Skip to contentSkip to conversation

LVM /dev and /proc problems and change proposal

linux-lvm Dale Kemp 2 participants 6 messages Jun 25, 2000 – Jun 26, 2000 1 message not in the archive

lore
Share & tools
lore

Dale Kemp

author
#0 lore raw

Hi all,

I've started using LVM thanks to Daniel Mehrmann 2.2.16 kernel patch and the debian lvm package (apt-get install lvm). I notice that vg's and

lv's are created in /dev this means the volume name is limited since it could
clash with similar named devices in the /dev directory. For example:

# ls -l /dev/par0
crw-rw----    1 root     lp         6,   0 Feb 23  1999 /dev/par0
# pvscan
pvscan -- reading all physical volumes (this may take a while...)
pvscan -- ACTIVE   PV "/dev/hda6" of VG "vdisk1" [3 GB / 3 GB free]
pvscan -- inactive PV "/dev/hde5"  is in no VG  [762.86 MB]
pvscan -- inactive PV "/dev/hde6"  is in no VG  [285.93 MB]
pvscan -- total: 3 [4.02 GB] / in use: 1 [3 GB] / in no VG: 2 [1.02 GB]

(create a volume group called par0)

# vgcreate par0 /dev/hde6
vgcreate -- INFO: using default physical extent size 4 MB
vgcreate -- INFO: maximum logical volume size is 255.99 Gigabyte
vgcreate -- doing automatic backup of volume group "par0"
vgcreate -- volume group "par0" successfully created and activated
# ls -ld /dev/par0
dr-xr-xr-x    2 root     root         1024 Jun 25 17:20 /dev/par0

Ouch! The device has been replace with a directory! Just adding protection here doesn't solve the problem of potential name clashes. I think the solution is to store these names in /etc/lvm* and use another directory level in /dev.

e.g.. /dev/lvm/

      |
      +---- lvm (109,0)
      |
      +---- <name1> +---- group (109,0)
      |             |
      |             +---- <lv_name1> (58,0)
      |             +---- <lv_name2> (58,1)
      |
      +---- <name2> +---- group (109,1)
      |             |
      |             +---- <lv_name3> (58,2)
      |
      +---- etc (not needed if add vg_ and lv_ to above names)
            |
            +---- future_item (890,0)

OR...

    /dev/lvm
      |
      |
      +-- <name1> (109,0) [optional forced vg_ prefix to stop
name-clash]
      +-- <name2> (109,1)
      |
      +-- <name1> (58,0) [optional forced lv_ prefix to stop name-clash]
      +-- <name2> (58,1)
      +-- <name3> (58,2)
      |
      +-- future_item (890,0)

OR... (my favorite)

    /dev/lvm
      |
      |
      +-- pv (example to show future expansion options)
      |    |
      |    +-- pv0 --> /dev/hda6
      |    +-- pv1 --> /dev/hde5
      |    +-- pv2 --> /dev/hde6
      |
      +-- vg
      |    |
      |    +-- <name1> (109,0)
      |    +-- <name2> (109,1)
      |
      +-- lv
      |    |
      |    +-- <name1> (58,0)
      |    +-- <name2> (58,1)
      |    +-- <name3> (58,2)
      |
      +-- future_item (890,0)

I use the prefix 'vg_' and 'lv_' to stop any future name clashes. I don't see
it being a good idea to list lv under the vg's since you might allow the transfer
of lv's between vg's. We use the tools to find out more info, or create readable
files in /etc/...

Also I think there should be a sub-directory for lvm itself in /proc ie. /proc/lvm/lvm instead of /proc/lvm.

Example:

    # pvcreate /dev/hde[56]
    # vgcreate vdisk2 /dev/hde5 /dev/hde6
    # lvcreate -n XFilesVol -L 150M vdisk2
    # mke2fs /dev/lvm/lv/XFilesVol
    # mount -t ext2 /dev/lvm/lv/XFileVol /mnt/XFiles

The current system has name space problems and will be more difficult to extend
in the future.

-- Dale ([email protected])

#1 lore raw

Hi Dale.

