NAT in an already established TCP connection

NAT in an already established TCP connection

[Diego Casado Mansilla]

Hello all!!!

This is my first mail in the list.

Hopefully the question is interesting and you can figure out how to help me.

I use iptables rules to manage the connections from internet to my local 
network. I know how to filter, do nat, etc...
But this days I'm trying to do NAT in connections that are already 
established. The problem is (as far as I know) the packets which pass 
throught the nat table are only the SYN packets (once), thus, the 
packets that are used to perform a NEW connection.

After that the connection is created, the maintenance and the resolution 
of the SNAT and DNAT are kept till the connection finish.
What I'm wondering is: how can I change the ports or IPs of an already 
established connection if my packets just go throught the nat table at 
the connection time?

**** Maybe doing packets' replication since those ones are redirected to 
annother machine?

**** NAT TCP Extensions??Patch-O-Matic --> window-tracking??

**** I read this in an interntet site:

--- NEW (and RELATED non-icmp)
    This is a very important part relevant for understanding the whole NAT
    subsystem. Only if the packet has the state NEW (i.e. it would establish
    a new connection, if we'd accept it), the NAT table is traversed by
    calling ip_nat_rule.c:ip_nat_rule_find(), which in turn calls
    ip_tables.c:ipt_do_table() for the actual IP table traversal. The 
traversal
    ends up in either ACCEPTing the packet as it is, or one of the nat 
targets
    (SNAT, DNAT and if loaded: REDIRECT, MASQUERADE) Please see
    chapter FIXME for further description of those targets.

--- ESTABLISHED
    This packet belongs to an already established connection. We don't need
    to traverse the NAT table again, as the necessary information
    (struct ip_nat_info) was already gained Hello everybody,


Thank you very much in advance and if my questions are not clear don't 
doubt to send me a message.

Diego.

Re: NAT in an already established TCP connection

[Julien Vehent]

On Tue, 28 Oct 2008 12:25:13 +0100, Diego Casado Mansilla
<diego.casadomansilla@gmail.com> wrote:
> Hello all!!!
> 
> This is my first mail in the list.
> 
> Hopefully the question is interesting and you can figure out how to help
> me.
> 
> I use iptables rules to manage the connections from internet to my local 
> network. I know how to filter, do nat, etc...
> But this days I'm trying to do NAT in connections that are already 
> established. The problem is (as far as I know) the packets which pass 
> throught the nat table are only the SYN packets (once), thus, the 
> packets that are used to perform a NEW connection.
> 
> After that the connection is created, the maintenance and the resolution 
> of the SNAT and DNAT are kept till the connection finish.
> What I'm wondering is: how can I change the ports or IPs of an already 
> established connection if my packets just go throught the nat table at 
> the connection time?
> 

What you want is to redirect an existing connection to a new destination.
If you use TCP protocol, the only way to do that is to record the current
connection and, in parralel, create a second connection to you new
destination and replay the payload on the packet in the new one.

If you use UDP, this doesn't apply since there's no connection tracking in
the UDP protocol. Netfilter, however, does some connection tracking on UDP
packets, so make some test to see if it's doable.

I had to solve the same problem in a honeypot project to redirect active
connections from low interaction honeypots to high interactions honeypots.
The solution I choosed was to queue the connections in userland using
netfilter_queue, process them and replay those I've selected to the new
destination in parrallel, and then drop the packet from the initial
connection.

It's tricky to do, and there's many issue to solve, but AFAIK this is the
most reliable solution.

> **** Maybe doing packets' replication since those ones are redirected to 
> annother machine?

Note : window-tracking is for the tracking of the window size in the tcp
header, it has nothing to do with this.


> 
> **** NAT TCP Extensions??Patch-O-Matic --> window-tracking??
> 
> **** I read this in an interntet site:
> 
> --- NEW (and RELATED non-icmp)
>     This is a very important part relevant for understanding the whole
NAT
>     subsystem. Only if the packet has the state NEW (i.e. it would
>     establish
>     a new connection, if we'd accept it), the NAT table is traversed by
>     calling ip_nat_rule.c:ip_nat_rule_find(), which in turn calls
>     ip_tables.c:ipt_do_table() for the actual IP table traversal. The 
> traversal
>     ends up in either ACCEPTing the packet as it is, or one of the nat 
> targets
>     (SNAT, DNAT and if loaded: REDIRECT, MASQUERADE) Please see
>     chapter FIXME for further description of those targets.
> 
> --- ESTABLISHED
>     This packet belongs to an already established connection. We don't
need
>     to traverse the NAT table again, as the necessary information
>     (struct ip_nat_info) was already gained Hello everybody,
> 
> 
> Thank you very much in advance and if my questions are not clear don't 
> doubt to send me a message.
> 
> Diego.
> 
> --
> To unsubscribe from this list: send the line "unsubscribe netfilter" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 



