Basic LVM Step-By-Step

A presentation by Dustin Laurence for the SGVLUG, 7/8/04.

What Is LVM? What can it do for me?

• LVM == Logical Volume Manager
• Flexibility
• High availability
• Reliability

Key Benefits Of LVM:

• Arbitrary "partition" resizing (flexibility).
• Migrate data easily--swap drives on a running system (availability).
• (Future) Filesystem snapshots for atomic backups. (Just being able to call your backups "atomic" around the water cooler is worth it all by itself, right?) ;-) (reliability)

Less interesting (to me) features:

• Can duplicate some software RAID functionality (level 0, maybe 1).

The Plan Of The Presentation

• (My) hands-on demonstration
• Tutorial
• Gripping drama--will I royally bork my home machine in front of an audience?

A Little Background

• The Big Picture

• 
  

Terminology Roadmap

DOS Disk Object/Concept	LVM Equivalent
(Physical) Partition	LVM Logical Volume
Disk	LVM Volume Group
Block	LVM Physical Extent
	LVM Physical Volume
	(Physical) Partition

• Operators on block devices:

  (dev1, dev2, ...) -> newdev

• Things work in layers. Forget where you are and mix levels and you'll probably shred the filesystem.

• For purposes of this talk:

  • RAID == Linux Software RAID
  • LVM == LVM2

• Important Points and Caveats

• LVM and RAID are independent, (almost) orthogonal storage tools.
• LVM is a work in progress.
• "Some assembly required, batteries not included, etc."

Are you comfortable with fdisk & mkfs?

These tools don't care of their powers are used for good or evil! It ain't hard to shred a filesystem with them. =8-O

Operator Error is the operator's problem, not the tool builder's. You know, unix.

• Will you know how to recover the system if you render it unbootable? (Particularly if / is on LVM.) How's your backup strategy?

Tools and Requirements

To use LVM, we need to make sure our installation supports it:

• Kernel Drivers: multiple devices support (CONFIG_MD) and device mapper support (CONFIG_BLK_DEV_DM) at the very least (plus drivers for any RAID levels you are going to use LVM with, of course). In menuconfig for 2.6 the drivers are under Device Drivers->Multi-Device Support (RAID and LVM). (Surprise)
• For / on LVM (or RAID), you need the drivers built-in or an appropriate initrd. (Again, may be done for you.)

• You need the LVM2 userland tools: pvcreate, lvscan, etc. In Gentoo, just type 'emerge lvm2'.)

• You might prefer a graphical interface. Sucks to be you, I guess. :-) (But see EVMS.)
• LVM needs to be activated in your init scripts (pvscan). (For Gentoo, add checkfs to the boot runlevel.)
• If you run a mainstream distro, this might all be done for you (especially a server-oriented distro). (RH, SuSE?)

Constructing a Logical Volume

Assuming we have the necessary tools, we are ready to use them. I will be using the tools on my home machine as we go (that's the drama part, remember?).

1. RTFM

       # man lvm
       # lvm help
       # man pvcreate
       
       

   Unfortunately, the man pages are out of date (e.g. lvm2 format is the default, 'man lvm' doesn't list all the commands).

   The LVM-HOWTO

   Daniel Robbin's LVM articles for IBM (Part 1, Part 2).

   Gentoo-specific info

   SuSE-specific info

2. Configure LVM

       # man lvm.conf
       # emacs /etc/lvm/lvm.conf
       

3. Create suitable devices to feed to LVM:

    # cfdisk /dev/hda
    

(Don't forget to set the partition type: 0x8E for LVM, 0xFD for RAID.)

or

    # emacs /etc/raidtab; mkraid
    

or

    # mdadm --create /dev/md0 --level=raid1 -n 2 /dev/hda5 /dev/hdc5
    

4.     # pvcreate /dev/hd[eg]13 /dev/hde14
       

5.     # pvs
       # pvscan
       # pvdisplay /dev/hde13
       

6.     # vgcreate -s 32M newsystem /dev/hd[eg]13
       

7.     # vgs
       # vgscan
       # vgdisplay newsystem
       

8.     # lvcreate -L 300M --contiguous y --name root newsystem
       

9.     # lvs
       # lvscan
       # lvdisplay newsystem
       # lvdisplay newsystem/root
       

   Note the logical volume size is a multiple of the 32M physical extent size of the volume group.

