Need guru help with data xfer corruption problem

I am experiencing a very odd problem, and I can't figure
out what could possibly be causing it.

Here's my network setup:

[ADSL Modem]
^
|
v
[ RedHat Linux 7.2 ]
[ PPPoE client ]
[Firewall (iptables)]
[ w/masquerading ]
[ DHCP, DNS, Samba ]
^
|
v
[Switch]
^
|
v
[Win2K SP4] x 2

In other words, a classic application of a Linux box as
a DHCP server/firewall/NAT/router/etc.

The problem is that I am experiencing corruption when
transferring data from the Internet to the Win2K boxes,
but not to the Linux box.

For example, if I download the same file to the Win2K
boxes and to the Linux box (using wget in all cases,)
all the transfers succeed, all the transferred files
are the same length, but the files downloaded to the
Win2K boxes are corrupt 80% of the time while that
received by the Linux box is OK 100% of the time.

The MTU setting for ppp0 on the Linux box is correct
(1492). The MTU for the ethernet connection from the
Linux box to the switch is 1500, but I tried changing it
to 1492 and that made no difference. Nor did changing
the MTU for the ethernet connections from the Win2K boxes
to the switch. PPPoE is configured to clamp MSS to MTU.
I also played with Path MTU Discovery on the Win2K boxes,
to no avail.

For the time being, I have implemented a work-around
by running Squid on the Linux box as a non-caching
proxy and all the problems have gone away.

But I would really like to find out what is going on.

Eternal gratitude to anyone who can figure this one out!

MV

--
Do not send e-mail to the above address. I do not read e-mail sent there.

In article <(E-Mail Removed) >, Martin Vuille wrote:

> all the transfers succeed, all the transferred files
> are the same length, but the files downloaded to the
> Win2K boxes are corrupt 80% of the time while that
> received by the Linux box is OK 100% of the time.

Dodgy network cable/card or switch port..

Phil

--
Phil Launchbury, Network/Unix Sysadmin
Triumph Tiger 955i
'Do not meddle in the affairs of sysadmins, for they
are quick to anger and have no need of subtlety'
Remove sick person to email me

I would of thought the same however the files would not arrive corrupt, only
download slowly, as the CRC checks in TCP would reject the packet; unless
Win2k is REALLY borked However I could be wrong.....

I am unsure whether you can do it, but can you do the equilivent of
'ifconfig' and get the output that mentions any ethernet frame errors or not?
I do not think 'ipconfig /all' can tell you this.

Regards

Alex

In article <(E-Mail Removed)> , Phil Launchbury wrote:
> In article <(E-Mail Removed) >, Martin Vuille wrote:
>
>> all the transfers succeed, all the transferred files
>> are the same length, but the files downloaded to the
>> Win2K boxes are corrupt 80% of the time while that
>> received by the Linux box is OK 100% of the time.
>
> Dodgy network cable/card or switch port..
>
> Phil
>

Alexander Clouter <alexander.junk-(E-Mail Removed)> wrote in message news:<rgesfb.rbc.ln@127.0.0.1>...
> I would of thought the same however the files would not arrive corrupt, only
> download slowly, as the CRC checks in TCP would reject the packet; unless
> Win2k is REALLY borked However I could be wrong.....
>
> I am unsure whether you can do it, but can you do the equilivent of
> 'ifconfig' and get the output that mentions any ethernet frame errors or not?
> I do not think 'ipconfig /all' can tell you this.
>
> Regards
>
> Alex

If it were a hardware problem on the LAN side of the Linux box,
then it seems to me that proxying the requests through Squid
wouldn't make any difference: the transfers would still fail.

Two additional pieces of information:

- File transfers from the Linux box to the Win2k boxes are 100%
reliable. I move a lot of data via SMB and have run into no
errors.

