funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
 ·Verizon Online DSL
 ·Skype
| reply to MxxCon
Re: Can we block the reset packets?
said by MxxCon  :impossible and idiotic request. RTS packets are an integral part of TCP protocol. blocking them will completely destroy your internet connection. realtime detection methods are inaccurate so any block rule will have HUGE false positive rate. the only accurate way to detect faked RST packets is by comparing traffic from both sides and see if one generated the given RST packet to the other. After a 2-way connection has been established, a Reset (RST) flag should only be generated by the far end if it has crashed. However, all of the forged RST packets generated by Sandvine come in within the same second as a previous genuine packet from the remote host. This being the case, it would be possible to detect a bogus RST packet and react to it differently or discard it.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
HTTP is the new Bandwidth Hog...
|
|
MxxCon
join:1999-11-19
Brooklyn, NY
clubs:  
| you know perfectly well if the far end crashed, it wouldn't generate RST. it wouldn't generate anything at all.
RST would be generated if far end decided to close the existing connection, which can be a valid request.
some bittorrent clients want to connect only to seeds, so the moment they see it's not a seed, they will abort such connection generating RST. blocking such packets will break your application |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
1 edit | said by MxxCon :you know perfectly well With respect, your unnecessary attitude toward others does not promote friendliness. You called the previous poster's idea idiotic and you're being condescending to me.
In this case, you misunderstand how RST works. These are just facts, nobody wins or loses on the strength of their debating skills. It is whatever it is. We both learn about it by being here.
said by MxxCon :if the far end crashed, it wouldn't generate RST. it wouldn't generate anything at all. Right, but when network devices crash, they often reboot. When they come back up with no memory of the pre-crash states, then start receiving TCP packets on a closed port, they send RST in response.
This specific example is explained in RFC 793, Figure 11 (either page 34 or 35).
said by MxxCon :RST would be generated if far end decided to close the existing connection, which can be a valid request. If the far end decided to end the existing connection, FIN should be sent, not RST.
Anything is possible (there's an older Microsoft webserver that sends RST to kill the FIN's TIME_WAIT interval), but these exceptions are rare and usually pointed out as mistaken implementations.
said by MxxCon :some bittorrent clients want to connect only to seeds, so the moment they see it's not a seed, they will abort such connection generating RST. blocking such packets will break your application As said above, FIN not RST, is what should be sent. FIN completes the connection gracefully. RST can cause the loss of sent-but-unacknowledged data.
You can see this using Wireshark.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
HTTP is the new Bandwidth Hog...
|
|
Hehe
@ssa.gov
| Well, looking at the RFC it seems like we should not receive a RST while the connection is in the ESTABLISHED STATE. So, maybe we can drop RST if ESTABLISHED?
Anyone know how to make iptables do that?
And I assume I am not 100% correct! 
It just can't be that easy. |
|
funchords
Hello
Premium,MVM
join:2001-03-11
Washington, DC
·Verizon Online DSL
·Skype
| said by Hehe :Well, looking at the RFC it seems like we should not receive a RST while the connection is in the ESTABLISHED STATE. So, maybe we can drop RST if ESTABLISHED? Anyone know how to make iptables do that? And I assume I am not 100% correct! It just can't be that easy. RFC 793, Figure 11 would happen with one end in the Established state, thinking it was in a full-open connection when it actually is only in a half-open connection due to the remote host's reboot.
--
Robb Topolski -= funchords.com =- Hillsboro, Oregon
HTTP is the new Bandwidth Hog...
|
|
