Dell 2950 w/ PERC 5i/Integrated, seeing RAID'd HDD's

Bug #55138 reported by Glenn Crawford
14
Affects Status Importance Assigned to Milestone
linux-source-2.6.15 (Ubuntu)
Fix Released
High
Kyle McMartin

Bug Description

Binary package hint: linux-image-server

I have a Dell 2950 with a PERC 5i/Integrated RAID controller.

I have created a logical disk a 300Gb comprised of 2 x 300Gb HDD's attached to the RAID controller.

When I install Ubuntu Server (32-bit) I can only see 1 HDD - /dev/sda

After installation when I upgrade the kernel to the latest version I can see 3 HDD's! /dev/sda, /dev/sdb, and /dev/sdc. sda and sdb are the HDD's, and sdc is now the RAID'd logical volume!

Current installed & working kernel:
rccadmin@mizar:~$ uname -a
Linux mizar 2.6.15-23-server #1 SMP Tue May 23 15:10:35 UTC 2006 i686 GNU/Linux

kernel version attempting to upgrade to:
urccadmin@mizar:~$ uname -a
Linux mizar 2.6.15-26-server #1 SMP Tue May 23 15:10:35 UTC 2006 i686 GNU/Linux

Additional info: In order to even see the HDD's after installation I need to add the "megaraid_sas" driver to /etc/mkinitramfs/modules and recompile the initrd file.

description: updated
Revision history for this message
Ben Collins (ben-collins) wrote :

I'm a little confused as to what the bug is. When you upgrade, you see all the drives, so what's the problem?

Revision history for this message
Ben Collins (ben-collins) wrote :

Ah, I think that you are saying that you shouldn't see the individual drives. I don't think that's a bug. If the hardware lets us see these drives, then we probe them. It's up to the hardware to hide or show these physical devices.

Changed in linux-source-2.6.15:
status: Unconfirmed → Rejected
Revision history for this message
Glenn Crawford (glenn-crawford) wrote :

But I *shouldn't* see the hard drives - if I see the hard drives during install (eg. 64-bit ubuntu) then the installer gets confused (actually grub gives me Error 21's and I can't boot)...

and if I don't see the hard drives during the install (eg. 32-bit ubuntu) then when I upgrade and I _can_ see the hard drives, everything goes up in smoke because whereas I installed on /dev/sda being the raid logical unit, /dev/sda is now the 1st physical hard drive. If I try to change the mountpoints to point to /dev/sdc (which is now the raid logical unit) then I end up getting all sorts of problems with file system corruption - not to mention grub starts having issues again

Revision history for this message
Ben Collins (ben-collins) wrote :

I'm not sure why there's a descrpancy between the versions, but as far as I understand this, if the raid is making them available to the controller, then the kernel can see them. It doesn't know which are physical and which is the raid device. It just probes and reads what the controller gives it.

I've seen this exact thing with a Sun Netra A1000. It had 8 drives, that were raid5'd. It showed all 8 drives, plus the raid5 drive. It worked just fine. We got errors trying to access the physical drives, but the raid5 drive was no problem.

Revision history for this message
Liam Bedford (lbedford-deactivatedaccount) wrote :

The reason it's a problem is because it's not consistent with the bios. It's also not consistent with itself. I have a 2950 here with 6 drives, all of which are in a RAID 10 volume.

So far, the dapper installer has presented me with:
1 drive (RAID volume only)
3 drives (2 physical drives + 1 RAID volume)
7 drives (6 physical drives + 1 RAID volume)

The BIOS sees the RAID volume as (hd0), but the driver is seeing it
as (hd6) and putting a broken grub config on it. The one time it booted
and saw just 1 drive, it worked properly.

This is fixed in other kernels. SUSE don't have the problem, and it seemed
to be okay in edgy, even though the installer was broken.

Revision history for this message
Ben Collins (ben-collins) wrote :

I guess the inconsistency can be seen as a problem. I'll check into it further.

Changed in linux-source-2.6.15:
status: Rejected → Confirmed
Revision history for this message
Liam Bedford (lbedford-deactivatedaccount) wrote :

any progress? this bug is causing installs on 2950s to be
broken, and dell have EOL'd the 2850.

L.

Revision history for this message
Joe eStara (joe-estara) wrote :

This is far worse than simply not seeing the devices: When all the devices are seen, writes are only sent to one hard drive effectively making your raid worthless.

This is fixed in FC5 by using the latest megaraid drivers 2.0.4 (?)

This is keeping me using ubuntu on 35 severs, and will keep every buyer of dell 1950's and 2950's from using ubuntu so it's very harmful and critical.

Joe

Revision history for this message
Ben Collins (ben-collins) wrote :

This should be fixed in latest Dapper security updates.

Can you make sure you have the latest from dapper-security and retest?

