From mboxrd@z Thu Jan 1 00:00:00 1970 From: NeilBrown Subject: Re: [PATCH] FIX: Cleanup after raid45->raid0 takeover Date: Thu, 7 Apr 2011 11:18:38 +1000 Message-ID: <20110407111838.2480b14c@notabene.brown> References: <20110328133156.10923.98647.stgit@gklab-128-111.igk.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20110328133156.10923.98647.stgit@gklab-128-111.igk.intel.com> Sender: linux-raid-owner@vger.kernel.org To: Krzysztof Wojcik Cc: linux-raid@vger.kernel.org, wojciech.neubauer@intel.com, adam.kwolek@intel.com, dan.j.williams@intel.com, ed.ciechanowski@intel.com List-Id: linux-raid.ids On Mon, 28 Mar 2011 15:31:57 +0200 Krzysztof Wojcik wrote: > Problem: > After raid4->raid0 takeover operation, another takeover operation > (e.g raid0->raid10) results "kernel oops". > Root cause: > Variables 'degraded' and 'plug' in mddev structure are not cleared > after raid45->raid0 takeover. > > This patch reset these variables. > > Signed-off-by: Krzysztof Wojcik > --- > drivers/md/raid0.c | 2 ++ > 1 files changed, 2 insertions(+), 0 deletions(-) > > diff --git a/drivers/md/raid0.c b/drivers/md/raid0.c > index c0ac457..5d47113 100644 > --- a/drivers/md/raid0.c > +++ b/drivers/md/raid0.c > @@ -603,6 +603,8 @@ static void *raid0_takeover_raid45(mddev_t *mddev) > mddev->new_chunk_sectors = mddev->chunk_sectors; > mddev->raid_disks--; > mddev->delta_disks = -1; > + mddev->degraded = 0; > + mddev->plug = NULL; > /* make sure it will be not marked as dirty */ > mddev->recovery_cp = MaxSector; > This is wrong. The raid0 'takeover' method is called while the raid5 array is still active, and mddev->plug and mddev->degraded are used by the active array. So you cannot just clear them like this. The new_* and delta_disks values are not used by an active array - also raid_disks isn't used - each personality keeps a copy of the effective raid_disks in it's private structure. So the ->plug must be cleared in the raid5 stop routine, and degraded should be cleared by level_store after calling ->stop on the raid5 and before calling ->run on the raid0. I have changed the patch to do that instead. Thanks, NeilBrown