Date: Fri, 14 Oct 2005 14:56:43 +0200 From: Nicolas KOWALSKI <Nicolas.Kowalski@imag.fr> To: freebsd-fs@FreeBSD.org Subject: Re: FreeBSD NFS server not responding to TCP SYN packets from Linux/SunOS clients Message-ID: <vqou0fkw92s.fsf@obiou.imag.fr> In-Reply-To: <20051014045824.V5343@odysseus.silby.com> References: <Pine.LNX.4.64.0510141021290.22064@corbeau.imag.fr> <20051014160128.hev160v52ossokg0@wwws.cs.ait.ac.th> <20051014045824.V5343@odysseus.silby.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Mike Silbersack <silby@silby.com> writes: > On Fri, 14 Oct 2005, on@cs.ait.ac.th wrote: > >> Nicolas KOWALSKI wrote: >>> Our FreeBSD 4.10 NFS server has some problems serving files by NFS >>> on TCP (no problem with UDP) when the Linux (2.6) or Solaris (5.9) >>> clients shut down in an unclean manner (power failure). When the >>> clients try to mount the shares from the server after an unclean >>> shutdown, the mount process hang during several minutes (delay is >>> varying), then succeeds. >> >> That is just a wild guess, but NFS mounting would happen always at >> the same stage of the boot, so maybe with the same source port >> number and you could be facing the problem that the connection is >> waiting for termination on the server (close_wait or fin_wait or >> something)... Se source port in working example is 798 and source >> port in failing example is 799 certainly not random. > > The socket on the server would still be in the ESTABLISHED state, > which is even worse than the close_wait or fin_wait states in this > case. The SYN will be accepted if it's greater than the previous > sequence number, so that's a 50% chance it'll work. Thanks for this explanation. > Assuming that port reuse is the problem, there is no quick fix for > this, just resetting connections when a SYN comes in would be a > really big security problem. Really? Are Linux and Solaris that insecure because of this behaviour? > Actually, there may be a quick fix for this specific machine. If you > set net.inet.tcp.keepidle to 1 minute (60*whatever kern.hz is), > that'll cause keepalive packets to be sent every minute to an idle > connection, rather than every 2 hours. That would kill the stuck > connections much quicker. Unfortunately, this does not work as expected. I just tested with my workstation (Linux 2.6), with NFS filesystems mounted with TCP; when the station rebooted abruptely, mounting the same NFS filesystems hung more than 1 minute (15 minutes just now). During this hang, I saw on the server, using netstat, the nfsd process related to my workstation in ESTABLISHED state. Any other tip? Many Thanks in advance, -- Nicolas
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?vqou0fkw92s.fsf>