AJ Lewis

lewis(at)sistina.com

0.5 2003-02-10 ajl Updated Redhat initscript information for 7.0 and above; Added information on removing a partition table from a disk if pvcreate fails; Default PE size is 32MB now; Updated method for snapshotting under XFS. 0.4 2002-12-16 ajl Updated for LVM 1.0.6 0.3 2002-09-16 ajl removed example pvmove from Command Operations section - we now just point to the more detailed recipe on pvmove that contains various warnings and such 0.2 2002-09-11 ajl Updated for LVM 1.0.5 and converted to DocBook XML 4.1.2. 0.1 2002-04-28 gf Initial conversion from Sistina's LaTeX source and import to tLDP in LinuxDoc format. 2002 Sistina Software, Inc This document is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY, either expressed or implied. While every effort has been taken to ensure the accuracy of the information documented herein, the author(s)/editor(s)/maintainer(s)/contributor(s) assumes NO RESPONSIBILITY for any errors, or for any damages, direct or consequential, as a result of the use of the information documented herein. This document describes how to build, install, and configure LVM for Linux. A basic description of LVM is also included. This version of the HowTo is for 1.0.6. This is an attempt to collect everything needed to know to get LVM up and running. The entire process of getting, compiling, installing, and setting up LVM will be covered. Pointers to LVM configurations that have been tested with will also be included. This version of the HowTo is for LVM 1.0.6. All previous versions of LVM are considered obsolete and are only kept for historical reasons. This document makes no attempt to explain or describe the workings or use of those versions. We will keep the latest version of this HOWTO in the CVS with the other LDP HowTos. You can get it by checking out ``LDP/howto/linuxdoc/LVM-HOWTO.sgml'' from the same CVS server as GFS. You should always be able to get a human readable version of this HowTo from the http://www.tldp.org/HOWTO/LVM-HOWTO.html This document is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY, either expressed or implied. While every effort has been taken to ensure the accuracy of the information documented herein, the author(s)/editor(s)/maintainer(s)/contributor(s) assumes NO RESPONSIBILITY for any errors, or for any damages, direct or consequential, as a result of the use of the information documented herein. List of everyone who has put words into this file. AJ Lewis Joe Thornber" Patrick Caulfield Please notify the HowTo maintainer if you believe you should be listed above. LVM is a Logical Volume Manager implemented by Heinz Mauelshagen for the Linux operating system. As of kernel version 2.4, LVM is incorporated in the main kernel source tree. This does not mean, however, that your 2.4.x kernel is up to date with the latest version of LVM. Look at the README for the latest information about which kernels have the latest code in them. Logical volume management provides a higher-level view of the disk storage on a computer system than the traditional view of disks and partitions. This gives the system administrator much more flexibility in allocating storage to applications and users. Storage volumes created under the control of the logical volume manager can be resized and moved around almost at will, although this may need some upgrading of file system tools. The logical volume manager also allows management of storage volumes in user-defined groups, allowing the system administrator to deal with sensibly named volume groups such as "development" and "sales" rather than physical disk names such as "sda" and "sdb". Logical volume management is traditionally associated with large installations containing many disks but it is equally suited to small systems with a single disk or maybe two. One of the difficult decisions facing a new user installing Linux for the first time is how to partition the disk drive. The need to estimate just how much space is likely to be needed for system files and user files makes the installation more complex than is necessary and some users simply opt to put all their data into one large partition in an attempt to avoid the issue. Once the user has guessed how much space is needed for /home /usr / (or has let the installation program do it) then is quite common for one of these partitions to fill up even if there is plenty of disk space in one of the other partitions. With logical volume management, the whole disk would be allocated to a single volume group and logical volumes created to hold the / /usr and /home file systems. If, for example the /home logical volume later filled up but there was still space available on /usr then it would be possible to shrink /usr by a few megabytes and reallocate that space to /home. Another alternative would be to allocate minimal amounts of space for each logical volume and leave some of the disk unallocated. Then, when the partitionsstart to fill up, they can be expanded as necessary. As an example: Joe buys a PC with an 8.4 Gigabyte disk on it and installs Linux using the following partitioning system: /boot /dev/hda1 10 Megabytes swap /dev/hda2 256 Megabytes / /dev/hda3 2 Gigabytes /home /dev/hda4 6 Gigabyes This, he thinks, will maximize the amount of space available for all his MP3 files. Sometime later Joe decides that he want to install the latest office suite and desktop UI available but realizes that the root partition isn't large enough. But, having archived all his MP3s onto a new writable DVD drive there is plenty of space on /home. His options are not good: Reformat the disk, change the partitioning scheme and reinstall. Buy a new disk and figure out some new partitioning scheme that will require the minimum of data movement. Set up a symlink farm from / to /home and install the new software on /home With LVM this becomes much easier: Jane buys a similar PC but uses LVM to divide up the disk in a similar manner: /boot /dev/vg00/boot 10 Megabytes swap /dev/vg00/swap 256 Megabytes / /dev/vg00/root 2 Gigabytes /home /dev/vg00/home 6 Gigabytes When she hits a similar problem she can reduce the size of /home by a gigabyte and add that space to the root partition. Suppose that Joe and Jane then manage to fill up the /home partition as well and decide to add a new 20 Gigabyte disk to their systems. Joe formats the whole disk as one partition (/dev/hdb1) and moves his existing /home data onto it and uses the new disk as /home. But he has 6 gigabytes unused or has to use symlinks to make that disk appear as an extension of /home, say /home/joe/old-mp3s. Jane simply adds the new disk to her existing volume group and extends her /home logical volume to include the new disk. Or, in fact, she could move the data from /home on the old disk to the new disk and then extend the existing root volume to cover all of the old disk. The benefits of logical volume management are more obvious on large systems with many disk drives. Managing a large disk farm is a time-consuming job, made particularly complex if the system contains many disks of different sizes. Balancing the (often conflicting) storage requirements of various users can be a nightmare. User groups can be allocated to volume groups and logical volumes and these can be grown as required. It is possible for the system administrator to "hold back" disk storage until it is required. It can then be added to the volume(user) group that has the most pressing need. When new drive are added to the system, it is no longer necessary to move users files around to make the best use of the new storage; simply add the new disk into an exiting volume group or groups and extend the logical volumes as necessary. It is also easy to take old drives out of service by moving the data from them onto newer drives - this can be done online, without disrupting user service. To learn more about LVM, please take a look at the other papers available at Logical Volume Manager: Publications, Presentations and Papers . This diagram gives a overview of the main elements in an LVM system: +-- Volume Group --------------------------------+ | | | +----------------------------------------+ | | PV | PE | PE | PE | PE | PE | PE | PE | PE | | | +----------------------------------------+ | | . . . . | | . . . . | | +----------------------------------------+ | | LV | LE | LE | LE | LE | LE | LE | LE | LE | | | +----------------------------------------+ | | . . . . | | . . . . | | +----------------------------------------+ | | PV | PE | PE | PE | PE | PE | PE | PE | PE | | | +----------------------------------------+ | | | +------------------------------------------------+ Another way to look at is this (courtesy of Erik Bågfors on the linux-lvm mailing list): hda1 hdc1 (PV:s on partitions or whole disks) \ / \ / diskvg (VG) / | \ / | \ usrlv rootlv varlv (LV:s) | | | ext2 reiserfs xfs (filesystems) The Volume Group is the highest level abstraction used within the LVM. It gathers together a collection of Logical Volumes and Physical Volumes into one administrative unit. A physical volume is typically a hard disk, though it may well just be a device that 'looks' like a hard disk (eg. a software raid device). The equivalent of a disk partition in a non-LVM system. The LV is visible as a standard block device; as such the LV can contain a file system (eg. /home). Each physical volume is divided chunks of data, known as physical extents, these extents have the same size as the logical extents for the volume group. Each logical volume is split into chunks of data, known as logical extents. The extent size is the same for all logical volumes in the volume group. A concrete example will help: Lets suppose we have a volume group called VG1, this volume group has a physical extent size of 4MB. Into this volume group we introduce 2 hard disk partitions, /dev/hda1 and /dev/hdb1. These partitions will become physical volumes PV1 and PV2 (more meaningful names can be given at the administrators discretion). The PV's are divided up into 4MB chunks, since this is the extent size for the volume group. The disks are different sizes and we get 99 extents in PV1 and 248 extents in PV2. We now can create ourselves a logical volume, this can be any size between 1 and 347 (248 + 99) extents. When the logical volume is created a mapping is defined between logical extents and physical extents, eg. logical extent 1 could map onto physical extent 51 of PV1, data written to the first 4 MB of the logical volume in fact be written to the 51st extent of PV1. The administrator can choose between a couple of general strategies for mapping logical extents onto physical extents: Linear mapping will assign a range of PE's to an area of an LV in order eg., LE 1 - 99 map to PV1 and LE 100 - 347 map onto PV2. Striped mapping will interleave the chunks of the logical extents across a number of physical volumes eg., 1st chunk of LE[1] -ent PV1[1], 2nd chunk of LE[1] -ent PV2[1], 3rd chunk of LE[1] -ent PV3[1], 4th chunk of LE[1] -ent PV1[2], and so on. In certain situations this strategy can improve the performance of the logical volume. Be aware however, that LVs created using striping cannot be extended past the PVs they were originally created on. A wonderful facility provided by LVM is 'snapshots'. This allows the administrator to create a new block device which is an exact copy of a logical volume, frozen at some point in time. Typically this would be used when some batch processing, a backup for instance, needs to be performed on the logical volume, but you don't want to halt a live system that is changing the data. When the snapshot device has been finished with the system administrator can just remove the device. This facility does require that the snapshot be made at a time when the data on the logical volume is in a consistent state, later sections of this document give some examples of this. More information on snapshots can be found in Taking a Backup Using Snapshots. The first thing you need to do is get a copy of LVM. Download via FTP a tarball of LVM. Download the source that is under active development via CVS There are source tarballs for the latest version . The LVM kernel patch must be generated using the LVM source. More information regarding this can be found at the section on Building the kernel module. Note: the state of code in the CVS repository fluctuates wildly. It will contain bugs. Maybe ones that will crash LVM or the kernel. It may not even compile. Consider it alpha-quality code. You could lose data. You have been warned. To follow the development progress of LVM, subscribe to the LVM mailing lists, lvm-devel and lvm-commit. To build LVM from the CVS sources, you must have several GNU tools: the CVS client version 1.9 or better GCC 2.95.2 GNU make 3.79 autoconf, version 2.13 or better To make life easier in the future with regards to updating the CVS tree create the file $HOME/.cvsrc and insert the following lines. This configures useful defaults for the three most commonly used CVS commands. Do this now before proceeding any further. diff -u -b -B checkout -P update -d -P Also, if you are on a slow net link (like a dialup), you will want to add a line containing cvs -z5 in this file. This turns on a useful compression level for all CVS commands. Before downloading the development source code for the first time it is required to log in to the server: # cvs -d :pserver:cvs@tech.sistina.com:/data/cvs login The password is `cvs1'. The command outputs nothing if successful and an error message if it fails. Only an initial login is required. All subsequent CVS commands read the password stored in the file $HOME/.cvspass for authentication. The following CVS checkout command will retrieve an initial copy of the code. # cvs -d :pserver:cvs@tech.sistina.com:/data/cvs checkout LVM This will create a new directory LVM in your current directory containing the latest, up-to-the-hour LVM code. CVS commands work from anywhere inside the source tree, and recurse downwards. So if you happen to issue an update from inside the `tools' subdirectory it will work fine, but only update the tools directory and it's subdirectories. In the following command examples it is assumed that you are at the top of the source tree. Code changes are made fairly frequently in the CVS repository. Announcements of this are automatically sent to the lvm-commit list. You can update your copy of the sources to match the master repository with the update command. It is not necessary to check out a new copy. Using update is significantly faster and simpler, as it will download only patches instead of entire files and update only those files that have changed since your last update. It will automatically merge any changes in the CVS repository with any local changes you have made as well. Just cd to the directory you'd like to update and then type the following. # cvs update If you did not specify a tag when you checked out the source, this will update your sources to the latest version on the main branch. If you specified a branch tag, it will update to the latest version on that branch. If you specified a version tag, it will not do anything. Discuss your ideas on the developers list before you start. Someone may be working on the same thing you have in mind or they may have some good ideas about how to go about it. So, have you found a bug you want to fix? Want to implement a feature from the TODO list? Got a new feature to implement? Hacking the code couldn't be easier. Just edit your copy of the sources. No need to copy files to .orig or anything. CVS has copies of the originals. When you have your code in a working state and have tested as best you can with the hardware you have, generate a patch against the current sources in the CVS repository. # cvs update # cvs diff ent patchfile Mail the patch to the lvm-devel list with a description of what changes or additions you implemented. If someone else has been working on the same files as you have, you may find that there are conflicting modifications. You'll discover this when you try to update your sources. # cvs update RCS file: LVM/tools/pvcreate.c,v retrieving revision 1.5 retrieving revision 1.6 Merging differences between 1.5 and 1.6 into pvcreate.c rcsmerge: warning: conflicts during merge cvs server: conflicts found in tools/pvcreate.c C tools/pvcreate.c Don't panic! Your working file, as it existed before the update, is saved under the filename .