I'm gonna catch hell for this, but this article is for the masses - newbies and elite both. And if you're elite, just remember you were a newbie once, so lighten up.
Nmap is a network scanner that allows you to specify various kinds of scans, like SYN, FIN, etc. written by Fyodor. At this point I only know of its existence on Unix, so don't go trolling through Infoseek for a Wintel version. And if there is a Wintel port, someone clue me in.
Newbie Exercise #1: What are SYN and FIN, and how do they relate to scanning? Check out the TCP/IP protocol (specifically, the structure of a TCP/IP packet). Also, hunt down Fyodor's webpage and read through the nmap docs to get more info.
Elite Exercise #1: Sit back and relax. The good info is coming. And in case you're rusty, brush up on your nmap switches.
Now, I revere nmap as a great piece of work, but I do have a few points I'd like to mention about it, and I think everyone can get something out of this. You see, sometimes a stealth scan isn't always a stealth scan.
First, I shouldn't even have to mention that a connect() scan is certainly loggable. This is the -t option, and is also the default. Now, if you're running on a network you have permission to, you're OK. But if your goal is to maintain some semblance of stealth, then make sure you specify -s, -u, or -U so you don't use normal connections.
Newbie Clue: A connect() scan uses normal connections to other systems, using no kinds of stealth and are most times logged (which is bad). It's called "connect()" because that's the name of the programming function that does it.
(Side note: on an NT 4.0 sp3 system, I found no logged referrals to anything after a connect() scan. But a firewall or router before the NT box could still grab anything off a connect() scan.)
Also, you should use the -F (fast scan) option in most cases. This will only check for services found in /etc/services (basically like "strobe"). This will minimize the actual packets sent, and really only check ports that count.
Newbie Exercise #2: On a Unix system, take a peek in /etc/services. You should learn the concept of a port, and common port assignments (ftp, telnet, smtp, dns, pop, imap). Also, what is "strobe"? Look it up - it's another scanning tool (a bit older). What does it do? Is it stealthy?
Elite Exercise #2: Show off your suave knowledge of port numbers by constructing a custom port list via the -p option. For instance, on most systems SSHd (test: which port is that?) isn't in /etc/services, so if you want to detect it, you'll need to 1) add it to etc/services, or 2) specify it with -p. By the way, typically installations of SSHd give off the version in the banner. And there's a problem with pre 1.2.26 versions... (as well as recent problems with the Kerberos code in 1.2.26).
So, what about that detectable part? I ran some tests against a few of my home systems, just to see what the systems detected. I ran NetXray to sniff off the network to watch what's going down the wire also.
Newbie Exercise #3: Do some research on network sniffers. What are some common ones out there? How does a switched network environment affect sniffing?
Elite Exercise #3: Tackle tcpdump. Read the raw output code, and be able to follow complete exchanges. If you can, you da man! (or woman)
Well, here's some simple results I've gotten on two systems:
SYN scan against RedHat Linux 5.0 box
Scan is accurate in determining open ports, but also leaves traces in the logs:
Jul 7 05:16:12 empri ftpd: getpeername (in.ftpd): Transport endpoint is n$
Jul 7 05:16:13 empri named: accept: Connection reset by peer
Jul 7 05:16:36 empri lpd: accept: Connection reset by peer
Jul 7 05:16:36 empri rlogind: Can't get peer name of remote host: Transpo$
Jul 7 05:16:36 empri rshd: get[peername: Transport endpoint is not connect$
Jul 7 05:16:12 empri in.telnetd: connect from unknown
Jul 7 05:16:36 empri in.rexecd: warning: can't get client address: Connec$
Jul 7 05:16:36 empri in.rexecd: connect from unkown
Jul 7 05:16:36 empri in.rlogind: warning: can't get client address: Conne$
Jul 7 05:16:36 empri in.rlogind: connect from unkown
Jul 7 05:16:36 empri in.rshd: warning: can't get client address: Connecti$
Jul 7 05:16:36 empri in.rshd: connect from unkown
FIN scan against RedHat Linux 5.0 box
No detectable signs in logs, and accurately returns port listing.
SYN scan against Win NT 4.0 sp3 box
Returns accurate port listing; however, MS DNS spits two events into the App event log, source: dns, event: 1 & 2. Both have "no description", and bogus insert strings. Unless you specifically knew that could be caused by port scanning, it's completely cryptic.
"The description of Event ID (1) in Source (DNS) could not be found. It contains the following insertion string(s):."
FIN scan against Win NT 4.0 sp3 box
Leaves nothing detectable in the event log, but also fails to detect any open ports.
Newbie Exercise #4: If you can, try to set up a Unix (Linux) box, and familiarize yourself with the logs (in /var/log/) and services (like ftpd, lpd, etc.). Or set up an NT 4.0 server. By the way, sp3 means service pack 3 was applied.
Elite Exercise #4: OK, time to show off. My list of sample scans is far from comprehensive. See what you can find out against Solaris, HPUX, AIX, etc. Bonus if you e-mail me the results.
My experience with the UDP scan seems to suck, majorly. It failed to report any accurate port listings vs. NT and Linux. However, a packet capture of nmap vs. NT shows that an ICMP "port unreachable" message is sent in response to a UDP sent to a non-open port, but no return message is sent in response to an open port. It's possible that this scan could work vs. NT, but the software isn't working right, or not expecting it.
Elite Exercise #5: Figure out how to fix it. It may be as simple as increasing the default time-out.
Note that NT seems to "take in" UDP packets to ports with TCP services; i.e., a UDP to port 80 won't get an ICMP "port unreachable" message, but on Linux it will (both running web servers). I think this is published already, so I'll move on.
An interesting point is that every packet sent out contains the data "blah"... this could be filtered at the firewall (any UDP packs containing only "blah" alert sysadmin to port scan).
