Re: [PATCH] TCP Idle Scan in IPv6

[David Fifield <david () bamsoftware com>]

On Thu, Aug 15, 2013 at 02:12:45PM +0200, Mathias Morbitzer wrote:
> > The change to get_ipid_sequence needs some more thought. When you are
> > dealing with 16-bit IPv4 ID values, and you start treating them as
> > 32-bit, you will have calculation errors at the 16-bit boundary.
> > Imagine
> > two consecutive IP ID values are 0xfffe and 0x0001.
> > (u16) 0x0001 - (u16) 0xfffe == 3
> > but
> > (u32) 0x0001 - (u32) 0xfffe == 4294901763
> > Do you have ideas as to how to account for this? Perhaps you need to
> > sequence calculation to be aware of whether it is handling 16-bit or
> > 32-bit values.
>
> Thank you for pointing that out, I did not think about that. 
>
> As far as I understand, this concerns only these three lines:
> 235: ipid_diffs[i - 1] = ipids[i] - ipids[i - 1]; 
> 237: ipid_diffs[i - 1] = (u16) (ipids[i] - ipids[i - 1] + 65536);
> 261: ipid_diffs[i] -= 256;
> 263: ipid_diffs[i]--;
>
> Those are the only ones were calculations are done with the ipid_diffs[] values. 
> If so, I would just put a  "if (o.af() == AF_INET4) ..." before those
> lines, and try to handle the values correctly depending on if they are
> 16 or 32 bits, like this:
>
> 237 in IPv4: ipid_diffs[i - 1] = (u16) (ipids[i] - ipids[i - 1] + 65536); 
> 237 in IPv6: ipid_diffs[i - 1] = (u32) (ipids[i] - ipids[i - 1] + 4294967296);
>
> 261 in IPv4: ipid_diffs[i] = (u16) ipid_diffs[i] - 256;
> 261 in IPv6: ipid_diffs[i] -= 256;
>
> 263 in IPv4: ipid_diffs[i] = (u16) ipid_diffs - 1;
> 263 in IPv6: ipid_diffs[i]--;
>
> Let me know what you think about that approach. 

I don't like to rely on spooky global state (o.af()) in a utility
function like this one. If o.af() must be used, let it be at a higher
level. You can call get_ipid_sequence_16 or get_ipid_sequence_32. They
will have the same interface, taking an array of u32. But
get_ipid_sequence_16 will do the calculation as if the values are u16.
Both these functions could be implemented in terms of an auxiliary
function; in fact I think get_ipid_sequence_16 can be written as a
postprocessing step that applies boolean AND with 0xffff to the results
of get_ipid_sequence_32. (For instance 4294901763 == 0xffff0003.)

> > I saw you added a new IPID_SEQ_INCR_BY_2, which is probably the
> > right way to do it. However, not only idle scan uses the IPID_SEQ
> > classifications; OS detection uses them as well. Specifically, we
> > need to decide what to do with IPID_SEQ_INCR_BY_2 in
> > make_aval_ipid_seq. This is the code that controls what gets printed
> > for the SEQ.TI, SEQ.CI, and SEQ.II IPv4 OS detection tests:
> > http://nmap.org/book/osdetect-methods.html#osdetect-ti.
> > We have two options: one is to add a new value for "increment by 2",
> > for instance to make TI=2 possible; and the other is to find out
> > what hosts having "increment by 2" behavior are currently being
> > classified as, and make hosts that increment by 2 map to this value,
> > for instance TI=I or TI=R. What did get_ipid_sequence return for
> > hosts that increment by 2, before your patch causing them to be
> > IPID_SEQ_INCR_BY_2?
>
> Before my patch, get_ipid_sequence returned "Incremental" instead of
> "Incremental by 2". (Which caused the scan to fail because it was
> expecting lower increments) So, hosts which increment by 2 are
> currently classified as "incremental". 
>
> It should not be much of an issue to apply the second solution and map
> the hosts to this value, but in my opinion it would be nicer to add a
> value for "increment by 2". This might also allow a better OS
> detection, as we have (in IPv6) for example Windows 7 that increments
> by 2, and Windows Vista that increments by 1. Also, there are
> operating systems like windows 7 which increment by 1 in IPv4, and by
> 2 in IPv6, which should allows allow to limit the pool of possible OS
> for OS detection. 
>
> And not to forget, in previous research I stumbled across network
> printers which assigned the IPID incremental by 9. They should
> probably also get their own value at a certain point instead of
> mapping all the "increment by x" to the same, simple "incremental". 

The TI=I, CI=I, II=I values are only for IPv4 OS fingerprints. IPv6
fingerprints look different, they are basically just the packet
contents. See the difference:
http://nmap.org/book/osdetect-fingerprint-format.html#osdetect-ex-typical-subject-fprint
http://nmap.org/book/osdetect-fingerprint-format.html#osdetect-ex-typical-reference-fprint-ipv6
I agree with you that it would be best to record the size of the
increment in fingerprints. However, to do so means potentially
invalidating many of our existing IPv4 fingerprints.

So what I propose is to have make_aval_ipid_seq return "I" for
incremental by 2. This will maintain compatibility with existing IPv4
fingerprints. IPv6 OS detection won't be affected. Please provide a new
patch implementing this when you can.

David Fifield
_______________________________________________
Sent through the dev mailing list
http://nmap.org/mailman/listinfo/dev
Archived at http://seclists.org/nmap-dev/