Re: Using TTL value of response packets on nmap port scans.

[Otto Airamo <otairamo () kapsi fi>]

Hello,

It seems that I was explaining in the details WHAT I would like from Nmap,
but I was missing WHY this is needed.

When port scan is executed over TCP, there is in most cases three relevant
cases that may occur in each port that is scanned.

1)
SYN-ACK packet is received. In most cases this means that port is open in
target host.

2)
No reply. In this case we do not know if target host is listening port. At
least it is known that from this scope* this port is not accessible in
target host. (*With scope I mean that either network firewall or personal
firewall blocks incoming SYN/outgoing RST/ICMP error message. It is
possible that port is open in target host and would answer with SYN-ACK if
packet would come from, for example, same network or different IP address)

Important point here is that WE DO NOT KNOW if port is actually open (from
different scope)!

3) RST packet is received. Typically this means that port is closed in
target host. If RST packet is generated by target host TCP/IP stack
because port is closed, it does not matter what is the scope where we are
performing scan. Result will remain same even if scan would be performed
from same network/using any source IP address. At the same time RST packet
may be generated by firewall located between scanner and target host. If
scanner is not able to tell if RST packet is generated by target host or
device between, result of the scan will fall into same category as case
#2. (We do not know if port is open)

If scanner would be able to detect difference between RST packet generated 
by target host TCP/IP stack and other cases, scanner would have better 
knowledge of target.

Options are:
a) Target host TCP/IP stack generated RST packet => we know that port is
closed
b) Somebody else than target host TCP/IP stack generated RST packet => we
do not know if port is actually open even we do not have access from this
scope.

Now we have following network setup:

scanner -- FW -- target

Firewall is configured to RST connections to port 22. Some ports are
allowd, mosts of the ports are dropped silently.

Result of nmap could be:
# nmap --reason target --top-ports=10
Interesting ports on target:
PORT     STATE    SERVICE      REASON           TTL
21/tcp   open     ftp          syn-ack          62
22/tcp   refused  ssh          reset (ttl?)     63
23/tcp   closed   telnet       reset            62
25/tcp   filtered smtp         no-response      N/A
80/tcp   open     http         syn-ack          62
110/tcp  filtered pop3         no-response      N/A
139/tcp  filtered netbios-ssn  no-response      N/A
443/tcp  filtered https        no-response      N/A
445/tcp  filtered microsoft-ds no-response      N/A
3389/tcp filtered ms-term-serv no-response      N/A

> From this result penetrator tester can assume that telnet port is closed
in target host (TTL is same in RST packet as it is in syn-ack packet). At
the same time ssh port seems interesting. Port might still be open even
RST packet is received. I think that current nmap implementation works
incorrectly by calling port closed even it seems evident that other device
is resetting connection.

I dont know what would be best term under STATE and REASON for this
situation. Now I used terms refused and reset(ttl?). Those terms are not
optimal and need to be renamed something better.

Does this make it more clear what I'm trying to propose?

Regards,
Otto Airamo

On Wed, 11 Apr 2012, David Fifield wrote:

> On Wed, Apr 11, 2012 at 07:52:22AM +0300, Otto Airamo wrote:
> > Hello,
> >
> > Currently Nmap is not detecting, by default, if there is
> > firewall/IDS resetting connections between target host/network and
> > scanner.
> >
> > Let's taken an example:
> > Between scanner and target host, there is a firewall blocking all
> > other ports than 80 and 443. Firewall refuses connections by sending
> > RST. Currently result of the scan is that ports 80 and 443 are open,
> > rest are closed. If Nmap would follow TTL values, it could trivially
> > detect that there is third party device generating RST packets as
> > TTL value of responses differs.
> >
> > I'm aware that Nmap already provides " -badsum" options to solve
> > same problem, but I believe that my suggested approach is more
> > elegant way to solve the problem. At least when option -sS is used
> > (raw sockets), it would be very easy to get TTL value of the packet
> > and do some automatic detection if there is firewall or IDS.
> > (Actually detecting firewall should be 100% sure. If firewall and
> > target host uses same initial TTL value, firewall should have at one
> > or more higher TTL. Detection with IDS is not as sure. IDS might
> > have same TTL value if it is located into same layer-two segment as
> > target host)
> >
> > I'm aware that there can be scenarios where altering TTL values from
> > single host do not automatically mean FW/IPS. Certain load balancing
> > scenarios and NAT decisions based on port information are likely to
> > produce results where port scan to single IP address will produce
> > reply packets with different TTL values. Reply packets may also be
> > routed via different route when TTL value will not be same.
>
> Nmap does in fact keep track of TTLs of response packets. I think it is
> only exposed in the XML output with --reason, though:
>
> # nmap --reason nmap.org -F -oX -
> <port protocol="tcp" portid="80"><state state="open" reason="syn-ack" reason_ttl="53"/>
>
> I'm afraid I don't understand what you're proposing. A new output for
> Nmap? A script that processes the TTL values?
>
> It would be very helpful if you could show us an example of what Nmap
> output would look like with your proposed change.
>
> David Fifield
> _______________________________________________
> Sent through the nmap-dev mailing list
> http://cgi.insecure.org/mailman/listinfo/nmap-dev
> Archived at http://seclists.org/nmap-dev/
_______________________________________________
Sent through the nmap-dev mailing list
http://cgi.insecure.org/mailman/listinfo/nmap-dev
Archived at http://seclists.org/nmap-dev/