#pvcreate.c.1.5. You can always recover it should things go horribly wrong. The file named `pvcreate.c' now contains both the old (i.e. your) version and new version of lines that conflicted. You simply edit the file and resolve each conflict by deleting the unwanted version of the lines involved. ententententententent pvcreate.c j++; ======= j--; ententententententent 1.6 Don't forget to delete the lines with all the ``ent'', ``='', and ``ent'' symbols. To use LVM you will have to build the LVM kernel module (recommended), or if you prefer rebuild the kernel with the LVM code statically linked into it. Your Linux system is probably based on one of the popular distributions (eg., Redhat, Debian) in which case it is possible that you already have the LVM module. Check the version of the tools you have on your system. You can do this by running any of the LVM command line tools with the '-h' flag. Use pvscan -h if you don't know any of the commands. If the version number listed at the top of the help listing is LVM 1.0.6, use your current setup and avoid the rest of this section. In order to patch the linux kernel to support LVM 1.0.6, you must do the following: Unpack LVM 1.0.6 # tar zxf lvm_1.0.6.tar.gz Enter the root directory of that version. # cd LVM/1.0.6 Run configure # ./configure You will need to pass the option --with-kernel_dir to configure if your linux kernel source is not in /usr/src/linux. (Run ./configure --help to see all the options available) Enter the PATCHES directory # cd PATCHES Run 'make' # make You should now have a patch called lvm-1.0.6-$KERNELVERSION.patch in the patches directory. This is the LVM kernel patch referenced in later sections of the howto. Patch the kernel # cd /usr/src/linux ; patch -pX ent /directory/lvm-1.0.6-$KERNELVERSION.patch The 2.2 series kernel needs to be patched before you can start building, look elsewhere for instructions on how to patch your kernel. Patches: rawio patch Stephen Tweedie's raw_io patch which can be found at http://www.kernel.org/pub/linux/kernel/people/sct/raw-io lvm patch The relevant LVM patch which should be built out of the PATCHES sub-directory of the LVM distribution. More information can be found in , Building a patch for your kernel. Once the patches have been correctly applied, you need to make sure that the module is actually built, LVM lives under the block devices section of the kernel config, you should probably request that the LVM /proc information is compiled as well. Build the kernel modules as usual. The 2.4 kernel comes with LVM already included although you should check at the Sistina web site for updates, (eg. v2.4.9 kernels and earlier must have the latest LVM patch applied ). When configuring your kernel look for LVM under Multi-device support (RAID and LVM). LVM can be compiled into the kernel or as a module. Build your kernel and modules and install then in the usual way. If you chose to build LVM as a module it will be called lvm-mod.o If you want to use snapshots with ReiserFS, make sure you apply the linux-2.4.x-VFS-lock patch (there are copies of this in the LVM/1.0.6/PATCHES directory.) If your kernel was compiled with the /proc file system (most are) then you can verify that LVM is present by looking for a /proc/lvm directory. If this doesn't exist then you may have to load the module with the command # modprobe lvm-mod If /proc/lvm still does not exist then check your kernel configuration carefully. When LVM is active you will see entries in /proc/lvm for all your physical volumes, volume groups and logical volumes. In addition there is a file called /proc/lvm/global which gives a summary of the LVM status and also shows just which version of the LVM kernel you are using. Boot-time scripts are not provided as part of the LVM distribution, however these are quite simple to do for yourself. The startup of LVM requires just the following two commands: # vgscan # vgchange -ay And the shutdown only one: # vgchange -an Follow the instructions below depending on the distribution of Linux you are running. It is necessary to edit the file /etc/rc.d/rc.boot. Look for the line that says Mounting local filesystems and insert the vgscan and vgchange commands just before it. You may also want to edit the the file /etc/rc.d/init.d/halt to deactivate the volume groups at shutdown. Insert the vgchange -an command near the end of this file just after the filesystems are unmounted or mounted read-only, before the comment that says Now halt or reboot. If you download the debian lvm tool package, an initscript should be installed for you. If you are installing LVM from source, you will still need to build your own initscript: Create a startup script in /etc/init.d/lvm containing the following: #!/bin/sh case "$1" in start) /sbin/vgscan /sbin/vgchange -ay ;; stop) /sbin/vgchange -an ;; restart|force-reload) ;; esac exit 0 Then execute the commands # chmod 0755 /etc/init.d/lvm # update-rc.d lvm start 26 S . stop 82 1 . Note the dots in the last command. No initscript modifications should be necessary for current versions of Mandrake. For Redhat 7.0 and up, you should not need to modify any initscripts to enable LVM at boot time if LVM is built into the kernel. If LVM is built as a module, it may be necessary to modify /etc/rc.d/rc.sysinit to load the LVM module before the section that reads: LVM initialization, take 2 (it could be on top of RAID) if [ -e /proc/lvm -a -x /sbin/vgchange -a -f /etc/lvmtab ]; then /sbin/vgchange -a y fi This init script fragment is from RedHat 7.3 - other versions of Redhat may look slightly different. For versions of Redhat older than 7.0, it is necessary to edit the file /etc/rc.d/rc.sysinit. Look for the line that says Mount all other filesystems and insert the vgscan and vgchange commands just before it. You should be sure that your root file system is mounted read/write before you run the LVM commands. You may also want to edit the the file /etc/rc.d/init.d/halt to deactivate the volume groups at shutdown. Insert the vgchange -an command near the end of this file just after the filesystems are mounted read-only, before the comment that says Now halt or reboot. Slackware 8.1 requires no updating of boot time scripts in order to make LVM work. For versions previous to Slackware 8.1, you should apply the following patch to /etc/rc.d/rc.S cd /etc/rc.d cp -a rc.S rc.S.old patch -p0 ent rc.S.diff (the cp part to make a backup in case). ----- snip snip file: rc.S.diff--------------- --- rc.S.or Tue Jul 17 18:11:20 2001 +++ rc.S Tue Jul 17 17:57:36 2001 @@ -4,6 +4,7 @@ # # Mostly written by: Patrick J. Volkerding, entvolkerdi@slackware.coment # +# Added LVM support enttgs@iafrica.coment PATH=/sbin:/usr/sbin:/bin:/usr/bin @@ -28,19 +29,21 @@ READWRITE=yes fi + # Check the integrity of all filesystems if [ ! READWRITE = yes ]; then - /sbin/fsck -A -a + /sbin/fsck -a / + # Check only the root fs first, but no others # If there was a failure, drop into single-user mode. if [ ? -gt 1 ] ; then echo echo - echo "*******************************************************" - echo "*** An error occurred during the file system check. ***" - echo "*** You will now be given a chance to log into the ***" - echo "*** system in single-user mode to fix the problem. ***" - echo "*** Running 'e2fsck -v -y entpartitionent' might help. ***" - echo "*******************************************************" + echo "************************************************************" + echo "*** An error occurred during the root file system check. ***" + echo "*** You will now be given a chance to log into the ***" + echo "*** system in single-user mode to fix the problem. ***" + echo "*** Running 'e2fsck -v -y entpartitionent' might help. ***" + echo "************************************************************" echo echo "Once you exit the single-user shell, the system will reboot." echo @@ -82,6 +85,44 @@ echo -n "get into your machine and start looking for the problem. " read junk; fi + # okay / fs is clean, and mounted as rw + # This was an addition, limits vgscan to /proc thus + # speeding up the scan immensely. + /sbin/mount /proc + + # Initialize Logical Volume Manager + /sbin/vgscan + /sbin/vgchange -ay + + /sbin/fsck -A -a -R + #Check all the other filesystem, including the LVM's, excluding / + + # If there was a failure, drop into single-user mode. + if [ ? -gt 1 ] ; then + echo + echo + echo "*******************************************************" + echo "*** An error occurred during the file system check. ***" + echo "*** You will now be given a chance to log into the ***" + echo "*** system