternative. Special thanks to my boss for giving me the opportunity to implement this in our organization. Thanks to my partner in crime Tony Inskeep for his editing and Patrick Girling. Also, thanks to Chris, Chrissy, Bob, and mutsman for their continued support for all that I do. Duane Dunston is a Unix Technical Specialist at Cape Fear Community College. He received his B.A. and M.S. degrees from Pfeiffer University and he has his GSEC certification from SANS. He writes poetry, hangs out at The General Assembly, and River Dogs in beautiful Wilmington, NC and wakes up every morning ready to go to work.