Regards,
Julien

-- 
www.linuxwall.info

Re: NAT in an already established TCP connection

[Jan Engelhardt]

On Tuesday 2008-10-28 08:53, Julien Vehent wrote:
>On Tue, 28 Oct 2008 12:25:13 +0100, Diego Casado Mansilla wrote:

>> But this days I'm trying to do NAT in connections that are already 
>> established. The problem is (as far as I know) the packets which pass 
>> throught the nat table are only the SYN packets (once), thus, the 
>> packets that are used to perform a NEW connection.

This is so wrong. Any connection that *Netfilter* (not specifically
the networking stack) does not yet know about is NEW. That might
even be in the middle of a TCP stream -- you can do it: conntrack -F
will clear all connections and thus they will begin with NEW. Not
that this is something you randomly do on a router which only allows
ESTABLISHED or NEW-combined-with-SYN -matched packets.

>> After that the connection is created, the maintenance and the resolution 
>> of the SNAT and DNAT are kept till the connection finish.
>> What I'm wondering is: how can I change the ports or IPs of an already 
>> established connection if my packets just go throught the nat table at 
>> the connection time?

Well they don't go through, hence this is moot. The TCP stack of
the final destination will also be confused. Different port --
different connection. Basic rules of networking.

>If you use UDP, this doesn't apply since there's no connection tracking in
>the UDP protocol. Netfilter, however, does some connection tracking on UDP
>packets, so make some test to see if it's doable.

Same here. A (src, sourceport, dst, dstport) tuple uniquely identifies a
connection. If you change something in this tuple, it is, by definition,
a new one.

Re: NAT in an already established TCP connection

[Diego Casado Mansilla]

Hi all,

Julien Vehent wrote:
> On Tue, 28 Oct 2008 12:25:13 +0100, Diego Casado Mansilla
> <diego.casadomansilla@gmail.com> wrote:
>   
>> Hello all!!!
>>
>> This is my first mail in the list.
>>
>> Hopefully the question is interesting and you can figure out how to help
>> me.
>>
>> I use iptables rules to manage the connections from internet to my local 
>> network. I know how to filter, do nat, etc...
>> But this days I'm trying to do NAT in connections that are already 
>> established. The problem is (as far as I know) the packets which pass 
>> throught the nat table are only the SYN packets (once), thus, the 
>> packets that are used to perform a NEW connection.
>>
>> After that the connection is created, the maintenance and the resolution 
>> of the SNAT and DNAT are kept till the connection finish.
>> What I'm wondering is: how can I change the ports or IPs of an already 
>> established connection if my packets just go throught the nat table at 
>> the connection time?
>>
>>     
>
> What you want is to redirect an existing connection to a new destination.
> If you use TCP protocol, the only way to do that is to record the current
> connection and, in parralel, create a second connection to you new
> destination and replay the payload on the packet in the new one.
>
> If you use UDP, this doesn't apply since there's no connection tracking in
> the UDP protocol. Netfilter, however, does some connection tracking on UDP
> packets, so make some test to see if it's doable.
>
> I had to solve the same problem in a honeypot project to redirect active
> connections from low interaction honeypots to high interactions honeypots.
> The solution I choosed was to queue the connections in userland using
> netfilter_queue, process them and replay those I've selected to the new
> destination in parrallel, and then drop the packet from the initial
> connection.
>
> It's tricky to do, and there's many issue to solve, but AFAIK this is the
> most reliable solution.
>   
I have also thought in that solution (or something like that), but I was 
expecting something easier with iptables-conntrack... :o).
So what I will do is to create a new connection between a C'/S' in my 
local machine, and when a packet from the real Server go trouhgt my 
transparent-proxi, I will enqueue it, record the pay-load, do something 
"tricky" with the ack and seq. number (I think that a substraction 
should be enought) and use all this stuff to send the data from S' to 
the Real Client...

