From mboxrd@z Thu Jan 1 00:00:00 1970 From: Phil Turmel Subject: Read-only mounts (was mdadm RAID6 "active" with spares and failed disks; need help) Date: Sat, 28 Mar 2015 11:11:29 -0400 Message-ID: <5516C4A1.9030604@turmel.org> References: <54ABEE54.6020707@sympatico.ca> <54B305B2.6000702@lentijn.sess.ink> <54B3F7ED.1000809@youngman.org.uk> <54C0C73A.4080409@lentijn.sess.ink> <55160B0A.1090405@turmel.org> <20150328151142.11db8c9a@natsu> Mime-Version: 1.0 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20150328151142.11db8c9a@natsu> Sender: linux-raid-owner@vger.kernel.org To: Roman Mamedov Cc: Matt Callaghan , linux-raid@vger.kernel.org List-Id: linux-raid.ids On 03/28/2015 06:11 AM, Roman Mamedov wrote: > On Fri, 27 Mar 2015 21:59:38 -0400 > Phil Turmel wrote: > >> I read through this. Given all of the destructive actions you took, I >> am doubtful you will ever get your data. Like mounting "readonly". >> That gives you a readonly filesystem, but it writes to the device. >> Possibly a great deal if there's a journal to replay. > > Are you sure, which FS does that? I remember some discussion on FS lists > (Btrfs?), and IIRC the consensus and the implemented behavior was that the > device shouldn't ever be touched with writes on RO mounts, no matter what. I remember people being burned by it with ext3/4 a couple years ago. Which is why all of the array disaster recoveries I've helped with called for fsck -n to verify a reconstruction attempt, not a mount -o ro. I will set up a small VM and see what recent kernels do. Phil