in single-user mode to fix the problem. ***" + echo "*** Running 'e2fsck -v -y entpartitionent' might help. ***" + echo "*** The root filesystem is ok and mounted readwrite ***" + echo "*******************************************************" + echo + echo "Once you exit the single-user shell, the system will reboot." + echo + + PS1="(Repair filesystem) #"; export PS1 + sulogin + + echo "Unmounting file systems." + umount -a -r + mount -n -o remount,ro / + echo "Rebooting system." + sleep 2 + reboot + fi + else echo "Testing filesystem status: read-write filesystem" if cat /etc/fstab | grep ' / ' | grep umsdos 1> /dev/null 2> /dev/null ; then @@ -111,14 +152,16 @@ echo -n "Press ENTER to continue. " read junk; fi + fi + # remove /etc/mtab* so that mount will create it with a root entry /bin/rm -f /etc/mtab* /etc/nologin /etc/shutdownpid # mount file systems in fstab (and create an entry for /) # but not NFS or SMB because TCP/IP is not yet configured -/sbin/mount -a -v -t nonfs,nosmbfs +/sbin/mount -a -v -t nonfs,nosmbfs,proc # Clean up temporary files on the /var volume: /bin/rm -f /var/run/utmp /var/run/*.pid /var/log/setup/tmp/* --snip snip snip end of file--------------- No changes should be necessary from 6.4 onward as LVM is included Change into the LVM directory and do a ./configure followed by make. This will make all of the libraries and programs. If the need arises you can change some options with the configure script. Do a ./configure --help to determine which options are supported. Most of the time this will not be necessary. There should be no errors from the build process. If there are, see the Reporting Errors and Bugs on how to report this. You are welcome to fix them and send us the patches too. Patches are generally sent to the lvm-devel list. After the LVM source compiles properly, simply run make install to install the LVM library and tools onto your system. To remove the library and tools you just installed, run make remove. You must have the original source tree you used to install LVM to use this feature. Transitioning from previous versions of LVM to LVM 1.0.6 should be fairly painless. We have come up with a method to read in PV version 1 metadata (LVM 0.9.1 Beta7 and earlier) as well as PV version 2 metadata (LVM 0.9.1 Beta8 and LVM 1.0). Warning: New PVs initialized with LVM 1.0.6 are created with the PV version 1 on-disk structure. This means that LVM 0.9.1 Beta8 and LVM 1.0 cannot read or use PVs created with 1.0.6. There are just a few simple steps to transition this setup, but it is still recommended that you backup your data before you try it. You have been warned. Build LVM kernel and modules Follow the steps outlined in - for instructions on how to get and build the necessary kernel components of LVM. Build the LVM user tools Follow the steps in to build and install the user tools for LVM. Setup your init scripts Make sure you have the proper init scripts setup as per . Boot into the new kernel Make sure your boot-loader is setup to load the new LVM-enhanced kernel and, if you are using LVM modules, put an insmod lvm-mod into your startup script OR extend /etc/modules.conf (formerly /etc/conf.modules) by adding alias block-major-58 lvm-mod alias char-major-109 lvm-mod to enable modprobe to load the LVM module (don't forget to enable kmod). Reboot and enjoy. This is relatively straightforward if you follow the steps carefully. It is recommended you have a good backup and a suitable rescue disk handy just in case. The normal way of running an LVM root file system is to have a single non-LVM partition called /boot which contains the kernel and initial RAM disk needed to start the system. The system I upgraded was as follows: # df Filesystem 1k-blocks Used Available Use% Mounted on /dev/rootvg/root 253871 93384 147380 39% / /dev/hda1 17534 12944 3685 78% /boot /dev/rootvg/home 4128448 4568 3914168 0% /home /dev/rootvg/usr 1032088 332716 646944 34% /usr /dev/rootvg/var 253871 31760 209004 13% /var /boot contains the old kernel and an initial RAM disk as well as the LILO boot files and the following entry in /etc/lilo.conf # ls /boot System.map lost+found vmlinux-2.2.16lvm map module-info boot.0300 boot.b os2_d.b chain.b initrd.gz # tail /etc/lilo.conf image=/boot/vmlinux-2.2.16lvm label=lvm08 read-only root=/dev/rootvg/root initrd=/boot/initrd.gz append="ramdisk_size=8192" Build LVM kernel and modules Follow the steps outlined in - for instructions on how to get and build the necessary kernel components of LVM. Build the LVM user tools Follow the steps in to build and install the user tools for LVM. Install the new tools. Once you have done this you cannot do any LVM manipulation as they are not compatible with the kernel you are currently running. Rename the existing initrd.gz This is so it doesn't get overwritten by the new one # mv /boot/initrd.gz /boot/initrd08.gz Edit /etc/lilo.conf Make the existing boot entry point to the renamed file. You will need to reboot using this if something goes wrong in the next reboot. The changed entry will look something like this: image=/boot/vmlinux-2.2.16lvm label=lvm08 read-only root=/dev/rootvg/root initrd=/boot/initrd08.gz append="ramdisk_size=8192" Run lvmcreate_initrd to create a new initial RAM disk # lvmcreate_initrd 2.4.9 Don't forget the put the new kernel version in there so that it picks up the correct modules. Add a new entry into /etc/lilo.conf This new entry is to boot the new kernel with its new initrd. image=/boot/vmlinux-2.4.9lvm label=lvm10 read-only root=/dev/rootvg/root initrd=/boot/initrd.gz append="ramdisk_size=8192" Re-run lilo This will install the new boot block # /sbin/lilo Reboot When you get the LILO prompt select the new entry name (in this example lvm10) and your system should boot into Linux using the new LVM version. If the new kernel does not boot, then simply boot the old one and try to fix the problem. It may be that the new kernel does not have all the correct device drivers built into it, or that they are not available in the initrd. Remember that all device drivers (apart from LVM) needed to access the root device should be compiled into the kernel and not as modules. If you need to do any LVM manipulation when booted back into the old version, then simply recompile the old tools and install them with # make install If you do this, don't forget to install the new tools when you reboot into the new LVM version. When you are happy with the new system remember to change the ``default='' entry in your lilo.conf file so that it is the default kernel. The following sections outline some common administrative tasks for an LVM system. This is no substitute for reading the man pages. Before you can use a disk or disk partition as a physical volume you will have to initialize it: For entire disks: Run pvcreate on the disk: # pvcreate /dev/hdb This creates a volume group descripter at the start of disk. If you get an error that LVM can't initialize a disk with a partition table on it, first make sure that the disk you are operating on is the correct one. If you are very sure that it is, run the following: The following commands will destroy the partition table on the disk being operated on. Be very sure it is the correct disk. # dd if=/dev/zero of=/dev/diskname bs=1k count=1 # blockdev --rereadpt /dev/diskname For partitions: Set the partition type to 0x8e using fdisk or some other similar program. Run pvcreate on the partition: # pvcreate /dev/hdb1 This creates a volume group descriptor at the start of the /dev/hdb1 partition. Use the 'vgcreate' program: # vgcreate my_volume_group /dev/hda1 /dev/hdb1 NOTE: If you are using devfs it is essential to use the full devfs name of the device rather than the symlinked name in /dev. so the above would be: # vgcreate my_volume_group /dev/ide/host0/bus0/target0/lun0/part1 \ /dev/ide/host0/bus0/target1/lun0/part1 You can also specify the extent size with this command if the default of 32MB is not suitable for you with the '-s' switch. In addition you can put some limits on the number of physical or logical volumes the volume can have. After rebooting the system or running vgchange -an, you will not be able to access your VGs and LVs. To reactivate the volume group, run: # vgchange -a y my_volume_group Make sure that no logical volumes are present in the volume group, see later section for how