Quoted 27 lines · Dale Kemp

Hi all,

I've started using LVM thanks to Daniel Mehrmann 2.2.16 kernel patch and the debian lvm package (apt-get install lvm). I notice that vg's and

lv's are created in /dev this means the volume name is limited since it could
clash with similar named devices in the /dev directory. For example:

# ls -l /dev/par0
crw-rw----    1 root     lp         6,   0 Feb 23  1999 /dev/par0
# pvscan
pvscan -- reading all physical volumes (this may take a while...)
pvscan -- ACTIVE   PV "/dev/hda6" of VG "vdisk1" [3 GB / 3 GB free]
pvscan -- inactive PV "/dev/hde5"  is in no VG  [762.86 MB]
pvscan -- inactive PV "/dev/hde6"  is in no VG  [285.93 MB]
pvscan -- total: 3 [4.02 GB] / in use: 1 [3 GB] / in no VG: 2 [1.02 GB]

(create a volume group called par0)

# vgcreate par0 /dev/hde6
vgcreate -- INFO: using default physical extent size 4 MB
vgcreate -- INFO: maximum logical volume size is 255.99 Gigabyte
vgcreate -- doing automatic backup of volume group "par0"
vgcreate -- volume group "par0" successfully created and activated
Quoted 4 lines
# ls -ld /dev/par0
dr-xr-xr-x    2 root     root         1024 Jun 25 17:20 /dev/par0

Ouch!

You are right, this has not to happen.

It has already been fixed in 0.8final but was still possible in the older prerelease you are using.

Quoted 2 lines

The device has been replace with a directory! Just adding protection here doesn't solve the problem of potential name clashes.

But It dissallows the user to do it and there's still plenty of free namespace to find a meaningfull name for the Volume Group.

Quoted 4 lines

I think the solution is to store these names in /etc/lvm* and use another directory level in /dev.

e.g.. /dev/lvm/

Quoted 15 lines
      |
      +---- lvm (109,0)
      |
      +---- <name1> +---- group (109,0)
      |             |
      |             +---- <lv_name1> (58,0)
      |             +---- <lv_name2> (58,1)
      |
      +---- <name2> +---- group (109,1)
      |             |
      |             +---- <lv_name3> (58,2)
      |
      +---- etc (not needed if add vg_ and lv_ to above names)
            |
            +---- future_item (890,0)

There's a preprocessor definition LVM_DIR_PREFIX in the code, which supports this already.

Not much testing though...

Quoted 32 lines

OR...

    /dev/lvm
      |
      |
      +-- <name1> (109,0) [optional forced vg_ prefix to stop
name-clash]
      +-- <name2> (109,1)
      |
      +-- <name1> (58,0) [optional forced lv_ prefix to stop name-clash]
      +-- <name2> (58,1)
      +-- <name3> (58,2)
      |
      +-- future_item (890,0)

OR... (my favorite)

    /dev/lvm
      |
      |
      +-- pv (example to show future expansion options)
      |    |
      |    +-- pv0 --> /dev/hda6
      |    +-- pv1 --> /dev/hde5
      |    +-- pv2 --> /dev/hde6
      |
      +-- vg
      |    |
      |    +-- <name1> (109,0)
      |    +-- <name2> (109,1)
Quoted 8 lines
      |
      +-- lv
      |    |
      |    +-- <name1> (58,0)
      |    +-- <name2> (58,1)
      |    +-- <name3> (58,2)
      |
      +-- future_item (890,0)

I don't like both, because they break the hierarchy of LVs (like virtual partitions) contained in VGs (like virtual disks containing physical disks).

Quoted 6 lines

I use the prefix 'vg_' and 'lv_' to stop any future name clashes. I don't see
it being a good idea to list lv under the vg's since you might allow the transfer
of lv's between vg's.

A command for LV transfer support is not planned.

You can achive it today by merging two volume groups, use pvmove to seperate allocated PEs to PVs you like and split the VG afterwards.

This can be wrapped by a script to have a single command to move the LV.

Quoted 6 lines

We use the tools to find out more info, or create
readable
files in /etc/...

