[svlug] Re: TCP checksum ignored?

Ira Abramov lists-svlug at ira.abramov.org
Sun Mar 23 11:56:10 PST 2003


Quoting J C Lawrence, from the post of Sun, 23 Mar:
> > Scenario: files pulled over TCP (http, ftp) to the machine (mobile
> > Pentium 3, Intel e1000 interface) get there with a different md5sum
> 
> Have you checked in the reverse direction -- files transfered FROM the
> suspect box?  I'd tend to be more suspicious of RAM or DASD errors.

forgot to mention:

eth1 on the same machine (on-board e1000 as well, possibly even on the
same chip) works OK while eth0 displays those problems. the corrupted
memory sounds odd to me when three different kernels have the same
problems (2.4.9-rh, 2.4.18-suse and 2.4.20-vanilla will probably use
different physical memory block for this function). all evidence show
it's specific to eth0. so what can corrupt the data but keep it looking
correct on the way from the eth0 interface to the kernel?

Tomorrow I'll have access to that machine personally, so I'll do some
transfers in the other direction as well.

-- 
The sig master
Ira Abramov
http://ira.abramov.org/email/



More information about the svlug mailing list