not all volume types export UUIDs (was: /dev/disk/by-id symlinks

not all volume types export UUIDs (was: /dev/disk/by-id symlinks

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 674 bytes --]

also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1000 +0200]:
> UUID's are the right solution here. Fix your mkswap, or re-run it, it
> has support for this for long:

Yes, you are right, this does work for many of my cases.
Unfortunately, there are still volume types out there without UUID
support, such as LVM (I think), mdadm, and non-LUKS dm-crypt. For
those, UUIDs will not be available.

-- 
 .''`.   martin f. krafft <madduck@debian.org>
: :'  :  proud Debian developer, author, administrator, and user
`. `'`   http://people.debian.org/~madduck - http://debiansystem.info
  `-  Debian - when you have better things to do than fixing systems

[-- Attachment #2: Digital signature (see http://martin-krafft.net/gpg/) --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Wed, 2008-07-02 at 12:28 +0200, martin f krafft wrote:
> also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1000 +0200]:
> > UUID's are the right solution here. Fix your mkswap, or re-run it, it
> > has support for this for long:
> 
> Yes, you are right, this does work for many of my cases.

> Unfortunately, there are still volume types out there without UUID
> support, such as LVM (I think), mdadm, and non-LUKS dm-crypt.

Then we should just add links for the raid-members. We already recognize
most of the metadata, just don't do anything with it.

> For
> those, UUIDs will not be available.

Oh, most of them have UUID's.

Anyway, the ID's are _bus_-id's, so the prefix is what we want. They are
in almost all cases unique on the same bus. Most devices which can move
from one bus to another will never share the same id's, like an
USB-storage device has no way to access ATA-data of the enclosed disk,
it will show the serial number of the usb-interface-device, which is
totally different.
ATA and SCSI are special, because of the Linux kernel using SCSI as a
transport for ATA, but we already create ata-* links for SCSI devices
which will show up if vendor=ATA, so this case is already covered.

What kind of id's do you have in mind, which would be the same across
buses?

Thanks,
Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 638 bytes --]

also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1619 +0200]:
> What kind of id's do you have in mind, which would be the same across
> buses?

I don't. I am trying to understand why the bus type is an integral
part of the /dev/disk/by-id/* links.

-- 
 .''`.   martin f. krafft <madduck@debian.org>
: :'  :  proud Debian developer, author, administrator, and user
`. `'`   http://people.debian.org/~madduck - http://debiansystem.info
  `-  Debian - when you have better things to do than fixing systems
 
dies ist eine manuell generierte email. sie beinhaltet
tippfehler und ist auch ohne großbuchstaben gültig.

[-- Attachment #2: Digital signature (see http://martin-krafft.net/gpg/) --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Greg KH]

On Wed, Jul 02, 2008 at 05:49:32PM +0200, martin f krafft wrote:
> also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1619 +0200]:
> > What kind of id's do you have in mind, which would be the same across
> > buses?
> 
> I don't. I am trying to understand why the bus type is an integral
> part of the /dev/disk/by-id/* links.

Because UUIDs are not guaranteed to be unique, except by bus type.  So
if you place them all in the same namespace, you need to add the bus
type to ensure there are no collisions.

thanks,

greg k-h

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Wed, 2008-07-02 at 17:49 +0200, martin f krafft wrote:
> also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1619 +0200]:
> > What kind of id's do you have in mind, which would be the same across
> > buses?
> 
> I don't. I am trying to understand why the bus type is an integral
> part of the /dev/disk/by-id/* links.

Because it's the USB-serial-number, the ATA-serial-string, the
SCSI-INQUIRY-serial-string, the FireWire-gid-number, the dm-name-string,
the md-name-string, the SD-flash-card serial, ..., and not a unique id
on its own. All are only unique per namespace/bus, and should better not
be mixed. 

The exception which would allow us moving the number around is ATA
devices using SCSI, because ATA synthesizes all SCSI data, but I think
we better just create two links in this case, instead of trying to move
all different id's into one namespace.

Thanks,
Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Karl O. Pinc]

On 07/02/2008 10:56:31 AM, Greg KH wrote:

> Because UUIDs are not guaranteed to be unique, except by bus type.

That seems a shame, given that they are
named Universally Unique IDentifiers and
are all supposed to reside in a single
namespace.  At least that's what I've
thought UUIDs are.  Oh well.


Karl <kop@meme.com>
Free Software:  "You don't pay back, you pay forward."
                  -- Robert A. Heinlein

Re: not all volume types export UUIDs (was: /dev/disk/by-id symlinks should not include bus) type

[Kay Sievers]

On Wed, Jul 2, 2008 at 18:17, Karl O. Pinc <kop@meme.com> wrote:
>
> On 07/02/2008 10:56:31 AM, Greg KH wrote:
>
>> Because UUIDs are not guaranteed to be unique, except by bus type.
>
> That seems a shame, given that they are
> named Universally Unique IDentifiers and
> are all supposed to reside in a single
> namespace.  At least that's what I've
> thought UUIDs are.  Oh well.

/by-id/* links are usually not using UUID's but serial numbers/names.

Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 863 bytes --]

also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> All are only unique per namespace/bus, and should better not be
> mixed.

Well, but that's my point I suppose. Why do we have to treat the
same drive differently, whether it's on the ATA bus or the USB bus?

My reason for bringing this up was that I was trying to boot off
a harddrive via USB and had to change all the links, rather than
just watching the system boot off the same disk it has always booted
from.

-- 
 .''`.   martin f. krafft <madduck@debian.org>
: :'  :  proud Debian developer, author, administrator, and user
`. `'`   http://people.debian.org/~madduck - http://debiansystem.info
  `-  Debian - when you have better things to do than fixing systems
 
the heineken uncertainty principle:
  you can never be sure how many
  beers you had last night.

[-- Attachment #2: Digital signature (see http://martin-krafft.net/gpg/) --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Thu, 2008-07-03 at 07:30 +0200, martin f krafft wrote:
> also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> > All are only unique per namespace/bus, and should better not be
> > mixed.
> 
> Well, but that's my point I suppose. Why do we have to treat the
> same drive differently, whether it's on the ATA bus or the USB bus?
> 
> My reason for bringing this up was that I was trying to boot off
> a harddrive via USB and had to change all the links, rather than
> just watching the system boot off the same disk it has always booted
> from.

So, what id's/symlinks is the same disk showing for ATA and USB in your
case?

Only advanced USB bridges read the number from the actual disk, I have a
bunch of enclosures here, which show all different id's for the same
disk inserted. Also usb-storage needs to encode the target and lun
number in the id, for multi-slot devices, so I do not think that will
work.

Thanks,
Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id symlinks should not include bus) type

[Matthew Dharm]

[-- Attachment #1: Type: text/plain, Size: 1407 bytes --]

On Thu, Jul 03, 2008 at 11:02:46AM +0200, Kay Sievers wrote:
> On Thu, 2008-07-03 at 07:30 +0200, martin f krafft wrote:
> > also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> > > All are only unique per namespace/bus, and should better not be
> > > mixed.
> > 
> > Well, but that's my point I suppose. Why do we have to treat the
> > same drive differently, whether it's on the ATA bus or the USB bus?
> > 
> > My reason for bringing this up was that I was trying to boot off
> > a harddrive via USB and had to change all the links, rather than
> > just watching the system boot off the same disk it has always booted
> > from.
> 
> So, what id's/symlinks is the same disk showing for ATA and USB in your
> case?
> 
> Only advanced USB bridges read the number from the actual disk, I have a
> bunch of enclosures here, which show all different id's for the same
> disk inserted. Also usb-storage needs to encode the target and lun
> number in the id, for multi-slot devices, so I do not think that will
> work.

Isn't devlabel the technology that is supposed to be used for this sort of
scenario?

Matt

-- 
Matthew Dharm                              Home: mdharm-usb@one-eyed-alien.net 
Maintainer, Linux USB Mass Storage Driver

C:  They kicked your ass, didn't they?
S:  They were cheating!
					-- The Chief and Stef
User Friendly, 11/19/1997

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

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Thu, 2008-07-03 at 06:42 -0700, Matthew Dharm wrote:
> On Thu, Jul 03, 2008 at 11:02:46AM +0200, Kay Sievers wrote:
> > On Thu, 2008-07-03 at 07:30 +0200, martin f krafft wrote:
> > > also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> > > > All are only unique per namespace/bus, and should better not be
> > > > mixed.
> > > 
> > > Well, but that's my point I suppose. Why do we have to treat the
> > > same drive differently, whether it's on the ATA bus or the USB bus?
> > > 
> > > My reason for bringing this up was that I was trying to boot off
> > > a harddrive via USB and had to change all the links, rather than
> > > just watching the system boot off the same disk it has always booted
> > > from.
> > 
> > So, what id's/symlinks is the same disk showing for ATA and USB in your
> > case?
> > 
> > Only advanced USB bridges read the number from the actual disk, I have a
> > bunch of enclosures here, which show all different id's for the same
> > disk inserted. Also usb-storage needs to encode the target and lun
> > number in the id, for multi-slot devices, so I do not think that will
> > work.
> 
> Isn't devlabel the technology that is supposed to be used for this sort of
> scenario?

No, that is replaced by /dev/disk/by-label/*

Thanks,
Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id symlinks should not include bus) type

[Matthew Dharm]

[-- Attachment #1: Type: text/plain, Size: 2018 bytes --]