Changed in linux-source-2.6.15:
assignee: nobody → kyle
importance: Undecided → High
status: Confirmed → Needs Info
Revision history for this message
Fabián Rodríguez (magicfab) wrote :

Was this a duplicate of # 57265 (or the other way around) ?

Revision history for this message
Fabián Rodríguez (magicfab) wrote :

I meant to have that reference linked. Here it goes: Bug # 57265

Revision history for this message
Scott Hilleard (scottles-deactivatedaccount) wrote :

I can confirm that on my Dell 2950 servers this bug is fixed with the new 2.6.15-28 kernel release. An updated installer disc would be nice though as at the moment you have do a work around (http://ubuntuforums.org/showthread.php?t=226114) to get ubuntu installed.

Revision history for this message
Rimas Kudelis (rq) wrote :

hm...
now that I've upgraded from 2.6.15-27 to 2.6.15-28, my system doesnt' boot with the new kernel. I guess it's because I have /dev/sdc all around in fstab. Will changing of sdc to sda in it fix the boot problem for me?

Revision history for this message
Scott Hilleard (scottles-deactivatedaccount) wrote :

yes, after applying the kernel update you will need to update your fstab accordingly to get the system booting again!

Revision history for this message
Rimas Kudelis (rq) wrote :

Thank you Scott,
The system now works as expected.

The devs should take a look at Bug #56854, and maybe fix that too, at least for the next version of *buntu...

Revision history for this message
Ondřej Surý (ondrej) wrote :

Hmm, strange. After upgrading to 2.6.15-28 (amd64) I cannot boot my system.

There are times when system boots with many write errors and sometimes it just prints:

run-init: /sbin/init: error 13
Kernel panic - not syncing: Ateempted to kill init!

But I can only guess, that it's connected to megaraid_sas.

Revision history for this message
Rimas Kudelis (rq) wrote :

Ondřej:
you must edit /etc/fstab and /boot/grub/menu.lst after upgrading. Change sdc to sda everywhere, then the system will boot. To do that, you should boot to your old kernel first. That's what I did, and the system works smoothly now.

Revision history for this message
Ondřej Surý (ondrej) wrote :

Ignore my last comment.

Kernel is broken on my Dells, but it's not connected to megaraid_sas (or rather scsi subsystem where this error was).

During boot I can catch on remote console that only sda is in my system and that's just fine.

Ondrej.

Revision history for this message
Ondřej Surý (ondrej) wrote :

2 Rimas: Nope, I was already using patches scsi-mod module (look at reporter of #57265). Must be some other error. Anyway I am going to investigate it now, I'll report back.

Revision history for this message
Martin Pitt (pitti) wrote :

So this is fixed in the actual kernel, it just needs to be rolled out to the install CDs.

Changed in linux-source-2.6.15:
status: Incomplete → Fix Released
Revision history for this message
Scott Hilleard (scottles-deactivatedaccount) wrote :

that is correct. It is fixed in the kernel, and now needs rolling out to the install CD's to make installing Ubuntu 6.06 LTS on a new Dell 2950 not a complete pain :)

Revision history for this message
Alan Tam (at) wrote :

I have to reject to the "Fix Release" status of this bug. Not until you roll out a 6.06.2 CD including the fix, basically there is no way for us to install the system smoothly. Even if we can install (by manually tweaking around), upgrading the kernel will break everything, since /dev/sd* will be renamed between the two kernel versions.

Revision history for this message
Fabián Rodríguez (magicfab) wrote :

Existing CD (ISO) images can't and won't be fixed. I only see 2 outstanding bug for a 6.06.2 release and new ISO images to be generated, see here:
https://edge.launchpad.net/ubuntu/+milestone/ubuntu-6.06.2

Having a 6.06.2 ISO image is a separate issue from this bug which is actually fixed according to the notes I see.

Revision history for this message
Alan Tam (at) wrote :

Thanks for the link showing the 6.06.2 blockers. I know existing CDs cannot be modified in any way.

My major complain is:
1. In June 2006, you released Ubuntu 6.06 ISO, which works.
2. In August 2006, you released Ubuntu 6.06.1 ISO, which doesn't work.
3. In November 2007, there is still no updated ISO image!

Do you think this is acceptable for a user-friendly OS?

Revision history for this message
Fabián Rodríguez (magicfab) wrote :

Alan, if you know then don't demand such an updated ISO. A bug report is not the place for this kind of discussion, nor is your comment any helpful towards resolution of this. Every comment here generates email to a number of people that need to read it, review it and decide what to do.

Contact me privately when you post this to the forums or to the ubuntu-devel-discuss mailing list and I'll gladly participate in the discussion, as may many other people. I invite anyone interested in such discussion to contact me directly and I'll let them know when/if this is posted to a forum or mailing list.

In case it's not clear enough, a bug report is NOT the place to discuss this. Thank you!

To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.