From: "J. Bruce Fields" Subject: Re: Turning off delegations in NFSv4 server? Date: Fri, 5 Mar 2010 10:30:30 -0500 Message-ID: <20100305153030.GA14862@fieldses.org> References: <4786788D.6040707@garzik.org> <20080110200034.GI2132@fieldses.org> <20100304143317.GA17191@fieldses.org> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Cc: NFS list , nfsv4@linux-nfs.org, Matt Bernstein To: Matt Bernstein <+systems.extlists.nfsv4@dcs.qmul.ac.uk> Return-path: In-Reply-To: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: nfsv4-bounces@linux-nfs.org Errors-To: nfsv4-bounces@linux-nfs.org List-ID: On Fri, Mar 05, 2010 at 03:12:46PM +0000, Matt Bernstein wrote: > On Mar 4 J. Bruce Fields wrote: >> On Thu, Mar 04, 2010 at 09:23:26AM +0000, Matt Bernstein wrote: >>> On Jan 10 2008 J. Bruce Fields wrote: >>>> The easiest thing is probably just to turn off leases: >>>> echo 0 >/proc/sys/fs/leases-enable >>> >>> I found this from two years ago, and am wondering >>> - is this still the easiest thing to do? >>> - is this on server, client or both? >>> - will it break anything else? >>> >>> We have a CentOS 5.4 NFS3/NFS4/samba server with about 300 clients on its >>> subnet, and another 100-200 on another subnet, which under high load has >>> started kernel-panicking (sometimes in nfsd4_cb_recall). >> >> Have you filed a bug with the backtraces? > > Duly nudged. I've posted the most recent one to > . OK, thanks. Unfortunately that code has changed quite a bit since 2.6.18. I don't recall this specific bug, but I wouldn't be suprised if it's something we've since fixed. Looking through 'gitk v2.6.18.. fs/nfsd/nfs4callback.c fs/nfsd/nfs4state.c' for delegation/callback fixes might be one approach. --b. > >>> I want to disable delegations to see if that cures our symptoms--but I >>> worry that turning leases off might cause other problems. >> >> No, it shouldn't cause problems, at least for NFSv4 clients. (Samba may >> be more reliant on leases, especially if Samba and NFSv4 clients are >> acting on the same files at the same time--but I don't know.) > > Thanks; no-one's grumbled just yet. > > Matt