nmap has the capability to guess the operating system with the option -O. I was highly surprised that it can find the correct kernel or service pack.

* How nmap guesses the operating system?

* What are common signatures for Windows 7,8, various Linux systems etc?


2 Answers 2


Heuristically. nmap observes the behaviour of the system during probes, develops a "fingerprint", which analagous to a real fingerprint test, looks mainly for the minutiae:

  • out of specification behaviour and extra-specification or undefined behaviour
  • retransmission times
  • response to fragmentation and various ICMP probes
  • patterns in TCP sequence numbers, IP ID numbers and TTLs
  • behaviour with various TCP options and flags

That's just a quick summary, you can read a more thorough description here http://nmap.org/book/osdetect-methods.html

The fingerprint is then compared with a database of known fingerprints. In the event that you scanned a system that has a unique fingerprint you will get a good match, otherwise you will get a guess based on the best fit (which is sometime wrong). It might be wrong anyway, some people like to play games (old but interesting). If there are no good matches, nmap will encourage you to find out what the system is and submit it.

You can find the fingerprints in /usr/share/nmap/nmap-os-db (probably), but you'll have to read the documentation to decipher the details.

In the current nmap database (6.47) there are 4485 fingerprints for 5009 CPE (platform) names, over 800 of which match various versions of MS-Windows, and over 1000 of which match various Linux systems (including appliances).

Here's an actual signature for a Linux system, an Asus Nexus 7 tablet:

# Linux 3.1.10-g52027f9 #1 SMP PREEMPT Thu Jun 28 16:19:26 PDT 2012 armv71 GNU/Linux, Asus Nexus 7 Android Tablet
Fingerprint Linux 3.1
Class Asus | Linux | 3.X | general purpose
CPE cpe:/o:asus:linux_kernel:3 auto

(Wow, that's not pretty, since you can find them in your nmap installation in nmap-os-db so I'm not going to show more.)

I'll pick a simple field IE() — ICMP echo behaviour (ping is simple, right?)


If you check you'll see that's shared with over 1250 other systems (about two-thirds of which are Linux systems).

From the documentation:

The IE test involves sending two ICMP echo request packets to the target. The first one has the IP DF bit set, a type-of-service (TOS) byte value of zero, a code of nine (even though it should be zero), the sequence number 295, a random IP ID and ICMP request identifier, and 120 bytes of 0x00 for the data payload.

The second ping query is similar, except a TOS of four (IP_TOS_RELIABILITY) is used, the code is zero, 150 bytes of data is sent, and the ICMP request ID and sequence numbers are incremented by one from the previous query values.

This decodes as:

  • DFI=N no ICMP reply will have DF (don't fragment) set
  • T=3B-45 TTL will be between 0x3B and 0x45
  • TG=40 IP initial TTL guess
  • CD=S ICMP response codes (same as ICMP request probe)

You can find all these codes and details here. The other fields are contain similarly detailed data for ECN, TCP sequence (SEQ, OPS, WIN, T1) and TCP options (T2-T7).


A list of the various tests used are provided in the mmap documentation at http://nmap.org/book/osdetect-methods.html#osdetect-probes-seq. For more specific details you'll have to read the source code.

Essentially various different probe packets are sent and the results analysed. Slight differences between TCP/IP software mean there are sometimes slight differences in the response to certain probe packets. With enough tests you can put together a fingerprint that identifies the OS very accurately.

An example might help. If you receive a TCP packet from a host, it contains a number of header options. The definition of the TCP protocol in RFC793 doesn't specify what order these options should be put in. So if the person who wrote the networking code for BeOS put those options in a particular order, and you receive a packet with the options in that order, then you can deduce that the packet might have come from a BeOS host. That may not enough on it's own to be sure it is BeOs; perhaps the TCP stack in Haiku puts the options in the same order. So you have to find some other test that gives different results for a packet from BeOS than it does from Haiku.

  • To expand on the probe packets: these work because the protocol only defines how communicating systems should response in normal communication. Most probe packets are nonsensical and break the established rules of communication and every system's software will react differently depending on how errors are handled. One example is the xmas or Christmas tree packet where every possible option in a packet is enabled (akin to the lights in a tree). This won't ever happen in normal communication but different system will respond (or stay silent) in different ways.
    – Lilienthal
    Commented Oct 16, 2014 at 14:27

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .