From mboxrd@z Thu Jan 1 00:00:00 1970 From: Shaohua Li Subject: Re: [md PATCH 0/2] Make it possible to disable create_on_open semantics. Date: Wed, 12 Apr 2017 12:24:07 -0700 Message-ID: <20170412192407.3hxadmmtfgpsn2bc@kernel.org> References: <149197804398.19936.12809382889200123725.stgit@noble> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <149197804398.19936.12809382889200123725.stgit@noble> Sender: linux-raid-owner@vger.kernel.org To: NeilBrown Cc: linux-raid@vger.kernel.org, Coly Li List-Id: linux-raid.ids On Wed, Apr 12, 2017 at 04:26:12PM +1000, Neil Brown wrote: > Currently, opening an md /dev node will create the array object. > This makes it hard to destroy the object as udev will typically > re-open the device node when handling REMOVE events. > > The "new_array" module parameter was created to work towards avoiding > this problem, and it can be used when > CREATE names=yes > > is given in /etc/mdadm.conf. > How this doesn't currently support names like "md%d", which lots of > people use and expect, so we need more work before we can transition > away from create_on_open. > > These patches add support to "new_array" so that md%d devices > can be created. This will make it, once again, possible to have > md%d devices with numbers > 511. (3.17 make this impossible). > > An enhancement to mdadm that uses this will cause new_array to always > be used (where available), and we can then disable create_on_open > completely (after suitable transition periods). Thanks, applied! The md device creation interface especially create_on_open is a disaster, hopefully the future sysfs/configfs interface deprecates all of these. Thanks, Shaohua