From: Denis Vlasenko Subject: Re: Unable to kill processes in D-state Date: Mon, 7 Oct 2002 10:24:43 -0200 Sender: linux-kernel-owner@vger.kernel.org Message-ID: <200210070731.g977VAp17211@Port.imtp.ilyichevsk.odessa.ua> References: <1033841462.1247.3716.camel@phantasy> <20021006164228.GB17170@vagabond> <20021006170932.GA23134@stud.ntnu.no> Reply-To: vda@port.imtp.ilyichevsk.odessa.ua Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Cc: nfs@lists.sourceforge.net Return-path: To: linux-kernel@vger.kernel.org, Thomas Lang?s In-Reply-To: <20021006170932.GA23134@stud.ntnu.no> List-ID: On 6 October 2002 15:09, Thomas Lang?s wrote: > Jan Hudec: > > If the shares were successfuly reloaded, then the processes should > > wake up. If they don't, it's a bug in NFS. > > They never wake up, and it happens every time. > > > Try to reproduce it (ie. reboot some machine, let it start > > everything and then restart the autofsd and see if processes lock > > up) and then talk to NFS maintainers about that. > > As I said above, it happens every time we encounter this, ie. it's a > bug that easy to reproduce (since I added nfs@lists.sourceforge.net > to the CC-list, I'm going to write some of what's already said in > this thread). > > Problem: > Processes entering D-state is unkillable. We have a problem with this > everytime we restart autofs (which automounts quite a few NFS-shares > on campus), ie. on our samba-boxes smbd hangs forever after this > (in D-state). Samba still works, it's just that all the D-state > processes is unkillable and will remain that way untill we reboot the > computer. Every D-state process increases the load on the machine, > and one of our 2-CPU intel-boxes currently remains at 430 (which > extremly high for such a box). > > Solution: > ? :) Please provide info about: * automounter startup options * automounter configuration file(s) * client/server kernel version and .config * ksymoopsed SysRq-T dump of a couple of hung processes (no, not all processes! please select 2-5 hung processes _only_ before feeding dump to ksymoops :-) -- vda