[3041] in linux-net channel archive
Linux TCP/IP wedges talking to DG AViiON
daemon@ATHENA.MIT.EDU (Kris Karas)
Wed May 29 00:46:50 1996
Date: Tue, 28 May 1996 15:13:41 -0400
From: Kris Karas <ktk@ktk.bih.harvard.edu>
To: alan@lxorguk.ukuu.org.uk
cc: linux-net@vger.rutgers.edu
Dear Alan, Linux-Net, et al.
We've got a server (pentium/linux 1.99.7) that cannot communicate via
TCP/IP with a mainframe Data General AViiON. Just upgraded from
1.2.13 a.out to an ELF based system. Wondering if you know what's
going on, and could suggest a patch/fix. (I don't want Linux to get a
bad name here at our hospital, now that people are starting to use it
in the clinical environment.)
I've included TCPDUMP output below that illustrates the problem.
Experiment showed that PCTCP compatibility mode fixed the problem
under 1.2.13, but does not do so under 1.3.x. In short:
AViiON advertises window size smaller than mss.
Linux sends a small packet equal in size to the window.
AViiON acks it, but with a larger window now >= mss.
Linux sends a new, larger packet, but fails to start the packet at the
sequence number most recently acked; it chooses the previous sequence
number instead, causing the AViiON to ignore the packet and re-ack.
This is now an endless loop that continues until one or the other
gives up. My understanding of the RFCs (and comer, _Internetworking
with TCP/IP_) is that the sender is supposed to start its packet at
the sequence most recently acked; perhaps this is wrong?
Without further ado...
14:16:40.703796 AViiON.smtp > Linux.1105: . ack 7788 win 511
14:16:40.903796 Linux.1105 > AViiON.smtp: P 7788:8299(511) ack 210 win 29696
14:16:40.913796 AViiON.smtp > Linux.1105: . ack 8299 win 963
14:16:40.913796 Linux.1105 > AViiON.smtp: P 7788:8556(768) ack 210 win 29696
14:16:40.913796 AViiON.smtp > Linux.1105: . ack 8299 win 963
14:16:41.103796 Linux.1105 > AViiON.smtp: P 7788:8556(768) ack 210 win 29696
14:16:41.113796 AViiON.smtp > Linux.1105: . ack 8299 win 963
..... ad infinitum.
Any suggestions/help would be gratefully appreciated.
Kris Karas, ktk@bih.harvard.edu