On Thu, Jul 03, 2008 at 05:12:57PM +0200, Kay Sievers wrote:
> On Thu, 2008-07-03 at 06:42 -0700, Matthew Dharm wrote:
> > On Thu, Jul 03, 2008 at 11:02:46AM +0200, Kay Sievers wrote:
> > > On Thu, 2008-07-03 at 07:30 +0200, martin f krafft wrote:
> > > > also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> > > > > All are only unique per namespace/bus, and should better not be
> > > > > mixed.
> > > > 
> > > > Well, but that's my point I suppose. Why do we have to treat the
> > > > same drive differently, whether it's on the ATA bus or the USB bus?
> > > > 
> > > > My reason for bringing this up was that I was trying to boot off
> > > > a harddrive via USB and had to change all the links, rather than
> > > > just watching the system boot off the same disk it has always booted
> > > > from.
> > > 
> > > So, what id's/symlinks is the same disk showing for ATA and USB in your
> > > case?
> > > 
> > > Only advanced USB bridges read the number from the actual disk, I have a
> > > bunch of enclosures here, which show all different id's for the same
> > > disk inserted. Also usb-storage needs to encode the target and lun
> > > number in the id, for multi-slot devices, so I do not think that will
> > > work.
> > 
> > Isn't devlabel the technology that is supposed to be used for this sort of
> > scenario?
> 
> No, that is replaced by /dev/disk/by-label/*

Well, yes... I mean, you can *use* the label on the partition in a number
of ways.

But, my point was that instead of trying to ID a disk by some name which
may change depending on how it's attached to the system, use the devlabel
(via /dev/disk/by-label or mount options or whatever), which doesn't change
regardless of how the disk is attached to the system.

Matt

-- 
Matthew Dharm                              Home: mdharm-usb@one-eyed-alien.net 
Maintainer, Linux USB Mass Storage Driver

Type "format c:"  That should fix everything.
					-- Greg
User Friendly, 12/18/1997

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

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Thu, 2008-07-03 at 08:33 -0700, Matthew Dharm wrote:
> On Thu, Jul 03, 2008 at 05:12:57PM +0200, Kay Sievers wrote:
> > On Thu, 2008-07-03 at 06:42 -0700, Matthew Dharm wrote:
> > > On Thu, Jul 03, 2008 at 11:02:46AM +0200, Kay Sievers wrote:
> > > > On Thu, 2008-07-03 at 07:30 +0200, martin f krafft wrote:
> > > > > also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.02.1806 +0200]:
> > > > > > All are only unique per namespace/bus, and should better not be
> > > > > > mixed.
> > > > > 
> > > > > Well, but that's my point I suppose. Why do we have to treat the
> > > > > same drive differently, whether it's on the ATA bus or the USB bus?
> > > > > 
> > > > > My reason for bringing this up was that I was trying to boot off
> > > > > a harddrive via USB and had to change all the links, rather than
> > > > > just watching the system boot off the same disk it has always booted
> > > > > from.
> > > > 
> > > > So, what id's/symlinks is the same disk showing for ATA and USB in your
> > > > case?
> > > > 
> > > > Only advanced USB bridges read the number from the actual disk, I have a
> > > > bunch of enclosures here, which show all different id's for the same
> > > > disk inserted. Also usb-storage needs to encode the target and lun
> > > > number in the id, for multi-slot devices, so I do not think that will
> > > > work.
> > > 
> > > Isn't devlabel the technology that is supposed to be used for this sort of
> > > scenario?
> > 
> > No, that is replaced by /dev/disk/by-label/*
> 
> Well, yes... I mean, you can *use* the label on the partition in a number
> of ways.
> 
> But, my point was that instead of trying to ID a disk by some name which
> may change depending on how it's attached to the system, use the devlabel
> (via /dev/disk/by-label or mount options or whatever), which doesn't change
> regardless of how the disk is attached to the system.

Sure, symlinks based on filesystem metadata are the way to go here.
Usually the fs UUID is used in fstab and such, and works pretty well.

Thanks,
Kay

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[martin f krafft]

[-- Attachment #1: Type: text/plain, Size: 684 bytes --]

also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.03.1736 +0200]:
> Sure, symlinks based on filesystem metadata are the way to go here.
> Usually the fs UUID is used in fstab and such, and works pretty well.

Iff there is a UUID for the contents of the partition.

-- 
 .''`.   martin f. krafft <madduck@debian.org>
: :'  :  proud Debian developer, author, administrator, and user
`. `'`   http://people.debian.org/~madduck - http://debiansystem.info
  `-  Debian - when you have better things to do than fixing systems
 
there are two groups of people in the world: those who believe that
the world can be divided into two groups of people, and those who
don't.

[-- Attachment #2: Digital signature (see http://martin-krafft.net/gpg/) --]
[-- Type: application/pgp-signature, Size: 197 bytes --]

Re: not all volume types export UUIDs (was: /dev/disk/by-id

[Kay Sievers]

On Sat, 2008-07-05 at 13:55 +0200, martin f krafft wrote:
> also sprach Kay Sievers <kay.sievers@vrfy.org> [2008.07.03.1736 +0200]:
> > Sure, symlinks based on filesystem metadata are the way to go here.
> > Usually the fs UUID is used in fstab and such, and works pretty well.
> 
> Iff there is a UUID for the contents of the partition.

Sure, that's obvoius. I kind of miss the point of this conversation now.
Blindly removing the prefixes for id's which are not unique otherwise,
and usually not even the same across buses for the same device, seems
not a reasonable solution to me.
Care to explain your ideas again, and give examples for real hardware
you have, and base it on the answers/question in this thread?

Thanks,
Kay