lkml.org 
[lkml]   [1998]   [Apr]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectThe Checksum Problems revisited [WAS: Re: "Greased TCPenguin"]
Date
Letting the chips fall where they may, I quote Andrea Arcangeli:
>On Tue, 24 Mar 1998, Richard Gooch wrote:
>>TCPv4 bad checksum from 130.155.198.47:03ff to 130.155.191.26:03fd, len=32/32/52
>>TCPv4 bad checksum from 130.155.198.47:03ff to 130.155.191.26:03fd, len=32/32/52
>>TCPv4 bad checksum from 130.155.198.47:0448 to 130.155.191.26:1770, len=32/32/52
>>TCPv4 bad checksum from 130.155.198.47:0201 to 130.155.191.26:03ff, len=20/20/4
>
>I read messages such us this a lot of times but I never related it to a
>possible TCP bug, I always think that the ppp/slip is a poor connection
>and can miss something (we use checksum for that cases no?).
>
>I think that also because _I_ never seen that messages over an ethernet.

I've seen those when ever their is a speed difference between nodes such
as PPP/SLIP and 100/10Mb ethernet. I have looked at this at great length
and can only conclude that the kernel has screwed something up in a very
subtle way. I have complete packet traces from more than one location on
the network and even have the kernel dumping the entire packet. When I
look at the data, the traffic was _not_ corrupted... it's just out of
sequence. The skb is assembled wrong somehow which places the TCP data
segment together with the wrong headers.

It has been blamed on the Intel ethernet hardware, but that's pure bull as it
has reproduced with numerous other vendor's hardware. I stopped looking into
this some time ago as I didn't have the time to dig deeper and there was no
interest from anyone else (and it's hard for me to explain without pointing
at the screen :-)) With 2.1.77 SMP + IO-APIC (D), I've not seen the problem
as much as I used to.

If anyone want the stuff, they are welcome to it:
-rw-r--r-- 1 cramer root 26328054 Jan 20 07:02 Checksum.tar.gz

method crc date time compressed uncompr. ratio uncompressed_name
defla 367a684b Jan 20 07:02 26328054 116899840 77.4% Archive/Checksum.tar

--Ricky


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu

\
 
 \ /
  Last update: 2005-03-22 13:42    [W:0.072 / U:0.140 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site