to do this. Deactivate the volume group: # vgchange -a n my_volume_group Now you actually remove the volume group: # vgremove my_volume_group Use 'vgextend' to add an initialized physical volume to an existing volume group. # vgextend my_volume_group /dev/hdc1 ^^^^^^^^^ new physical volume Make sure that the physical volume isn't used by any logical volumes by using then 'pvdisplay' command: # pvdisplay /dev/hda1 --- Physical volume --- PV Name /dev/hda1 VG Name myvg PV Size 1.95 GB / NOT usable 4 MB [LVM: 122 KB] PV# 1 PV Status available Allocatable yes (but full) Cur LV 1 PE Size (KByte) 4096 Total PE 499 Free PE 0 Allocated PE 499 PV UUID Sd44tK-9IRw-SrMC-MOkn-76iP-iftz-OVSen7 If the physical volume is still used you will have to migrate the data to another physical volume. Then use 'vgreduce' to remove the physical volume: # vgreduce my_volume_group /dev/hda1 Decide which physical volumes you want the logical volume to be allocated on, use 'vgdisplay' and 'pvdisplay' to help you decide. To create a 1500MB linear LV named 'testlv' and its block device special '/dev/testvg/testlv': # lvcreate -L1500 -ntestlv testvg To create a 100 LE large logical volume with 2 stripes and stripesize 4 KB. # lvcreate -i2 -I4 -l100 -nanothertestlv testvg If you want to create an LV that uses the entire VG, use vgdisplay to find the Total PE size, then use that when running lvcreate. # vgdisplay testvg | grep "Total PE" Total PE 10230 # lvcreate -l 10230 testvg -n mylv This will create an LV called mylv filling the testvg VG. A logical volume must be closed before it can be removed: # umount /dev/myvg/homevol # lvremove /dev/myvg/homevol lvremove -- do you really want to remove "/dev/myvg/homevol"? [y/n]: y lvremove -- doing automatic backup of volume group "myvg" lvremove -- logical volume "/dev/myvg/homevol" successfully removed To extend a logical volume you simply tell the lvextend command how much you want to increase the size. You can specify how much to grow the volume, or how large you want it to grow to: # lvextend -L12G /dev/myvg/homevol lvextend -- extending logical volume "/dev/myvg/homevol" to 12 GB lvextend -- doing automatic backup of volume group "myvg" lvextend -- logical volume "/dev/myvg/homevol" successfully extended will extend /dev/myvg/homevol to 12 Gigabytes. # lvextend -L+1G /dev/myvg/homevol lvextend -- extending logical volume "/dev/myvg/homevol" to 13 GB lvextend -- doing automatic backup of volume group "myvg" lvextend -- logical volume "/dev/myvg/homevol" successfully extended will add another gigabyte to /dev/myvg/homevol. After you have extended the logical volume it is necessary to increase the file system size to match. how you do this depends on the file system you are using. By default, most file system resizing tools will increase the size of the file system to be the size of the underlying logical volume so you don't need to worry about specifying the same size for each of the two commands. ext2 Unless you have patched your kernel with the ext2online patch it is necessary to unmount the file system before resizing it. # umount /dev/myvg/homevol/dev/myvg/homevol # resize2fs /dev/myvg/homevol # mount /dev/myvg/homevol /home If you don't have e2fsprogs 1.19 or later, you can download the ext2resize command from ext2resize.sourceforge.net and use that: # umount /dev/myvg/homevol/dev/myvg/homevol # resize2fs /dev/myvg/homevol # mount /dev/myvg/homevol /home For ext2 there is an easier way. LVM ships with a utility called e2fsadm which does the lvextend and resize2fs for you (it can also do file system shrinking, see the next section) so the single command # e2fsadm -L+1G /dev/myvg/homevol is equivalent to the two commands: # lvextend -L+1G /dev/myvg/homevol # resize2fs /dev/myvg/homevol You will still need to unmount the file system before running e2fsadm. reiserfs Reiserfs file systems can be resized when mounted or unmounted as you prefer: Online: # resize_reiserfs -f /dev/myvg/homevol Offline: # umount /dev/myvg/homevol # resize_reiserfs /dev/myvg/homevol # mount -treiserfs /dev/myvg/homevol /home xfs XFS file systems must be mounted to be resized and the mount-point is specified rather than the device name. # xfs_growfs /home Logical volumes can be reduced in size as well as increased. However, it is very important to remember to reduce the size of the file system or whatever is residing in the volume before shrinking the volume itself, otherwise you risk losing data. ext2 If you are using ext2 as the file system then you can use the e2fsadm command mentioned earlier to take care of both the file system and volume resizing as follows: # umount /home # e2fsadm -L-1G /dev/myvg/homevol # mount /home If you prefer to do this manually you must know the new size of the volume in blocks and use the following commands: # umount /home # resize2fs /dev/myvg/homevol 524288 # lvreduce -L-1G /dev/myvg/homevol # mount /home reiserfs Reiserfs seems to prefer to be unmounted when shrinking # umount /home # resize_reiserfs -s-1G /dev/myvg/homevol # lvreduce -L-1G /dev/myvg/homevol # mount -treiserfs /dev/myvg/homevol /home xfs There is no way to shrink XFS file systems. To take a disk out of service it must first have all of its active physical extents moved to one or more of the remaining disks in the volume group. There must be enough free physical extents in the remaining PVs to hold the extents to be copied from the old disk. For further detail see . LVM allows you to create PVs (physical volumes) out of almost any block device so, for example, the following are all valid commands and will work quite happily in an LVM environment: # pvcreate /dev/sda1 # pvcreate /dev/sdf # pvcreate /dev/hda8 # pvcreate /dev/hda6 # pvcreate /dev/md1 In a normal production system it is recommended that only one PV exists on a single real disk, for the following reasons: Administrative convenience It's easier to keep track of the hardware in a system if each real disk only appears once. This becomes particularly true if a disk fails. To avoid striping performance problems LVM can't tell that two PVs are on the same physical disk, so if you create a striped LV then the stripes could be on different partitions on the same disk resulting in a decrease in performance rather than an increase. However it may be desirable to do this for some reasons: Migration of existing system to LVM On a system with few disks it may be necessary to move data around partitions to do the conversion (see ) Splitting one big disk between Volume Groups If you have a very large disk and want to have more than one volume group for administrative purposes then it is necessary to partition the drive into more than one area. If you do have a disk with more than one partition and both of those partitions are in the same volume group, take care to specify which partitions are to be included in a logical volume when creating striped volumes. The recommended method of partitioning a disk is to create a single partition that covers the whole disk. This avoids any nasty accidents with whole disk drive device nodes and prevents the kernel warning about unknown partition types at boot-up. You need to be especially careful on SPARC systems where the disks have Sun disk labels on them. The normal layout for a Sun disk label is for the first partition to start at block zero of the disk, thus the first partition also covers the area containing the disk label itself. This works fine for ext2 filesystems (and is essential for booting using SILO) but such partitions should not be used for LVM. This is because LVM starts writing at the very start of the device and will overwrite the disk label. If you want to use a disk with a Sun disklabel with LVM, make sure that the partition you are going to use starts at cylinder 1 or higher. This section details several different recipes for setting up lvm. The hope is that the reader will adapt these recipes to their own system and needs. For this recipe, the setup has three SCSI disks that will be put into a logical volume using LVM. The disks are at /dev/sda, /dev/sdb, and /dev/sdc. Before you can use a disk in a volume group you will have to prepare it: The following will destroy any data on /dev/sda, /dev/sdb, and /dev/sdc Run pvcreate on the disks # pvcreate /dev/sda # pvcreate /dev/sdb # pvcreate /dev/sdc This creates a volume group descriptor area (VGDA) at the start of the disks. Create a volume group # vgcreate my_volume_group /dev/sda /dev/sdb /dev/sdc/ Run vgdisplay to verify volume group # vgdisplay # vgdisplay --- Volume Group --- VG Name my_volume_group VG Access read/write VG Status available/resizable VG # 1 MAX LV 256 Cur LV 0 Open LV 0 MAX LV Size 255.99 GB Max PV 256 Cur PV 3 Act PV 3 VG Size 1.45 GB PE Size 4 MB Total PE 372 Alloc PE / Size 0 / 0 Free PE / Size 372/ 1.45 GB VG UUID nP2PY5-5TOS-hLx0-FDu0-2a6N-f37x-0BME0Y The most important things to verify are that the first three items are correct and that the VG Size item is the proper size for the amount of space in all four of your disks. If the volume group looks correct, it is time to create a logical volume on top of the volume group. You can make the logical volume any size you like. (It is similar to a partition on a non LVM setup.) For this example we will create just a single logical volume of size 1GB on the volume group. We will not use striping because it is not currently possible to add a disk to a stripe set after the logical volume is created. # lvcreate -L1G -nmy_logical_volume my_volume_group lvcreate -- doing automatic backup of "my_volume_group" lvcreate -- logical volume "/dev/my_volume_group/my_logical_volume" successfully created Create an ext2 file system on the logical volume # mke2fs /dev/my_volume_group/my_logical_volume mke2fs 1.19, 13-Jul-2000 for EXT2 FS 0.5b, 95/08/09 Filesystem label= OS type: Linux Block size=4096 (log=2) Fragment size=4096 (log=2) 131072 inodes, 262144 blocks 13107 blocks (5.00%) reserved for the super user First data block=0 9 block groups 32768 blocks per group, 32768 fragments per group 16384 inodes per group Superblock backups stored on blocks: 32768, 98304, 163840, 229376 Writing inode tables: done Writing superblocks and filesystem accounting information: done Mount the logical volume and check to make sure everything looks correct # mount /dev/my_volume_group/my_logical_volume /mnt # df Filesystem 1k-blocks Used Available Use% Mounted on /dev/hda1 1311552 628824 616104 51% / /dev/my_volume_group/my_logical_volume 1040132 20 987276 0% /mnt If everything worked properly, you should now have a logical volume with and ext2 file system mounted at /mnt. For this recipe, the setup has three SCSI disks that will be put into a logical volume using LVM. The disks are at /dev/sda, /dev/sdb, and /dev/sdc. It is not currently possible to add a disk to a striped logical volume. Do not use LV striping if you wish to be able to do so. Before you can use a disk in a volume group you will have to prepare it: The following will destroy any data on /dev/sda, /dev/sdb, and /dev/sdc Run pvcreate on the disks: # pvcreate /dev/sda # pvcreate /dev/sdb # pvcreate /dev/sdc This creates a volume group descriptor area (VGDA) at the start of the disks. Create a volume group # vgcreate my_volume_group /dev/sda /dev/sdb /dev/sdc Run vgdisplay to verify volume group # vgdisplay --- Volume Group --- VG Name my_volume_group VG Access read/write VG Status available/resizable VG # 1 MAX LV 256 Cur LV 0 Open LV 0 MAX LV Size 255.99 GB Max PV 256 Cur PV 3 Act PV 3 VG Size 1.45 GB PE Size 4 MB Total PE 372 Alloc PE / Size 0 / 0 Free PE / Size 372/ 1.45 GB VG UUID nP2PY5-5TOS-hLx0-FDu0-2a6N-f37x-0BME0Y The most important things to verify are that the first three items are correct and that the VG Size item is the proper size for the amount of space in all four of your disks. If the volume group looks correct, it is time to create a logical volume on top of the volume group. You can make the logical volume any size you like (up to the size of the VG you are creating it on; it is similar to a partition on a non LVM setup). For this example we will create just a single logical volume of size 1GB on the volume group. The logical volume will be a striped set using for the 4k stripe size. This should increase the performance of the logical volume. # lvcreate -i3 -I4 -L1G -nmy_logical_volume my_volume_group lvcreate -- rounding 1048576 KB to stripe boundary size 1056768 KB / 258 PE lvcreate -- doing automatic backup of "my_volume_group" lvcreate -- logical volume "/dev/my_volume_group/my_logical_volume" successfully created If you create the logical volume with a '-i2' you will only use two of the disks in your volume group. This is useful if you want to create two logical volumes out of the same physical volume, but we will not touch that in this recipe. Create an ext2 file system on the logical volume # mke2fs /dev/my_volume_group/my_logical_volume mke2fs 1.19, 13-Jul-2000 for EXT2 FS 0.5b, 95/08/09 Filesystem label= OS type: Linux Block size=4096 (log=2) Fragment size=4096 (log=2) 132192 inodes, 264192 blocks 13209 blocks (5.00%) reserved for the super user First data block=0 9 block groups 32768 blocks per group, 32768 fragments per group 14688 inodes per group Superblock backups stored on blocks: 32768, 98304, 163840, 229376 Writing inode tables: done Writing superblocks and filesystem accounting information: done Mount the file system on the logical volume # mount /dev/my_volume_group/my_logical_volume /mnt and check to make sure everything looks correct # df Filesystem 1k-blocks Used Available Use% Mounted on /dev/hda1 1311552 628824 616104 51% / /dev/my_volume_group/my_logical_volume 1040132 20 987276 0% /mnt If everything worked properly, you should now have a logical volume mounted at /mnt. A data centre machine has 6 disks attached as follows: # pvscan pvscan -- ACTIVE PV "/dev/sda" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sdb" of VG "sales" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sdc" of VG "ops" [1.95 GB / 44 MB free] pvscan -- ACTIVE PV "/dev/sdd" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sde1" of VG "ops" [996 MB / 52 MB free] pvscan -- ACTIVE PV "/dev/sde2" of VG "sales" [996 MB / 944 MB free] pvscan -- ACTIVE PV "/dev/sdf1" of VG "ops" [996 MB / 0 free] pvscan -- ACTIVE PV "/dev/sdf2" of VG "dev" [996 MB / 72 MB free] pvscan -- total: 8 [11.72 GB] / in use: 8 [11.72 GB] / in no VG: 0 [0] # df Filesystem 1k-blocks Used Available Use% Mounted on /dev/dev/cvs 1342492 516468 757828 41% /mnt/dev/cvs /dev/dev/users 2064208 2060036 4172 100% /mnt/dev/users /dev/dev/build 1548144 1023041 525103 66% /mnt/dev/build /dev/ops/databases 2890692 2302417 588275 79% /mnt/ops/databases /dev/sales/users 2064208 871214 1192994 42% /mnt/sales/users /dev/ops/batch 1032088 897122 134966 86% /mnt/ops/batch As you can see the "dev" and "ops" groups are getting full so a new disk is purchased and added to the system. It becomes /dev/sdg. The new disk is to be shared equally between ops and dev so it is partitioned into two physical volumes /dev/sdg1 and /dev/sdg2 : # fdisk /dev/sdg Device contains neither a valid DOS partition table, nor Sun or SGI disklabel Building a new DOS disklabel. Changes will remain in memory only, until you decide to write them. After that, of course, the previous content won't be recoverable. Command (m for help): n Command action e extended p primary partition (1-4) p Partition number (1-4): 1 First cylinder (1-1000, default 1): Using default value 1 Last cylinder or +size or +sizeM or +sizeK (1-1000, default 1000): 500 Command (m for help): n Command action e extended p primary partition (1-4) p Partition number (1-4): 2 First cylinder (501-1000, default 501): Using default value 501 Last cylinder or +size or +sizeM or +sizeK (501-1000, default 1000): Using default value 1000 Command (m for help): t Partition number (1-4): 1 Hex code (type L to list codes): 8e Changed system type of partition 1 to 8e (Unknown) Command (m for help): t Partition number (1-4): 2 Hex code (type L to list codes): 8e Changed system type of partition 2 to 8e (Unknown) Command (m for help): w The partition table has been altered! Calling ioctl() to re-read partition table. WARNING: If you have created or modified any DOS 6.x partitions, please see the fdisk manual page for additional information. Next physical volumes are created on this partition: # pvcreate /dev/sdg1 pvcreate -- physical volume "/dev/sdg1" successfully created # pvcreate /dev/sdg2 pvcreate -- physical volume "/dev/sdg2" successfully created The volumes are then added to the dev and ops volume groups: # vgextend ops /dev/sdg1 vgextend -- INFO: maximum logical volume size is 255.99 Gigabyte vgextend -- doing automatic backup of volume group "ops" vgextend -- volume group "ops" successfully extended # vgextend dev /dev/sdg2 vgextend -- INFO: maximum logical volume size is 255.99 Gigabyte vgextend -- doing automatic backup of volume group "dev" vgextend -- volume group "dev" successfully extended # pvscan pvscan -- reading all physical volumes (this may take a while...) pvscan -- ACTIVE PV "/dev/sda" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sdb" of VG "sales" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sdc" of VG "ops" [1.95 GB / 44 MB free] pvscan -- ACTIVE PV "/dev/sdd" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sde1" of VG "ops" [996 MB / 52 MB free] pvscan -- ACTIVE PV "/dev/sde2" of VG "sales" [996 MB / 944 MB free] pvscan -- ACTIVE PV "/dev/sdf1" of VG "ops" [996 MB / 0 free] pvscan -- ACTIVE PV "/dev/sdf2" of VG "dev" [996 MB / 72 MB free] pvscan -- ACTIVE PV "/dev/sdg1" of VG "ops" [996 MB / 996 MB free] pvscan -- ACTIVE PV "/dev/sdg2" of VG "dev" [996 MB / 996 MB free] pvscan -- total: 10 [13.67 GB] / in use: 10 [13.67 GB] / in no VG: 0 [0] The next thing to do is to extend the file systems so that the users can make use of the extra space. There are tools to allow online-resizing of ext2 file systems but here we take the safe route and unmount the two file systems before resizing them: # umount /mnt/ops/batch # umount /mnt/dev/users We then use the e2fsadm command to resize the logical volume and the ext2 file system on one operation. We are using ext2resize instead of resize2fs (which is the default command for e2fsadm) so we define the environment variable E2FSADM_RESIZE_CMD to tell e2fsadm to use that command. # export E2FSADM_RESIZE_CMD=ext2resize # e2fsadm /dev/ops/batch -L+500M e2fsck 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09 Pass 1: Checking inodes, blocks, and sizes Pass 2: Checking directory structure Pass 3: Checking directory connectivity Pass 4: Checking reference counts Pass 5: Checking group summary information /dev/ops/batch: 11/131072 files (0.0ent!-- non-contiguous), 4127/262144 blocks lvextend -- extending logical volume "/dev/ops/batch" to 1.49 GB lvextend -- doing automatic backup of volume group "ops" lvextend -- logical volume "/dev/ops/batch" successfully extended ext2resize v1.1.15 - 2000/08/08 for EXT2FS 0.5b e2fsadm -- ext2fs in logical volume "/dev/ops/batch" successfully extended to 1.49 GB # e2fsadm /dev/dev/users -L+900M e2fsck 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09 Pass 1: Checking inodes, blocks, and sizes Pass 2: Checking directory structure Pass 3: Checking directory connectivity Pass 4: Checking reference counts Pass 5: Checking group summary information /dev/dev/users: 12/262144 files (0.0% non-contiguous), 275245/524288 blocks lvextend -- extending logical volume "/dev/dev/users" to 2.88 GB lvextend -- doing automatic backup of volume group "dev" lvextend -- logical volume "/dev/dev/users" successfully extended ext2resize v1.1.15 - 2000/08/08 for EXT2FS 0.5b e2fsadm -- ext2fs in logical volume "/dev/dev/users" successfully extended to 2.88 GB We can now remount the file systems and see that the is plenty of space. # mount /dev/ops/batch # mount /dev/dev/users # df Filesystem 1k-blocks Used Available Use% Mounted on /dev/dev/cvs 1342492 516468 757828 41% /mnt/dev/cvs /dev/dev/users 2969360 2060036 909324 69% /mnt/dev/users /dev/dev/build 1548144 1023041 525103 66% /mnt/dev/build /dev/ops/databases 2890692 2302417 588275 79% /mnt/ops/databases /dev/sales/users 2064208 871214 1192994 42% /mnt/sales/users /dev/ops/batch 1535856 897122 638734 58% /mnt/ops/batch Following on from the previous example we now want to use the extra space in the "ops" volume group to make a database backup every evening. To ensure that the data that goes onto the tape is consistent we use an LVM snapshot logical volume. This type of volume is a read-only copy of another volume that contains all the data that was in the volume at the time the snapshot was created. This means we can back up that volume without having to worry about data being changed while the backup is going on, and we don't have to take the database volume offline while the backup is taking place. There is a little over 500 Megabytes of free space in the "ops" volume group, so we will use all of it to allocate space for the snapshot logical volume. A snapshot volume can be as large or a small as you like but it must be large enough to hold all the changes that are likely to happen to the original volume during the lifetime of the snapshot. So here, allowing 500 megabytes of changes to the database volume which should be plenty. # lvcreate -L592M -s -n dbbackup /dev/ops/databases lvcreate -- WARNING: the snapshot must be disabled if it gets full lvcreate -- INFO: using default snapshot chunk size of 64 KB for "/dev/ops/dbbackup" lvcreate -- doing automatic backup of "ops" lvcreate -- logical volume "/dev/ops/dbbackup" successfully created If the snapshot is of an XFS filesystem, the xfs_freeze command should be used to quiesce the filesystem before creating the snapshot. (if the filesystem is mounted) # xfs_freeze -f /mnt/point; lvcreate -L592M -s -n dbbackup /dev/ops/databases; xfs_freeze -u /mnt/point If the snapshot logical volume becomes full it will become unusable so it is vitally important to allocate enough space. We can now create a mount-point and mount the volume # mkdir /mnt/ops/dbbackup # mount /dev/ops/dbbackup /mnt/ops/dbbackup mount: block device /dev/ops/dbbackup is write-protected, mounting read-only If you are using XFS as the filesystem you will need to add the nouuid option to the mount command: # mount /dev/ops/dbbackup /mnt/ops/dbbackup -onouuid,ro Previously, the norecovery option was suggested to allow the mounting of XFS snapshots. It has been recommended not to use this option, but to instead use xfs_freeze to quiesce the filesystem before creating the snapshot. I assume you will have a more sophisticated backup strategy than this! # tar -cf /dev/rmt0 /mnt/ops/dbbackup tar: Removing leading `/' from member names When the backup has finished you can now unmount the volume and remove it from the system. You should remove snapshot volume when you have finished with them because they take a copy of all data written to the original volume and this can hurt performance. # umount /mnt/ops/dbbackup # lvremove /dev/ops/dbbackup lvremove -- do you really want to remove "/dev/ops/dbbackup"? [y/n]: y lvremove -- doing automatic backup of volume group "ops" lvremove -- logical volume "/dev/ops/dbbackup" successfully removed Say you have an old IDE drive on /dev/hdb. You want to remove that old disk but a lot of files are on it. You should always backup your system before attempting a pvmove operation. If you have enough free extents on the other disks in the volume group, you have it easy. Simply run # pvmove /dev/hdb pvmove -- moving physical extents in active volume group "dev" pvmove -- WARNING: moving of active logical volumes may cause data loss! pvmove -- do you want to continue? [y/n] y pvmove -- 249 extents of physical volume "/dev/hdb" successfully moved This will move the allocated physical extents from /dev/hdb onto the rest of the disks in the volume group. Be aware that pvmove is quite slow as it has to copy the contents of a disk block by block to one or more disks. If you want more steady status reports from pvmove, use the -v flag. We can now remove the old IDE disk from the volume group. # vgreduce dev /dev/hdb vgreduce -- doing automatic backup of volume group "dev" vgreduce -- volume group "dev" successfully reduced by physical volume: vgreduce -- /dev/hdb The drive can now be either physically removed when the machine is next powered down or reallocated to other users. If you do not have enough free physical extents to distribute the old physical extents to, you will have to add a disk to the volume group and move the extents to it. First, you need to pvcreate the new disk to make it available to LVM. In this recipe we show that you don't need to partition a disk to be able to use it. # pvcreate /dev/sdf pvcreate -- physical volume "/dev/sdf" successfully created As developers use a lot of disk space this is a good volume group to add it into. # vgextend dev /dev/sdf vgextend -- INFO: maximum logical volume size is 255.99 Gigabyte vgextend -- doing