Arch Linux

dobradude

Member

Registered: 2019-10-15

Posts: 7

Kernels > 5.3.1 and RAID 0 (mdadm) not working

Hi folks

I can't see this mentioned elsewhere but  I can't seem to get my RAID 0 array working after upgrade to kernel 5.3.5/ 6

the HDD's are recognized but RAID won't start and says devices busy or something -- note this is a data array not a boot device.

I've also installed arch on 2 systems -- one with AMD asnd operon, and another on an Intel machine -- same problem so it's not specifically machine related.

I installed arch on 5.3.1 (coming from manjaro so understand a bit about arch linux philosophy.

BTW if it's allowed to mention other disros here -- tried ubuntu and manjaro -- same problem with kernels > 5.3.1 so AFAIK it seems to be a generic kernel problem

Any ideas folks

currently I've backlevelled linux and linux-headers to 5.3.1 and blocked those updating

pacman -U <package>

I had also to backlevel gcc and gcc-libs (that was to keep VMWare workstation 15.5 working - although there's no basic problem with the newer kernel --it's just RAID tyhat seems to be the problem.

Thanks people --hope the knowlegable people on this forum can come up with things to get RAID 0 working on these kernels.

loqs

Member

Registered: 2014-03-06

Posts: 17,321

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

Please post the command that fails along with its full output in code tags.
What is the first version between 5.31 and 5.3.5 that fails?
If you do not have all the versions in the system's package cache you can obtain them from the ALA

Last edited by loqs (2019-10-15 16:15:37)

stronnag

Member

Registered: 2011-01-25

Posts: 61

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

See https://www.reddit.com/r/linuxquestions … lt_layout/ for a description and workaround (I used this to recover an old raid0 installation).

Then convert your raid0 setup into an LVM.

dobradude

Member

Registered: 2019-10-15

Posts: 7

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

Hi there
on 5.3.1 lsblk shows the following :

NAME   MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
sda      8:0    0 465.8G  0 disk 
├─sda2   8:2    0   315M  0 part  /boot
├─sda3   8:3    0  31.7G  0 part  /
├─sda4   8:4    0   418G  0 part  /home
└─sda5   8:5    0  15.8G  0 part  [SWAP]
sdb      8:16   0   5.5T  0 disk 
└─md0    9:0    0    20T  0 raid0 /mnt/DV1
sdc      8:32   0   5.5T  0 disk 
└─md0    9:0    0    20T  0 raid0 /mnt/DV1
sdd      8:48   0   4.6T  0 disk 
└─md0    9:0    0    20T  0 raid0 /mnt/DV1
sde      8:64   0   4.6T  0 disk 
└─md0    9:0    0    20T  0 raid0 /mnt/DV1

on kernels > 5.3.3 -- haven't got a system currently available but the devices are simply shown as  discrete disks /dev/sdb, /dev/sdc, /dev/sdd and /dev/sde and no mdadm commands will work to get the array working

I certainly don't want to mess around with LVM etc -- there's 20TB of storage space on this array that I don't want to lose (although I do have backup - it would take ages to re-build this.

In any case it's all worked 100% perfectly until now -- I also run VMWare on this machine with  3 Windows VM's which I don't want to lose either) .

Seems strange when a minor kernel update breaks this. Mdadm has been working perfectly even as far back as kernels 3.x

Just checked also on a colleagues machine --he's got the same problem (although he uses RAID 10) on kernel 5.3.6  and also on ubuntu it's broken as well.

frostschutz

Member

Registered: 2013-11-15

Posts: 1,417

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

Seems strange when a minor kernel update breaks this.

It's a bug, recently discovered, that caused the raid0 layout to change unintentionally a while back, so now they're in a pickle... they have two layouts and the metadata doesn't say which is correct.

If they fix the unintentional change, they corrupt all raid0 using the unintended new layout; if they don't fix it, they corrupt all raid0 using the old layout. Both choices are wrong so they decided instead to make it "stop working entirely" until the sysadmin decides how they want to corrupt their raid0.

so you have to add a kernel paramter: raid0.default_layout=1 for new or =2 for old 3.x kernel raid0s

although he uses RAID 10

as far as I'm aware this issue is specific to raid0 with mixed drive sizes (like yours: 5.5T and 4.6T in one raid0)

and there is a very specific message in dmesg so maybe check dmesg what is going on exactly

Last edited by frostschutz (2019-10-16 08:57:01)

dobradude

Member

Registered: 2019-10-15

Posts: 7

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

Hi there

Thanks

I'll try that

another suggestion was to switch to LVM - but I chose RAID 0 because of superior I/O speed -- OK there's a rsik but I do have backups so if necessary I can re-build.

here's a decent review over performance  of RAID 0 vs LVM which could be of interest
http://fibrevillage.com/storage/429-per … ed-mapping

anyway I'll try your suggestion and post back. If it doesn't work I've got the old kernel which I can reload

EjDriver

Member

Registered: 2016-11-27

Posts: 7

Re: Kernels > 5.3.1 and RAID 0 (mdadm) not working

I ran into this issue last week. Took 2hrs to figure out which package cause the issue. I had to downgrade linux,linux-headers, and gcc. I think this should be under "Latest News".