Date: Tue, 17 Jun 1997 09:40:59 -0700 From: John Polstra <jdp@polstra.com> To: fenner@parc.xerox.com Cc: current@freebsd.org Subject: Re: loopback i/f bug (?) [ was Re: Some trouble with SCSI HD] Message-ID: <199706171641.JAA02737@austin.polstra.com> In-Reply-To: <97Jun16.173857pdt.177512@crevenia.parc.xerox.com> References: <97Jun16.173857pdt.177512@crevenia.parc.xerox.com>
next in thread | previous in thread | raw e-mail | index | archive | help
In article <97Jun16.173857pdt.177512@crevenia.parc.xerox.com>, Bill Fenner <fenner@parc.xerox.com> wrote: > I use the "ssh -L 5999:localhost:5999 -R 9898:localhost:9898" trick > along with "cvsup -h localhost -P 9898" and haven't experienced any > problems with a -current vintage 5/20/97. Yes, it doesn't happen when you tunnel through ssh. It turns out that you have to set up quite an uncommon state of affairs in order to make it fail. I spent a couple days on it with tcpdump and the kernel debugger, and was able to track it down to a bug in the TCP stack. The bug is exposed by (but not caused by) the large MTU and buffer sizes chosen when localhost is used. (Compare "route get localhost" with "route get yourhost.domain.com".) I have a small test program that reliably makes it fail even when localhost isn't used. I also have a simple fix which is being reviewed. It's not likely to be related to the NFS hangs, since this bug is in the TCP stack and not in the UDP code. -- John Polstra jdp@polstra.com John D. Polstra & Co., Inc. Seattle, Washington USA "Self-knowledge is always bad news." -- John Barth
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199706171641.JAA02737>