SATA disk device naming ?

SATA disk device naming ?

[Robert M. Stockmann]

Hi,

After a rather tiresome nightly sit through, we discovered that when
going from kernel 2.6.3 to kernel 2.6.7 the SATA disk device naming
on at least the AMD64 platform changes from : /dev/hde and up 
to /dev/sda and up. What a total disaster. 

Effectively this means that your installed AMD64 platform based
linux distro cannot mount its root filesystem anymore.

We installed mandrake 10.0 amd64 and had to go back todo a initial
install on a UDMA IDE disk on /dev/hda.

Is there in such cases a smart workaround available, maybe as an extra
GRUB/LILO boot option ? And why was the device naming changed in 
such a fatal way, effectively going from IDE to SCSI device names.

Googling for "kernel init no init found" results in :

Results 11 - 20 of about 323,000 for kernel init no init found. 
ouch :(

Regards,

Robert
-- 
Robert M. Stockmann - RHCE
Network Engineer - UNIX/Linux Specialist
crashrecovery.org  stock@stokkie.net

Re: SATA disk device naming ?

[Jeff Garzik]

Robert M. Stockmann wrote:
> Hi,
> 
> After a rather tiresome nightly sit through, we discovered that when
> going from kernel 2.6.3 to kernel 2.6.7 the SATA disk device naming
> on at least the AMD64 platform changes from : /dev/hde and up 
> to /dev/sda and up. What a total disaster. 


Whoever builds your kernels changed around the kernel configuration on you.

SATA "disk naming" (what driver you use) did not change from 2.6.3 to 2.6.7.

	Jeff

Re: SATA disk device naming ?

[Eric D. Mudama]

On Tue, Jul 13 at  4:25, Robert M. Stockmann wrote:
>Is there in such cases a smart workaround available, maybe as an extra
>GRUB/LILO boot option ? And why was the device naming changed in 
>such a fatal way, effectively going from IDE to SCSI device names.

Google on the "root=LABEL=/" syntax supported by both LILO and grub (I
think), should point you in the proper direction.  Then it won't
matter where/how you mount your drives, they'll still boot and mount
into the proper places in the filesystem.

--eric


-- 
Eric D. Mudama
edmudama@mail.bounceswoosh.org

Re: SATA disk device naming ?

[Ian Soboroff]

"Eric D. Mudama" <edmudama@bounceswoosh.org> writes:

> On Tue, Jul 13 at  4:25, Robert M. Stockmann wrote:
>>Is there in such cases a smart workaround available, maybe as an extra
>> GRUB/LILO boot option ? And why was the device naming changed in
>> such a fatal way, effectively going from IDE to SCSI device names.
>
> Google on the "root=LABEL=/" syntax supported by both LILO and grub (I
> think), should point you in the proper direction.  Then it won't
> matter where/how you mount your drives, they'll still boot and mount
> into the proper places in the filesystem.

Did the patch for labeling the swap partition ever make it into the
kernel?

Also, if your filesystems are modular (e.g. XFS) then labels may not
help at boot time.

Ian

Re: SATA disk device naming ?

[Ricky Beam]

On Tue, 13 Jul 2004, Eric D. Mudama wrote:
>... "root=LABEL=/" ...

I've seen the LABEL method not work at all. (2.6.7-rc3 on the wikipedia
servers.)

--Ricky

Re: SATA disk device naming ?

[Jeff Garzik]

On Tue, Jul 13, 2004 at 12:23:32PM -0400, Ricky Beam wrote:
> On Tue, 13 Jul 2004, Eric D. Mudama wrote:
> >... "root=LABEL=/" ...
> 
> I've seen the LABEL method not work at all. (2.6.7-rc3 on the wikipedia
> servers.)

For LABEL to work, your filesystem needs to support it.

For LABEL to work on root filesystem, you need an initrd.

	Jeff

Re: SATA disk device naming ?

[Muli Ben-Yehuda]

[-- Attachment #1: Type: text/plain, Size: 543 bytes --]

On Tue, Jul 13, 2004 at 06:15:04PM +0100, Tigran Aivazian wrote:

> I assumed that "LABEL=/" thing is a RedHat-specific kernel's feature but 
> never bothered to check.

This is a common misconception. The LABEL magic is actually done by
nash, which is part of RH's (and others) mkinitrd package. This also
means that LABEL relies on having a working initrd. 

I have a short writeup of my understanding of how this works at
http://www.livejournal.com/users/mulix/84768.html. All AFAIUI,
corrections welcome. 

Cheers, 
Muli 


[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: SATA disk device naming ?

[Tigran Aivazian]

On Tue, 13 Jul 2004, Ricky Beam wrote:

> On Tue, 13 Jul 2004, Eric D. Mudama wrote:
> >... "root=LABEL=/" ...
> 
> I've seen the LABEL method not work at all. (2.6.7-rc3 on the wikipedia
> servers.)

right, the LABEL=/ never works for me so I always replace it with explicit 
"root=/dev/sda2" thing, but this is problematic if "/dev/sda2" becomes 
"/dev/hda2" if you boot with IDE driver compiled in, as opposed to SATA 
driver.

I assumed that "LABEL=/" thing is a RedHat-specific kernel's feature but 
never bothered to check. The reason I assumed so was that the problems 
always happen after installation, i.e. as soon as I replace the vendor's 
generic kernel with the one properly optimized for my specific needs.

Kind regards
Tigran

Re: SATA disk device naming ?

[Tigran Aivazian]

On Tue, 13 Jul 2004, Jeff Garzik wrote:
> For LABEL to work on root filesystem, you need an initrd.

Ok, now I know why it doesn't work for me --- I always disable initrd
after fresh installation. Thank you for the explanation.

Kind regards
Tigran

Re: SATA disk device naming ?

