Skip to content
Tutorials Center
My learning

Topics

Lesson 19 of 36 ยท Service and OS Detection

OS Fingerprinting (-O)

Understand how Nmap -O guesses an operating system from TCP/IP behaviour: TTL, window size, options, accuracy and limits, with runnable scoring examples.

Different operating systems implement the TCP/IP protocols in slightly different ways. The standards leave many details open: the starting value of the time-to-live counter, the initial window size, which TCP options appear and in which order. These small choices form a fingerprint. Nmap's -O option sends a set of carefully built packets, measures the replies and compares the result with a database of known fingerprints.

OS detection is useful for asset inventories ("which of these 300 addresses are Windows servers?") and for spotting devices that should not exist on a network, such as an unexpected printer or camera. As always, use it only on networks you are authorised to scan.

Nmap and raw packet scanning cannot run in a web page. Example scan output is labelled as example output; the Python blocks are stand-ins that show how a fingerprint can be scored.

Running OS detection

OS detection needs raw packets, so on most systems it requires administrator rights (root, or the Windows Npcap driver with an elevated prompt):

Terminal
# Example command (read-only): OS detection against a lab host you own
sudo nmap -O 192.0.2.20

Example output (illustrative):

TEXT
Nmap scan report for 192.0.2.20
Host is up (0.00041s latency).
Not shown: 997 closed tcp ports (reset)
PORT    STATE SERVICE
22/tcp  open  ssh
80/tcp  open  http
443/tcp open  https
Device type: general purpose
Running: Linux 5.X|6.X
OS CPE: cpe:/o:linux:linux_kernel:5 cpe:/o:linux:linux_kernel:6
OS details: Linux 5.4 - 6.2
Network Distance: 1 hop

Notice the range "Linux 5.4 - 6.2". Nmap is honest about uncertainty: when several fingerprints match equally well, it lists them all.

What Nmap measures

Nmap needs at least one open and one closed TCP port to get a good fingerprint, because open and closed ports reply differently. It then sends a sequence of probes:

  • Six TCP SYN probes (called T1 to T7 in the database) with different option sets, to measure the sequence numbers, window size and options of the replies.
  • ICMP echo requests with unusual codes, to see how the stack answers.
  • TCP and UDP probes to closed ports, to see the shape of the reset or "port unreachable" reply.
  • IP ID and timestamp sampling, to check how the target numbers its packets.

From the replies Nmap builds a compact fingerprint, a list of tests such as SEQ, OPS, WIN, ECN, T1 to T7, U1 and IE. Each test has fields (for example W=FAF0 for the window size). The database scores your fingerprint against each known one and reports the best matches.

flowchart LR
    S["Scanner"] -->|"SYN probes, ICMP,<br/>closed-port probes"| T["Target stack"]
    T -->|"replies with TTL, window,<br/>options, IP ID"| S
    S --> F["Build fingerprint"]
    F --> D["Compare with<br/>nmap-os-db"]
    D --> R["Best matches plus<br/>confidence"]

The OS detection loop. More matching fields mean higher confidence.

Simple clues you can compute yourself

Before Nmap sends anything special, two very simple facts already help: the TTL and the window size of a normal reply. Systems start TTL at a round number (commonly 64, 128 or 255) and routers subtract 1 per hop, so the TTL you see is a little lower than the start value.

Python Stand-in: guess the initial TTL from what you see

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

These guesses are weak on their own: an administrator can change the TTL, and a device behind a NAT or proxy shows the proxy's behaviour. Nmap combines dozens of such signals, which is why it is far more reliable than TTL alone.

Scoring a fingerprint

The following stand-in shows the matching idea. Each known system has expected values for several fields. We count how many fields of an observation agree and turn the share into a confidence figure. The data are invented for teaching and are not Nmap's database.

Python Stand-in: score observed fields against known stacks

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Useful options

Option Effect
-O Turn on OS detection.
--osscan-limit Only try OS detection on hosts that have at least one open and one closed port. Saves time on big scans.
--osscan-guess / --fuzzy Print a guess even when the match is not perfect.
--max-os-tries 1 Try only once instead of up to five times. Faster, less reliable.
-A Aggressive bundle: -sV -O -sC --traceroute (see the next lessons).

Limits and good practice

  • Firewalls and NAT can change or hide the replies. If every port is filtered, Nmap cannot build a fingerprint and says No exact OS matches.
  • Virtual machines and containers show the stack of the host kernel or the hypervisor network layer, not of the guest's marketing name.
  • Modern Windows and Linux builds are close to each other in many fields, so versions are given as ranges.
  • Do not use a single OS guess to make decisions on its own. Combine it with service detection and your asset records.
  • Unexpected OS detection results are a good prompt to check your inventory: a "Linux" on an address you recorded as a printer deserves a look.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. Why does OS detection work?

    The standards leave details open, so each stack answers specially built probes in a slightly different way. Nmap compares these answers with a fingerprint database.
  2. 2. What does a host with all ports filtered usually produce under -O?

    Nmap needs at least one open and one closed TCP port for a good fingerprint. Without them the result is weak or absent.
  3. 3. You observe a TTL of 117. Which initial TTL is the most likely?

    Initial values are usually 64, 128 or 255. 117 is just below 128, which suggests about 11 hops from a system that starts at 128.
  4. 4. What does --osscan-limit do?

    It saves time on large scans by not attempting OS detection where it is likely to fail.

Summary

  • -O compares measured TCP/IP behaviour with a database of fingerprints and reports best matches with ranges.
  • It needs raw-packet privileges and, ideally, one open and one closed TCP port.
  • TTL and window size are weak clues alone; Nmap combines many signals.
  • Treat results as evidence to check against your inventory, not as proof.

Next: Banner Grabbing and Probes