10.     # mkreiserfs -l '/' /dev/newsystem/root
        

11.     # mkdir /mnt/newsystem
        # emacs /etc/fstab
        # mount /mnt/newsystem
        

A Few Simple Acts Of Humanity

•     # lvremove newsystem/root
      

  You can't remove a mounted logical volume, fortunately.

•     # vgremove newsystem
      

  You can't remove a volume group that contains logical volumes, either.

Extending A Logical Volume

1.     # vgs
       # lvcreate -L 1.1G --contiguous y --name usr newsystem
       

   There weren't enough contiguous extents on a single physical volume. (Note that LVM uses SI prefixes. It's probably a commie plot.)

2.     # lvcreate -L 1.1G --name usr newsystem
       # mkreiserfs -l '/usr' /dev/newsystem/usr
       # mkdir /mnt/newsystem/usr
       # emacs /etc/fstab
       # mount /mnt/newsystem/usr/
       

   All is well.

3.     # lvdisplay newsystem/usr
       # lvextend  -L +200M newsystem/usr
       # resize_reiserfs /dev/newsystem/usr
       

   What didn't I do first?

4.     # resize_reiserfs -s -500M /dev/newsystem/usr
       

   But must umount to shrink.

5.     # pvcreate /dev/hde14
       # vgextend newsystem /dev/hde14
       # lvextend -L +1G newsystem/usr
       # resize_reiserfs /dev/newsystem/usr
       # pvs
       

   Note that newsystem/usr is larger than any of the physical volumes in the volume group. As long as it isn't forced to be contiguous, LVM can extend a logical volume with any extent on any physical volume in the volume group.

Reducing A Logical Volume

1.     # umount /mnt/newsystem/usr
       # resize_reiserfs -s -900M /dev/newsystem/usr
       # lvreduce -L -800M /dev/newsystem/usr
       # resize_reiserfs /dev/newsystem/usr
       # mount /mnt/newsystem/usr/
       # df -h
       # pvs
       

Removing (Physical) Partitions

We should be able to do this:

1. If necessary, shrink some more

       # umount /mnt/newsystem/usr
       # resize_reiserfs -s -900M /dev/newsystem/usr
       # lvreduce -L -800M /dev/newsystem/usr
       # resize_reiserfs /dev/newsystem/usr
       # mount /mnt/newsystem/usr/
       # df -h
       # pvs
       

2.     # pvmove -v /dev/hde13
       

But it uses a temporary snapshot to do it, and guess what doesn't work yet?

Snapshotting and Backups

• Simple backup systems are NOT ATOMIC! =8-O
• In general, you can't unmount before backing up (root, servers, etc.)
• Snapshotting is the answer. Or it will be. It doesn't work in lvm2 yet.

LVM and RAID

• Essentially nothing changes.
• Order matters: RAID-on-LVM != LVM-on-RAID
• Much greater potential for confusion.

Booting From LVM (/boot on LVM)

• It can't be done. End of story.
• It wouldn't matter that much if it could.

root On LVM

The Problem With Root-On-LVM

• Hardest and most poorly documented aspect of LVM.
• Not necessary, and not that big a deal if you properly fine-grain your partition scheme. Just use RAID or a simple partition outside of your volume group (just like for /boot).
• There are chicken-and-egg problems both at installation time and at boot time.
• Your distro might have done this for you too.
• One way or the other, you need to use an initrd.

• Definitely RTFM:

  How to set up LVM on root under Gentoo

  Root-on-LVM-on-RAID HOWTO (from cold install)

  Root-on-RAID and Root-on-LVM-on-RAID HOWTO (from existing system

Alternatives to LVM

Right now, LVM seems like the best alternative. But on the horizon:

• EVMS
  • Totally integrated tools.
  • Command-line, curses, X-windows...
  • Great support.
  • No LVM2 support yet.
  • Still a full-contact administration tool, not a user-friendly shortcut for n00bs.

  • Daniel Robbins EVMS guide for IBM, Part 1 & Part 2.

    EVMS User Guide

• Dynamic Disks from the Evil Empire?!? Horrors!!!