Newbie Exercise #5: The line responsible for the "blah" is 920 in nmap.c. Modify the source to have NULLs (0x00) instead of "blah". If need be, get a little intro to C.
Elite Exercise #6: Be more creative. Shove random() junk in for "blah". Again, line 920 in nmap.c.
On the same token, the SYN & FIN scan is detectable too. First, every packet comes from the same port (49724).
Newbie Exercise #6: Both nmap.h and tcpip.h have a #define for MAGIC_PORT as 49724. Change it to another port. Careful! Make sure you know what port numbers are valid (What port ranges are reserved? What's the highest port possible?).
Elite Exercise #7: Obviously, add extra functions to change the MAGIC_PORT for every packet sent. And a hint: sequential increases are detectable. Be creative... randomly increase between 1-5 ports, etc.
Also, every packet there's typically some bytes of frame padding, being "\nhelp\nquit\n".
Newbie Exercise #7: Again, change the "\nhelp\nquit\n" to some other random data. This time, I'm not going to tell you where to look for it. I recommend you use the Unix command "grep" to find it. If you need more info on this command, use the command "man grep".
Elite Exercise #8: Find and change that to something unknown, preferably random() data.
Remember that it is very feasible to set up filters to detect a vanilla nmap scan (vanilla being unmodified source). As simple as: from port 49724 and contains "QUIT"... (pseudo filter language).
From the sample scans above, you can see there's a dilemma. If you don't know what OS a system is running and you did a FIN scan, you'd get accurate results against a Linux box but not against an NT box. And if you did a SYN scan, the Linux box would log it, but you'd get accurate results against the NT box. What's this mean?
It's very important to know what OS you're scanning against! OS's respond differently to stealth scans, so you have to be creative and figure it out beforehand. This is the concept behind a newer program called "queso".
Newbie Exercise #8: Locate queso and try to get it up and running. Again, it's for Unix platform.
Elite Exercise #9: Is queso itself stealthy, or loggable? Are there any telltale signs of a queso scan (other than raw packet dumps)? I havn't played with this much, so bonus points if you e-mail me your findings.
Also, not too long ago (as of me writing this), there was a public post by Shadow concerning certain findings in regards to scanning.
Newbie Exercise #9: Who's Shadow? Give you a hint: they're government. Do a look for them.
One very interesting point I would like to highlight from that document is that it is possible to detect scans as small as two packets a day! Granted, this isn't a hard feat, and detecting one packet a day scans would lead to tons of false alarms. I'll give you a hint... the Shadow system involves a few systems running tcpdump with massive hard drive space, and they just log every packet and then analyze the data for the past few days to put scans together. No amount of stealth will avoid this. You need to waste another brain cell and figure out how to still stay low under radar.
And, at this point, I want to make a public gripe:
Shadow reports that "hackers are cooperating in scanning efforts." I'm sorry, but I saw no evidence supporting this claim within that document.
Point 1: If two hackers truly were coordinating scans from different locations against a common target, there shouldn't be any overlap in IP and port assignments (i.e., the same port should not be scanned twice). Either these hackers are severely sloppy (which I find hard to believe if they're doing coordinated stealth scans against.gov installations), or they weren't working together. They just happened to be scanning the same .gov at the same time.
Point 2: Just because there are two separate geographic sources for a scan doesn't mean there are two people cooperating on the effort. Nothing stops me from firing up two telnet sessions to two different (geographically separated) boxes, and launching scans back to the same target from those points. It could be one person splitting his scanning across two sources.
OK, so what did we learn here? Hopefully something of use. And I hope some newbies now have an inkling on what to do next. Let me finish this scanning article with a few tips:
1. Scanning any system, any port twice is sloppy. Be organized, and minimize the packets you send out.
2. Patterns can be mined and deduced. Sending packets at a fixed interval is stupid. Make large amounts of possible randomness between packets (and make sure that randomness doesn't result in two packets being sent close together).
3. Patience is a virtue. One packet a day total is good.
4. Dispersed sources (geographically or not, but not same organization) is practically a must. And tip #3 doesn't apply per source; it applies to sources as a whole (meaning if you have five source systems, you should coordinate so one packet per system every five days, leading to one packet per day, with no overlap).
5. We are simple creatures, and usually order things in a linear fashion, but there's no reason you should scan ports in order (or reverse order.) Kinda goes with tip #2.
Remember: in this day and age, network efficiency and reliability is increasing. It's hard to even say that one packet could be misrouted, let alone several. The concept of a "completely random packet" is becoming rare - and paranoia can easily deduce that the packet was actually planned.
I don't want to quote Mentor's Manifesto, but remember, it's all about seeking info, and learning. Use this info wisely. No, it won't help you change your grade in your school's computer. No, it won't help you crash your buddies' Win9x box. If you're a "newbie," and you're truly in it for non-destructive purposes, good for you. If you want to e-mail me a question (notice the singular), if I can help, I'll try. But don't expect detailed instructions on how to do anything. If you want to learn, I'll try to point you in the right direction.
The lego program by Miff of 9mm.com (issue 15:3) can be adapted to spew packets as described above. Plus, it's in perl... which is my interpreter of choice. Kudos to Miff.
Armageddon wrote an article in the same issue about probing remote networks. A good read for newbies. He mentions use of WS Ping, which has a great UI, but remember, WS Ping does connect()-type scans (and if you analyze the packets' output, it actually does more than just connect... But I'll leave this as another exercise). Kudos to Armageddon.
Let me digress about 10 years and do my greets to JM working the Doc in Rogers Park. Take care kiddies. (I use that term literally, since I'm probably pushing the "old" brink of the average reader.)
2600 Magazine, Volume 16, Number 1, Spring 1999