After that the client replies with an ACK to the real server...thus.. I 
have to do a replication of that ACK in order to send one copy to the 
real Server and annother one to the S' in order to keep the 
TCP-connection state and flow control.
So...Do I must to use annother time the QUEUE to do so?? I think so.
>   
>> **** Maybe doing packets' replication since those ones are redirected to 
>> annother machine?
>>     
>
> Note : window-tracking is for the tracking of the window size in the tcp
> header, it has nothing to do with this.
>
>
>   
>> **** NAT TCP Extensions??Patch-O-Matic --> window-tracking??
>>
>> **** I read this in an interntet site:
>>
>> --- NEW (and RELATED non-icmp)
>>     This is a very important part relevant for understanding the whole
>>     
> NAT
>   
>>     subsystem. Only if the packet has the state NEW (i.e. it would
>>     establish
>>     a new connection, if we'd accept it), the NAT table is traversed by
>>     calling ip_nat_rule.c:ip_nat_rule_find(), which in turn calls
>>     ip_tables.c:ipt_do_table() for the actual IP table traversal. The 
>> traversal
>>     ends up in either ACCEPTing the packet as it is, or one of the nat 
>> targets
>>     (SNAT, DNAT and if loaded: REDIRECT, MASQUERADE) Please see
>>     chapter FIXME for further description of those targets.
>>
>> --- ESTABLISHED
>>     This packet belongs to an already established connection. We don't
>>     
> need
>   
>>     to traverse the NAT table again, as the necessary information
>>     (struct ip_nat_info) was already gained Hello everybody,
>>
>>
>> Thank you very much in advance and if my questions are not clear don't 
>> doubt to send me a message.
>>
>> Diego.
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe netfilter" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
>>     
>
>
>
> Regards,
> Julien
>
>   
Thanks and regards,
Diego.

Re: NAT in an already established TCP connection

[Julien Vehent]

Hi,


On Wed, 29 Oct 2008 11:01:04 +0100, Diego Casado Mansilla
<diego.casadomansilla@gmail.com> wrote:
> Hi all,
> 
> Julien Vehent wrote:
>> On Tue, 28 Oct 2008 12:25:13 +0100, Diego Casado Mansilla
>> <diego.casadomansilla@gmail.com> wrote:
>>   
>>> Hello all!!!
>>>
>>> This is my first mail in the list.
>>>
>>> Hopefully the question is interesting and you can figure out how to
help
>>> me.
>>>
>>> I use iptables rules to manage the connections from internet to my
local
>>>
>>> network. I know how to filter, do nat, etc...
>>> But this days I'm trying to do NAT in connections that are already 
>>> established. The problem is (as far as I know) the packets which pass 
>>> throught the nat table are only the SYN packets (once), thus, the 
>>> packets that are used to perform a NEW connection.
>>>
>>> After that the connection is created, the maintenance and the
resolution
>>>
>>> of the SNAT and DNAT are kept till the connection finish.
>>> What I'm wondering is: how can I change the ports or IPs of an already 
>>> established connection if my packets just go throught the nat table at 
>>> the connection time?
>>>
>>>     
>>
>> What you want is to redirect an existing connection to a new
destination.
>> If you use TCP protocol, the only way to do that is to record the
current
>> connection and, in parralel, create a second connection to you new
>> destination and replay the payload on the packet in the new one.
>>
>> If you use UDP, this doesn't apply since there's no connection tracking
>> in
>> the UDP protocol. Netfilter, however, does some connection tracking on
>> UDP
>> packets, so make some test to see if it's doable.
>>
>> I had to solve the same problem in a honeypot project to redirect active
>> connections from low interaction honeypots to high interactions
>> honeypots.
>> The solution I choosed was to queue the connections in userland using
>> netfilter_queue, process them and replay those I've selected to the new
>> destination in parrallel, and then drop the packet from the initial
>> connection.
>>
>> It's tricky to do, and there's many issue to solve, but AFAIK this is
the
>> most reliable solution.
>>   
> I have also thought in that solution (or something like that), but I was 
> expecting something easier with iptables-conntrack... :o).
> So what I will do is to create a new connection between a C'/S' in my 
> local machine, and when a packet from the real Server go trouhgt my 
> transparent-proxi, I will enqueue it, record the pay-load, do something 
> "tricky" with the ack and seq. number (I think that a substraction 
> should be enought) and use all this stuff to send the data from S' to 
> the Real Client...
> 
> After that the client replies with an ACK to the real server...thus.. I 
> have to do a replication of that ACK in order to send one copy to the 
> real Server and annother one to the S' in order to keep the 
> TCP-connection state and flow control.
> So...Do I must to use annother time the QUEUE to do so?? I think so.