[Chris Wedgwood]

On Tue, Jul 13, 2004 at 12:49:11PM -0400, Jeff Garzik wrote:

> For LABEL to work on root filesystem, you need an initrd.

initrd is such a PITA at times, I wondered about something hacky like
sticking LABEL parsing for rootfs (marked init) into the kernel but
it's really gross.

Ideally the initrd/initramfs process just needs better (userspace)
infrastructure to make it more reliable/easier.



  --cw

Re: SATA disk device naming ?

[DaMouse]

This is why you need to keep a rescue disk of something like gentoo 
(normally up to date with driver imho) which you can edit fstab with 
instead of getting stuck...

-DaMouse

Re: SATA disk device naming ?

[P]

Chris Wedgwood wrote:
> On Tue, Jul 13, 2004 at 12:49:11PM -0400, Jeff Garzik wrote:
> 
> 
>>For LABEL to work on root filesystem, you need an initrd.
> 
> 
> initrd is such a PITA at times, I wondered about something hacky like
> sticking LABEL parsing for rootfs (marked init) into the kernel but
> it's really gross.
> 
> Ideally the initrd/initramfs process just needs better (userspace)
> infrastructure to make it more reliable/easier.

Something like this?
http://www-124.ibm.com/pipermail/evms/2001-March/000119.html

initrd is awkward but I've found I need the flexibility.
For e.g. I modified nash to support an index as well as
a label so one could have multiple redundant filesystem
images to choose from.

Pádraig.

Re: SATA disk device naming ?

[Arnd Bergmann]

[-- Attachment #1: Type: text/plain, Size: 553 bytes --]

On Dienstag, 13. Juli 2004 05:04, Jeff Garzik wrote:
> Whoever builds your kernels changed around the kernel configuration on you.
> 
> SATA "disk naming" (what driver you use) did not change from 2.6.3 to 2.6.7.
> 
The thing that changed is the new BLK_DEV_IDE_SATA option that defaults
to 'n', while before it was enabled implicitly.

I was bitten by this as well. Unfortunately I have an md RAID-0 volume
which I can not start after hde/hdg become sda/sdb, so I'll probably
have to keep using the old IDE driver indefinitely.

	Arnd <><

[-- Attachment #2: signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: SATA disk device naming ?

[Joel Becker]

On Tue, Jul 13, 2004 at 03:35:41PM -0700, Chris Wedgwood wrote:
> initrd is such a PITA at times, I wondered about something hacky like
> sticking LABEL parsing for rootfs (marked init) into the kernel but
> it's really gross.
> 
> Ideally the initrd/initramfs process just needs better (userspace)
> infrastructure to make it more reliable/easier.

	I'm waiting for udev to give me consistent device names easily.
Then I can specify root=/dev/disk1 and not have to scan-all-100-disks
for LABEL mounts.

Joel

-- 

	f/8 and be there.

Joel Becker
Senior Member of Technical Staff
Oracle Corporation
E-mail: joel.becker@oracle.com
Phone: (650) 506-8127

Re: SATA disk device naming ?

[Bartlomiej Zolnierkiewicz]

On Wednesday 14 of July 2004 13:20, Arnd Bergmann wrote:
> On Dienstag, 13. Juli 2004 05:04, Jeff Garzik wrote:
> > Whoever builds your kernels changed around the kernel configuration on
> > you.
> >
> > SATA "disk naming" (what driver you use) did not change from 2.6.3 to
> > 2.6.7.
>
> The thing that changed is the new BLK_DEV_IDE_SATA option that defaults
> to 'n', while before it was enabled implicitly.

This is a quite fresh change (post 2.6.7) and should be fixed before 2.6.8.
This option should default to 'y', we can put some runtime warning instead.

Jeff, do you agree?

> I was bitten by this as well. Unfortunately I have an md RAID-0 volume
> which I can not start after hde/hdg become sda/sdb, so I'll probably
> have to keep using the old IDE driver indefinitely.
>
> 	Arnd <><

Re: SATA disk device naming ?

[Jeff Garzik]

Bartlomiej Zolnierkiewicz wrote:
> On Wednesday 14 of July 2004 13:20, Arnd Bergmann wrote:
> 
>>On Dienstag, 13. Juli 2004 05:04, Jeff Garzik wrote:
>>
>>>Whoever builds your kernels changed around the kernel configuration on
>>>you.
>>>
>>>SATA "disk naming" (what driver you use) did not change from 2.6.3 to
>>>2.6.7.
>>
>>The thing that changed is the new BLK_DEV_IDE_SATA option that defaults
>>to 'n', while before it was enabled implicitly.
> 
> 
> This is a quite fresh change (post 2.6.7) and should be fixed before 2.6.8.
> This option should default to 'y', we can put some runtime warning instead.
> 
> Jeff, do you agree?


It's a really tough question...  default=n will eliminate a lot of 
problems people are reporting on lkml, but default=y changes the disks 
for the few users not having problems.

It's a tough call...

	Jeff

Re: SATA disk device naming ?

[Greg KH]

On Wed, Jul 14, 2004 at 07:44:14PM -0700, Joel Becker wrote:
> On Tue, Jul 13, 2004 at 03:35:41PM -0700, Chris Wedgwood wrote:
> > initrd is such a PITA at times, I wondered about something hacky like
> > sticking LABEL parsing for rootfs (marked init) into the kernel but
> > it's really gross.
> > 
> > Ideally the initrd/initramfs process just needs better (userspace)
> > infrastructure to make it more reliable/easier.
> 
> 	I'm waiting for udev to give me consistent device names easily.
> Then I can specify root=/dev/disk1 and not have to scan-all-100-disks
> for LABEL mounts.

What's missing from udev to let you do this today?

thanks,

greg k-h