Also I think there should be a sub-directory for lvm itself in /proc ie. /proc/lvm/lvm instead of /proc/lvm.

Yep.
Already on the TODO list.

Quoted 11 lines

Example:

    # pvcreate /dev/hde[56]
    # vgcreate vdisk2 /dev/hde5 /dev/hde6
    # lvcreate -n XFilesVol -L 150M vdisk2
    # mke2fs /dev/lvm/lv/XFilesVol
    # mount -t ext2 /dev/lvm/lv/XFileVol /mnt/XFiles

The current system has name space problems and will be more difficult to extend in the future.

If you refer to /dev IMO you are wrong because there's plenty of namespace.

I share your point of view in regard to /proc.

Signature · 12 lines
Regards,
Heinz      -- The LVM guy --

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Sistina Software Inc.
Senior Consultant/Developer                       Bartningstr. 12
                                                  64289 Darmstadt
                                                  Germany
[email protected]                           +49 6151 7103 86
                                                       FAX 7103 96
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
author
#2 lore raw
Quoted 2 lines

But It dissallows the user to do it and there's still plenty of free namespace to find a meaningfull name for the Volume Group.

I'm refering to the future when a new device gets added that just might clash with a volume group name that a user has chosen. And we don't know what users are going to call their volume groups and we don't know what device names will be created in the future.

Quoted 2 lines

There's a preprocessor definition LVM_DIR_PREFIX in the code, which supports this already.

So I see. The simplest solution is to change this default prefix, and have the lvm stuff self-contained in its own subdirectory. Even without the name clash this is just tidier, already /dev is very wide but not very deep.

Quoted 2 lines

I don't like both, because they break the hierarchy of LVs (like virtual partitions) contained in VGs (like virtual disks containing physical disks).

Fair enough :-) Just throwing some ideas around.

Can I propose' that simply by changing the default (LVM_DIR_PREFIX) in the original LVM code and we begin standardizing' now the LVM directories.

eg.

/dev/lvm/ (all previuos

    |
    +-- lvm [c:109,0] (need to change constant LVM_DIR in "liblvm.h" too]
    |
    +-- vdisk1/
         |
         +-- group [c:109,0]
         +-- lvol1 [b:58,0]
         +-- lvol2 [b:58,1]
Quoted 6 lines
Quoted 2 lines

Also I think there should be a sub-directory for lvm itself in /proc ie. /proc/lvm/lvm instead of /proc/lvm.

Yep.
Already on the TODO list.
I share your point of view in regard to /proc.

The need for /proc/lvm/... is simular to that for /dev/lvm/...

-- Dale.

#3 lore raw
Quoted 7 lines · Dale Kemp
Quoted 2 lines

But It dissallows the user to do it and there's still plenty of free namespace to find a meaningfull name for the Volume Group.

I'm refering to the future when a new device gets added that just might clash with a volume group name that a user has chosen. And we don't know what users are going to call their volume groups and we don't know what device names will be created in the future.

vgrename(8) is able to address that.

Quoted 6 lines
Quoted 3 lines

There's a preprocessor definition LVM_DIR_PREFIX in the code, which supports this already.

So I see. The simplest solution is to change this default prefix, and have the lvm stuff self-contained in its own subdirectory.

Yes.
Please give it a try and tell me if it works.

Quoted 7 lines

Even without the name
clash this is just tidier, already /dev is very wide but not very deep.

Quoted 2 lines

I don't like both, because they break the hierarchy of LVs (like virtual partitions) contained in VGs (like virtual disks containing physical disks).

Fair enough :-) Just throwing some ideas around.

;-{)

Quoted 7 lines

Can I propose' that simply by changing the default (LVM_DIR_PREFIX) in the original LVM code and we begin standardizing' now the LVM directories.

eg.

/dev/lvm/ (all previuos

Quoted 8 lines
    |
    +-- lvm [c:109,0] (need to change constant LVM_DIR in "liblvm.h" too]
    |
    +-- vdisk1/
         |
         +-- group [c:109,0]
         +-- lvol1 [b:58,0]
         +-- lvol2 [b:58,1]

Sure.
Please try the LVM_DIR_PREFIX change.