It's even more tricky than this, because if you wait for the ACK of the
client, the delay may be too long and you will receive retries from the new
server. So you need to ack this packet before you forward it, and then drop
the real ack received.

Also, this needs to be done using raw packets, because you will have to
manipulate all the header. So, to make it short, you will have to code an
almost complete TCP/IP stack to use in your queue...


Have fun !

>>   
>>> **** Maybe doing packets' replication since those ones are redirected
to
>>>
>>> annother machine?
>>>     
>>
>> Note : window-tracking is for the tracking of the window size in the tcp
>> header, it has nothing to do with this.
>>
>>
>>   
>>> **** NAT TCP Extensions??Patch-O-Matic --> window-tracking??
>>>
>>> **** I read this in an interntet site:
>>>
>>> --- NEW (and RELATED non-icmp)
>>>     This is a very important part relevant for understanding the whole
>>>     
>> NAT
>>   
>>>     subsystem. Only if the packet has the state NEW (i.e. it would
>>>     establish
>>>     a new connection, if we'd accept it), the NAT table is traversed by
>>>     calling ip_nat_rule.c:ip_nat_rule_find(), which in turn calls
>>>     ip_tables.c:ipt_do_table() for the actual IP table traversal. The 
>>> traversal
>>>     ends up in either ACCEPTing the packet as it is, or one of the nat 
>>> targets
>>>     (SNAT, DNAT and if loaded: REDIRECT, MASQUERADE) Please see
>>>     chapter FIXME for further description of those targets.
>>>
>>> --- ESTABLISHED
>>>     This packet belongs to an already established connection. We don't
>>>     
>> need
>>   
>>>     to traverse the NAT table again, as the necessary information
>>>     (struct ip_nat_info) was already gained Hello everybody,
>>>
>>>
>>> Thank you very much in advance and if my questions are not clear don't 
>>> doubt to send me a message.
>>>
>>> Diego.
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe netfilter" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>>>     
>>
>>
>>
>> Regards,
>> Julien
>>
>>   
> Thanks and regards,
> Diego.
> 


Regards,

Julien


-- 
www.linuxwall.info

[OffTopic] Re: NAT in an already established TCP connection

[Diego Casado Mansilla]

Hi all,

Julien Vehent wrote:
> Hi,
>
>
> On Wed, 29 Oct 2008 11:01:04 +0100, Diego Casado Mansilla
> <diego.casadomansilla@gmail.com> wrote:
>   
>> Hi all,
>>
>> Julien Vehent wrote:
>>     
>>> On Tue, 28 Oct 2008 12:25:13 +0100, Diego Casado Mansilla
>>> <diego.casadomansilla@gmail.com> wrote:
>>>   
>>>       
>>>> Hello all!!!
>>>>
>>>> This is my first mail in the list.
>>>>
>>>> Hopefully the question is interesting and you can figure out how to
>>>>         
> help
>   
>>>> me.
>>>>
>>>> I use iptables rules to manage the connections from internet to my
>>>>         
> local
>   
>>>> network. I know how to filter, do nat, etc...
>>>> But this days I'm trying to do NAT in connections that are already 
>>>> established. The problem is (as far as I know) the packets which pass 
>>>> throught the nat table are only the SYN packets (once), thus, the 
>>>> packets that are used to perform a NEW connection.
>>>>
>>>> After that the connection is created, the maintenance and the
>>>>         
> resolution
>   
>>>> of the SNAT and DNAT are kept till the connection finish.
>>>> What I'm wondering is: how can I change the ports or IPs of an already 
>>>> established connection if my packets just go throught the nat table at 
>>>> the connection time?
>>>>
>>>>     
>>>>         
>>> What you want is to redirect an existing connection to a new
>>>       
> destination.
>   
>>> If you use TCP protocol, the only way to do that is to record the
>>>       
> current
>   
>>> connection and, in parralel, create a second connection to you new
>>> destination and replay the payload on the packet in the new one.
>>>
>>> If you use UDP, this doesn't apply since there's no connection tracking
>>> in
>>> the UDP protocol. Netfilter, however, does some connection tracking on
>>> UDP
>>> packets, so make some test to see if it's doable.
>>>
>>> I had to solve the same problem in a honeypot project to redirect active
>>> connections from low interaction honeypots to high interactions
>>> honeypots.
>>> The solution I choosed was to queue the connections in userland using
>>> netfilter_queue, process them and replay those I've selected to the new
>>> destination in parrallel, and then drop the packet from the initial
>>> connection.
>>>
>>> It's tricky to do, and there's many issue to solve, but AFAIK this is
>>>       
> the
>   
>>> most reliable solution.
>>>   
>>>       
>> I have also thought in that solution (or something like that), but I was 
>> expecting something easier with iptables-conntrack... :o).
>> So what I will do is to create a new connection between a C'/S' in my 
>> local machine, and when a packet from the real Server go trouhgt my 
>> transparent-proxi, I will enqueue it, record the pay-load, do something 
>> "tricky" with the ack and seq. number (I think that a substraction 
>> should be enought) and use all this stuff to send the data from S' to 
>> the Real Client...
>>
>> After that the client replies with an ACK to the real server...thus.. I 
>> have to do a replication of that ACK in order to send one copy to the 
>> real Server and annother one to the S' in order to keep the 
>> TCP-connection state and flow control.
>> So...Do I must to use annother time the QUEUE to do so?? I think so.
>>     
>
>
> It's even more tricky than this, because if you wait for the ACK of the
> client, the delay may be too long and you will receive retries from the new
> server. So you need to ack this packet before you forward it, and then drop
> the real ack received.
>
> Also, this needs to be done using raw packets, because you will have to
> manipulate all the header. So, to make it short, you will have to code an
> almost complete TCP/IP stack to use in your queue...
>
>   
Bufff....sounds interesting. But I think there's something that I don't 
get very well.
Outline:

