On Tue, Jun 11, 2019 at 5:21 AM Xiao Ni <xni@xxxxxxxxxx> wrote:
>
>
>
> On 06/11/2019 10:53 AM, Colt Boyd wrote:
> > Is there anything that can be done?
> >
> > -----Original Message-----
> > From: Colt Boyd <coltboyd@xxxxxxxxx>
> > Sent: Saturday, June 8, 2019 10:48 AM
> > To: linux-raid@xxxxxxxxxxxxxxx
> > Subject: RAID-6 aborted reshape
> >
> > I was resizing a raid6 array with a internal write intent bitmap from
> > 5 3TB drives (in RAID6) to 6 drives. It was aborted very early in reshape via reboot and then reassembled with:
> > 'mdadm -A /dev/md0 --force --verbose --update=revert-reshape --invalid-backup /dev/sdb1 /dev/sdc1 /dev/sdd1 /dev/sde1 /dev/sdf1 /dev/sdg1'
>
> Does this command finish? I tried this myself with this command. The
> filesystem is good after this command.
> I interrupt the reshape by `mdadm -S /dev/md0` before assembling it.
It did finish, but did not leave a mountable XFS file system.
>
> Then I did another test. I interrupted the reshape by `echo b >
> /proc/sysrq-trigger`. Then tried to assemble the raid
> by your command. It gave the error message:
>
> [root@dell-per720-08 home]# mdadm -A /dev/md0 --force --verbose
> --update=revert-reshape --invalid-backup /dev/loop[0-4]
> mdadm: looking for devices for /dev/md0
> mdadm: Reshape position is not suitably aligned.
> mdadm: Try normal assembly and stop again
>
>
> Then I used this command to try to assemble it:
> mdadm -A /dev/md0 --verbose --invalid-backup /dev/loop[0-4]
>
> The filesystem is good too.
>
> By the way I used the latest upstream version.
I was using:
root@OMV1:/tmp# mdadm --version
mdadm - v3.4 - 28th January 2016
root@OMV1:/tmp# uname -r
4.19.0-0.bpo.2-amd64
I can try the upstream version.
I was not using overlay devices at first and initially tried the above
create on the actual raid devices. I still have the superblock intact
on the 6th device. Is there a way to reconstruct the super blocks on
devices 1-5 (raid devices 0-4) based on the superblock on the 6th
device and/or the backup file?
Thanks,
Colt
>
> Regards
> Xiao
> >
> > When I reassembled it this way I incorrectly thought the backup file was zero bytes. It wasn't. I still have the intact backup file.
> >
> > I’ve also since tried to reassemble it with the following create but the XFS fs is not accessible:
> > 'mdadm --create /dev/md0 --data-offset=1024 --level=6 --raid-devices=5 --chunk=1024K --name=OMV:0 /dev/sdb1 /dev/sde1 /dev/sdc1 /dev/sdd1
> > /dev/sdf1 --assume-clean --readonly'
> >
> > I can see the XFS FS on the drives, example:
> > root@OMV1:~# dd if=/dev/sde1 bs=512k count=5 | hexdump -C <snip>
> > 00200000 58 46 53 42 00 00 10 00 00 00 00 00 82 f2 c3 00 |XFSB............|
> > 00200010 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 |................|
> > 00200020 4e 2b 04 64 e8 1b 49 d9 a5 20 b5 74 79 94 52 f8 |N+.d..I.. .ty.R.| <snip>
> >
> > This is what it looked like immediately following the aborted reshape and before attempting to recreate it. This is from the drive that was being added at the time.
> >
> > /dev/sdg1:
> > Magic : a92b4efc
> > Version : 1.2
> > Feature Map : 0x1
> > Array UUID : f8fdf8d4:d173da32:eaa97186:eaf88ded
> > Name : OMV:0
> > Creation Time : Mon Feb 24 18:19:36 2014
> > Raid Level : raid6
> > Raid Devices : 6
> >
> > Avail Dev Size : 5858529280 (2793.56 GiB 2999.57 GB)
> > Array Size : 11717054464 (11174.25 GiB 11998.26 GB)
> > Used Dev Size : 5858527232 (2793.56 GiB 2999.57 GB)
> > Data Offset : 2048 sectors
> > Super Offset : 8 sectors
> > Unused Space : before=1960 sectors, after=2048 sectors
> > State : clean
> > Device UUID : 92e022c9:ee6fbc26:74da4bcc:5d0e0409
> >
> > Internal Bitmap : 8 sectors from superblock
> > Update Time : Thu Jun 6 10:24:34 2019
> > Bad Block Log : 512 entries available at offset 72 sectors
> > Checksum : 8f0d9eb9 - correct
> > Events : 1010399
> >
> > Layout : left-symmetric
> > Chunk Size : 1024K
> >
> > Device Role : Active device 5
> > Array State : AAAAAA ('A' == active, '.' == missing, 'R' == replacing)
> >
> > Where can I go from here to get this back?
> >
>
