Date: Fri, 15 Mar 2013 22:13:39 -0400 (EDT) From: Rick Macklem <rmacklem@uoguelph.ca> To: Andre Oppermann <andre@freebsd.org> Cc: freebsd-current <freebsd-current@freebsd.org>, Lars Eggert <lars@netapp.com> Subject: Re: NewNFS vs. oldNFS for 10.0? Message-ID: <1138560306.3963522.1363400019941.JavaMail.root@erie.cs.uoguelph.ca> In-Reply-To: <51432F0D.2040002@freebsd.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Andre Oppermann wrote: > On 15.03.2013 15:01, Eggert, Lars wrote: > > Hi, > > > > this reminds me that I ran into an issue lately with the new NFS and > > locking for NFSv3 mounts on a client that ran -CURRENT and a server > > that ran > > -STABLE. > > > > When I ran "portmaster -a" on the client, which mounted /usr/ports > > and /usr/local, as well as the location of the respective sqlite > > databases over > > NFSv3, the client network stack became unresponsive on all > > interfaces for 30 or so seconds and e.g. SSH connections broke. The > > serial console > > remained active throughout, and the system didn't crash. About a > > minute after the wedgie I could SSH into the box again, too. > > The total wedge of all interfaces certainly shouldn't happen. This > smells like blocking on a lock on a socket_upcall() thereby wedging > tcp_input. I don't know the lockd code so maybe Rick knows how this > could happen. > I'm afraid I can't help much. The lockd and all the NFS stuff (old and new) share the kernel rpc (sys/rpc) code and that is where the upcalls are. At a quick look at them, they do have to acquire a mutex, but it would be hard to believe that would take 30sec? Also, I think (I'm not familiar with the NLM, so I'm not sure) that the NLM normally uses UDP and not TCP. It is a separate protocol with different port#s etc than NFS. rick > -- > Andre > > > The issue went away when I killed lockd on the client, but that > > caused the sqlite database to become corrupted over time. The > > workaround for me > > was to move to NFSv4, which has been working fine. (One more reason > > to make it the default...) > > > > I'm not really sure how to debug this further, but would be willing > > to work with someone off-list who'd tell me what tests to run. > > > > Lars _______________________________________________ > > freebsd-current@freebsd.org mailing list > > http://lists.freebsd.org/mailman/listinfo/freebsd-current To > > unsubscribe, send any mail to > > "freebsd-current-unsubscribe@freebsd.org" > > > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1138560306.3963522.1363400019941.JavaMail.root>