       ** C-S Connection "sockets TCP".
       ** Proxi store the connection info (iptables FORWARD -j QUEUE)
       ** Proxi wants to replace Server, so, a new TCP Connexion is 
created locally C'/S'.
       ** C wants a content, hence, socket C -->recv();
       ** S-->send(content); proxi catchs payload and drops.
       ** S' do send with new payload (new port, IP, ack, seq). And 
obviously C' do nothing.
       ** C receives the content, since It wants to ACK to S, then it 
send ACK.(S doesn't realize their packets are drop)
       ** Also S' wants the ACK...to maintain both flow control.
                   So.....I don't understand very well the raw packets 
that you has suggested me.
                   I was thinking to get the ack reply  and a the same 
trick as before with the original payload.
                   Do you suggest me to create myself a new ack packet 
in my local machine to ACK the content what S' has sent?
                   Raw packets.....The kernel doesn't process them and I 
can create my own IP/TCP headers, isn't it?
       To summarize...one Client-2 servers, nobody knows who is in 
communication with whom, but the flow control is maintained.

Thanks.
> Have fun !
>
>   
>>>   
>>>       
>>>> **** Maybe doing packets' replication since those ones are redirected
>>>>         
> to
>   
>>>> annother machine?
>>>>     
>>>>         
>>> Note : window-tracking is for the tracking of the window size in the tcp
>>> header, it has nothing to do with this.
>>>
>>>
>>>   
>>>       
>>>> **** NAT TCP Extensions??Patch-O-Matic --> window-tracking??
>>>>
>>>> **** I read this in an interntet site:
>>>>
>>>> --- NEW (and RELATED non-icmp)
>>>>     This is a very important part relevant for understanding the whole
>>>>     
>>>>         
>>> NAT
>>>   
>>>       
>>>>     subsystem. Only if the packet has the state NEW (i.e. it would
>>>>     establish
>>>>     a new connection, if we'd accept it), the NAT table is traversed by
>>>>     calling ip_nat_rule.c:ip_nat_rule_find(), which in turn calls
>>>>     ip_tables.c:ipt_do_table() for the actual IP table traversal. The 
>>>> traversal
>>>>     ends up in either ACCEPTing the packet as it is, or one of the nat 
>>>> targets
>>>>     (SNAT, DNAT and if loaded: REDIRECT, MASQUERADE) Please see
>>>>     chapter FIXME for further description of those targets.
>>>>
>>>> --- ESTABLISHED
>>>>     This packet belongs to an already established connection. We don't
>>>>     
>>>>         
>>> need
>>>   
>>>       
>>>>     to traverse the NAT table again, as the necessary information
>>>>     (struct ip_nat_info) was already gained Hello everybody,
>>>>
>>>>
>>>> Thank you very much in advance and if my questions are not clear don't 
>>>> doubt to send me a message.
>>>>
>>>> Diego.
>>>>
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe netfilter" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>
>>>>     
>>>>         
>>>
>>> Regards,
>>> Julien
>>>
>>>   
>>>       
>> Thanks and regards,
>> Diego.
>>
>>     
>
>
> Regards,
>
> Julien
>
>
>   
Regards,

Diego.