Martin,

I think I am onto something here. Is it possible for the FTP server to
switch IP addresses mid stream? By that I mean between files. Without NAT
would the client be happy about it? In my last email, I think the dropped
packet has a source IP which is different from that in
/proc/net/ip_conntrack. I think that would certainly fool the NATing. I have
captured some data from another session using PORT mode which seems to
illustrate the point.

ip_conntrack reports

EXPECTING: proto=6 src=203.2.75.213 dst=203.164.4.37 sport=0 dport=2060

setup by the PORT command I think?

/var/log/messages reports

Feb 20 13:45:26 multimedia kernel: IN=eth1 OUT=
MAC=00:04:e2:10:72:2a:00:03:e3:31:b9:8c:08:00 SRC=203.2.75.28
DST=203.164.4.37 LEN=60 TOS=0x00 PREC=0x00 TTL=58 ID=34460 DF PROTO=TCP
SPT=18 DPT=2060 WINDOW=32120 RES=0x00 SYN URGP=0

This implies to me, I was expecting a response on 203.2.75.213 and actually
got one on 203.2.75.28

This is further supported by the two packets dropped in an earlier unrelated
FTP session shown below. If I read this right I have the same MAC address
reported as 203.2.75.29 and 203.2.75.28 an easy thing to do, but confusing
for NAT.

Feb 20 13:34:13 multimedia kernel: IN=eth1 OUT=
MAC=00:04:e2:10:72:2a:00:03:e3:31:b9:8c:08:00 SRC=203.2.75.29
DST=203.164.4.37 LEN=101 TOS=0x00 PREC=0x00 TTL=58 ID=42379 DF PROTO=TCP
SPT=19 DPT=1821 WINDOW=32120 RES=0x00 ACK PSH URGP=0
Feb 20 13:45:26 multimedia kernel: IN=eth1 OUT=
MAC=00:04:e2:10:72:2a:00:03:e3:31:b9:8c:08:00 SRC=203.2.75.28
DST=203.164.4.37 LEN=60 TOS=0x00 PREC=0x00 TTL=58 ID=34460 DF PROTO=TCP
SPT=18 DPT=2060 WINDOW=32120 RES=0x00 SYN URGP=0

If I am correct, can IPTABLES be setup to handle this type of situation?

Kind Regards

Steve McAllister

-----Original Message-----
From: Martin Josefsson [mailto:[EMAIL PROTECTED]]
Sent: Wednesday, 20 February 2002 6:53 AM
To: Steve McAllister
Cc: [EMAIL PROTECTED]
Subject: Re: Batching of FTP Transfers Fails


On Wed, 20 Feb 2002, Steve McAllister wrote:

> The kernel is 2.4.7-10 and the masquerading is done via IPTABLES 1.2.4-2.
I

> Module                  Size  Used by
> ip_conntrack_ftp        4080   0  (unused)
> ip_nat_ftp              3680   0  (unused)

I've seen problems like this but I can't remember which kernelversions. I
know there was a problem with the rewriting of the ACK number in packets
which have been modified, it sometimes calculated the wrong number and the
transfer dies. This has been fixed in later kernels, please upgrade your
kernel to 2.4.17 or something a lot more recent then that ancient 2.4.7
you are running.

/Martin

Never argue with an idiot. They drag you down to their level, then beat you
with experience.




Reply via email to