- I monitored both successful and corrupt data transfers by using
ethereal and capturing the messaging over ppp0. The data arrives
down the PPPoE pipe already corrupt, so if there's a hardware
problem in my equipment, it would have to be in the ADSL modem
or ethernet card. But then why would it only affect transfers
to the Win2k boxes and not to the Linux box?

MV

I was thinking the same. How are you down loading the files, via FTP?
Unlikely but probably worth trying another FTP client on the windoze box,
LeechFTP is my favourite when I have to endure M$ 'Goodness'
(http://stud.fh-heilbronn.de/~jdebis/...les/lftp13.zip 620k).

I would be surprised if the arrival of corrupt packets was due to your PPPoE
interface and the remote end's exchange; otherwise the packet would of been
dropped and never presented to your LAN, or rather should of not been. What
are the differences in identical downloads, compare the linux downloaded file
to the windows one. Is there a pattern, is it only a few bytes or is the
whole thing toasted. If the whole thing is toasted you might be downloading
binary files in text mode (if using FTP).
If you are downloading the files in the 'correct' manner then it
would start pointing to a strange device beyond your exchange that goes 'hmmm
windoze packets....corrupt and adjust CRCs....'; where you would get such a
device I could only dream about.....

Run 'tethereal' on your gateway box and compare the two downloads; if you do
not understand how to use it, get in touch and we can run the diagnostics and
I can interpret them for you.

Regards

Alex

In article <(E-Mail Removed) >, Martin Vuille wrote:
> Alexander Clouter <alexander.junk-(E-Mail Removed)> wrote in message news:<rgesfb.rbc.ln@127.0.0.1>...
>> I would of thought the same however the files would not arrive corrupt, only
>> download slowly, as the CRC checks in TCP would reject the packet; unless
>> Win2k is REALLY borked However I could be wrong.....
>>
>> I am unsure whether you can do it, but can you do the equilivent of
>> 'ifconfig' and get the output that mentions any ethernet frame errors or not?
>> I do not think 'ipconfig /all' can tell you this.
>>
>> Regards
>>
>> Alex
>
> If it were a hardware problem on the LAN side of the Linux box,
> then it seems to me that proxying the requests through Squid
> wouldn't make any difference: the transfers would still fail.
>
> Two additional pieces of information:
>
> - File transfers from the Linux box to the Win2k boxes are 100%
> reliable. I move a lot of data via SMB and have run into no
> errors.
>
> - I monitored both successful and corrupt data transfers by using
> ethereal and capturing the messaging over ppp0. The data arrives
> down the PPPoE pipe already corrupt, so if there's a hardware
> problem in my equipment, it would have to be in the ADSL modem
> or ethernet card. But then why would it only affect transfers
> to the Win2k boxes and not to the Linux box?
>
> MV

Alexander Clouter <alexander.junk-(E-Mail Removed)> wrote
in newsjvtfb.udf.ln@127.0.0.1:

> I was thinking the same. How are you down loading the files, via FTP?
> Unlikely but probably worth trying another FTP client on the windoze
> box, LeechFTP is my favourite when I have to endure M$ 'Goodness'
> (http://stud.fh-heilbronn.de/~jdebis/...les/lftp13.zip 620k).
>

Most of my downloads are via HTTP. For test purposes, I have been using
wget on both Linux and Win2k to eliminate the possibility of this being
a client-related issue.

>
> Run 'tethereal' on your gateway box and compare the two downloads; if
> you do not understand how to use it, get in touch and we can run the
> diagnostics and I can interpret them for you.

I did do some captures with ethereal, and posted the results in an
earlier message, which I quote below:

>> - I monitored both successful and corrupt data transfers by using
>> ethereal and capturing the messaging over ppp0. The data arrives
>> down the PPPoE pipe already corrupt, [...]

I have not had a chance to review the messaging in detail yet, but
a cursory check did not show any significant differences.

> Regards
>
> Alex

Thanks for your suggestions.


MV


--
Do not send e-mail to the above address. I do not read e-mail sent there.