Quoted 10 lines
Quoted 8 lines
Quoted 4 lines

Also I think there should be a sub-directory for lvm itself in /proc ie. /proc/lvm/lvm instead of /proc/lvm.

Yep.
Already on the TODO list.
I share your point of view in regard to /proc.

The need for /proc/lvm/... is simular to that for /dev/lvm/...

More or less.

/proc/* is for programs to parse the contained information.

Therfore its not primarily addressing a real name clash problem but the answer to the request for easy parsing.

Quoted 2 lines

-- Dale.

Signature · 12 lines
Regards,
Heinz      -- The LVM guy --

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Sistina Software Inc.
Senior Consultant/Developer                       Bartningstr. 12
                                                  64289 Darmstadt
                                                  Germany
[email protected]                           +49 6151 7103 86
                                                       FAX 7103 96
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
author
#4 lore raw
Quoted 5 lines
Quoted 3 lines

And we don't know
what users are going to call their volume groups and we don't know what device names will be created in the future.

vgrename(8) is able to address that.

That's not the point I'm making here.

    # vgcreate yup0
    # etc...

Creates /dev/yup0 and all is well...

Then an upgrade occurs with support for the new yup devices from Yups(tm) technology. Devices /dev/yup[0-5] are created, or it fails(?). If your yup0 was in /dev/lvm/yup0 then things would carry on like normal. We don't know what the user will name the vg's and we don't know the names of new devices to be added to linux. By using a sub directory:

- You keep all the stuff together in one place (well almost /etc/lvm/ /dev/lvm/
...)
- You get around any future name clash problems
- Its easier to find what your looking for
- Its just plain tidier
- Its a simple to do

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

There's a preprocessor definition LVM_DIR_PREFIX in the code, which supports this already.

So I see. The simplest solution is to change this default prefix, and have the lvm stuff self-contained in its own subdirectory.

Yes.
Please give it a try and tell me if it works.

Ok - I will. I suspect it will work fine looking at the code.

Quoted 4 lines
Quoted 4 lines
Quoted 3 lines

Yep.
Already on the TODO list.
I share your point of view in regard to /proc.

Quoted 8 lines
Quoted 1 line

The need for /proc/lvm/... is simular to that for /dev/lvm/...

More or less.

/proc/* is for programs to parse the contained information.

Therfore its not primarily addressing a real name clash problem but the answer to the request for easy parsing.

Yes for user and machine.

-- Dale.

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw
Quoted 23 lines · Dale Kemp
Quoted 12 lines · Heinz J. Mauelshagen
Quoted 7 lines · Dale Kemp
Quoted 2 lines

But It dissallows the user to do it and there's still plenty of free namespace to find a meaningfull name for the Volume Group.

I'm refering to the future when a new device gets added that just might clash with a volume group name that a user has chosen. And we don't know what users are going to call their volume groups and we don't know what device names will be created in the future.

vgrename(8) is able to address that.

Quoted 1 line
Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

There's a preprocessor definition LVM_DIR_PREFIX in the code, which supports this already.

So I see. The simplest solution is to change this default prefix, and have the lvm stuff self-contained in its own subdirectory.

Yes.
Please give it a try and tell me if it works.

Quoted 6 lines

Just changing LVM_DIR_PREFIX does not work. I updated code to automatically move the old files and directories around but it still doesn't work. The reason is it

does consistency checks and finds /dev/lvm/.. is not there but is /dev/... and it refuses. If the system stored relative names not absolute paths things might be ok, ie <VG>/<LV> and never /dev/.../<VG>/<LV>.

We need to change lvrename.c to address this.
As you say the absolute patch of the LV name is stored in the VGDA but lvrename will complain because it checks for LVM_DIR_PREFIX in the LV path too.

<SNIP>

Signature · 12 lines
Regards,
Heinz      -- The LVM guy --

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Sistina Software Inc.
Senior Consultant/Developer                       Bartningstr. 12
                                                  64289 Darmstadt
                                                  Germany
[email protected]                           +49 6151 7103 86
                                                       FAX 7103 96
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-