automatic backup of volume group "dev" vgextend -- volume group "dev" successfully extended Next we move the data from the old disk onto the new one. Note that it is not necessary to unmount the file system before doing this. Although it is *highly* recommended that you do a full backup before attempting this operation in case of a power outage or some other problem that may interrupt it. The pvmove command can take a considerable amount of time to complete and it also exacts a performance hit on the two volumes so, although it isn't necessary, it is advisable to do this when the volumes are not too busy. # pvmove /dev/hdb /dev/sdf pvmove -- moving physical extents in active volume group "dev" pvmove -- WARNING: moving of active logical volumes may cause data loss! pvmove -- do you want to continue? [y/n] y pvmove -- 249 extents of physical volume "/dev/hdb" successfully moved We can now remove the old IDE disk from the volume group. # vgreduce dev /dev/hdb vgreduce -- doing automatic backup of volume group "dev" vgreduce -- volume group "dev" successfully reduced by physical volume: vgreduce -- /dev/hdb The drive can now be either physically removed when the machine is next powered down or reallocated to some other users. It is quite easy to move a whole volume group to another system if, for example, a user department acquires a new server. To do this we use the vgexport and vgimport commands. First, make sure that no users are accessing files on the active volume, then unmount it # unmount /mnt/design/users Marking the volume group inactive removes it from the kernel and prevents any further activity on it. # vgchange -an design vgchange -- volume group "design" successfully deactivated It is now necessary to export the volume group. This prevents it from being accessed on the ``old'' host system and prepares it to be removed. # vgexport design vgexport -- volume group "design" sucessfully exported When the machine is next shut down, the disk can be unplugged and then connected to it's new machine When plugged into the new system it becomes /dev/sdb so an initial pvscan shows: # pvscan pvscan -- reading all physical volumes (this may take a while...) pvscan -- inactive PV "/dev/sdb1" is in EXPORTED VG "design" [996 MB / 996 MB free] pvscan -- inactive PV "/dev/sdb2" is in EXPORTED VG "design" [996 MB / 244 MB free] pvscan -- total: 2 [1.95 GB] / in use: 2 [1.95 GB] / in no VG: 0 [0] We can now import the volume group (which also activates it) and mount the file system. # vgimport design /dev/sdb1 /dev/sdb2 vgimport -- doing automatic backup of volume group "design" vgimport -- volume group "design" successfully imported and activated # mkdir -p /mnt/design/users # mount /dev/design/users /mnt/design/users The file system is now available for use. There is a new group of users "design" to add to the system. One way of dealing with this is to create a new volume group to hold their data. There are no new disks but there is plenty of free space on the existing disks that can be reallocated. # pvscan pvscan -- reading all physical volumes (this may take a while...) pvscan -- ACTIVE PV "/dev/sda" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sdb" of VG "sales" [1.95 GB / 1.27 GB free] pvscan -- ACTIVE PV "/dev/sdc" of VG "ops" [1.95 GB / 564 MB free] pvscan -- ACTIVE PV "/dev/sdd" of VG "dev" [1.95 GB / 0 free] pvscan -- ACTIVE PV "/dev/sde" of VG "ops" [1.95 GB / 1.9 GB free] pvscan -- ACTIVE PV "/dev/sdf" of VG "dev" [1.95 GB / 1.33 GB free] pvscan -- ACTIVE PV "/dev/sdg1" of VG "ops" [996 MB / 432 MB free] pvscan -- ACTIVE PV "/dev/sdg2" of VG "dev" [996 MB / 632 MB free] pvscan -- total: 8 [13.67 GB] / in use: 8 [13.67 GB] / in no VG: 0 [0] We decide to reallocate /dev/sdg1 and /dev/sdg2 to design so first we have to move the physical extents into the free areas of the other volumes (in this case /dev/sdf for volume group dev and /dev/sde for volume group ops). Some space is still used on the chosen volumes so it is necessary to move that used space off onto some others. Move all the used physical extents from /dev/sdg1 to /dev/sde and from /dev/sdg2 to /dev/sde # pvmove /dev/sdg1 /dev/sde pvmove -- moving physical extents in active volume group "ops" pvmove -- WARNING: moving of active logical volumes may cause data loss! pvmove -- do you want to continue? [y/n] y pvmove -- doing automatic backup of volume group "ops" pvmove -- 141 extents of physical volume "/dev/sdg1" successfully moved # pvmove /dev/sdg2 /dev/sdf pvmove -- moving physical extents in active volume group "dev" pvmove -- WARNING: moving of active logical volumes may cause data loss! pvmove -- do you want to continue? [y/n] y pvmove -- doing automatic backup of volume group "dev" pvmove -- 91 extents of physical volume "/dev/sdg2" successfully moved Now, split /dev/sdg2 from dev and add it into a new group called "design". it is possible to do this using vgreduce and vgcreate but the vgsplit command combines the two. # vgsplit dev design /dev/sdg2 vgsplit -- doing automatic backup of volume group "dev" vgsplit -- doing automatic backup of volume group "design" vgsplit -- volume group "dev" successfully split into "dev" and "design" Next, remove /dev/sdg1 from ops and add it into design. # vgreduce ops /dev/sdg1 vgreduce -- doing automatic backup of volume group "ops" vgreduce -- volume group "ops" successfully reduced by physical volume: vgreduce -- /dev/sdg1 # vgextend design /dev/sdg1 vgextend -- INFO: maximum logical volume size is 255.99 Gigabyte vgextend -- doing automatic backup of volume group "design" vgextend -- volume group "design" successfully extended Now create a logical volume. Rather than allocate all of the available space, leave some spare in case it is needed elsewhere. # lvcreate -L750M -n users design lvcreate -- rounding up size to physical extent boundary "752 MB" lvcreate -- doing automatic backup of "design" lvcreate -- logical volume "/dev/design/users" successfully created # mke2fs /dev/design/users mke2fs 1.18, 11-Nov-1999 for EXT2 FS 0.5b, 95/08/09 Filesystem label= OS type: Linux Block size=4096 (log=2) Fragment size=4096 (log=2) 96384 inodes, 192512 blocks 9625 blocks (5.00ent!-- ) reserved for the super user First data block=0 6 block groups 32768 blocks per group, 32768 fragments per group 16064 inodes per group Superblock backups stored on blocks: 32768, 98304, 163840 Writing inode tables: done Writing superblocks and filesystem accounting information: done # mkdir -p /mnt/design/users mount /dev/design/users /mnt/design/users/ It's also a good idea to add an entry for this file system in your /etc/fstab file as follows: /dev/design/user /mnt/design/users ext2 defaults 1 2 It is strongly recommended that you take a full backup of your system before attempting to convert to root on LVM. Having your root filesystem on LVM can significantly complicate upgrade procedures (depending on your distribution) so it should not be attempted lightly. Particularly, you must consider how you will insure that the LVM kernel module (if you do not have LVM compiled into the kernel) as well as the vgscan/vgchange tools are available before, during, and after the upgrade. Having your root filesystem on LVM can significantly complicate recovery of damaged filesystems. If you lose your initrd, it will be very difficult to boot your system. You will need to have a recover disk that contains the kernel, LVM module, and LVM tools, as well as any tools necessary to recover a damaged filesystem. Be sure to make regular backups and have an up-to-date alternative boot method that allows for recovery of LVM. In this example the whole system was installed in a single root partition with the exception of /boot. The system had a 2 gig disk partitioned as: /dev/hda1 /boot /dev/hda2 swap /dev/hda3 / The / partition covered all of the disk not used by /boot and swap. An important prerequisite of this procedure is that the root partition is less that half full (so that a copy of it can be created in a logical volume). If this is not the case then a second disk drive should be used. The procedure in that case is similar but there is no need to shrink the existing root partition and /dev/hda4 should be replaced with (eg) /dev/hdb1 in the examples. To do this it is easiest to use GNU parted. This software allows you to grow and shrink partitions that contain filesystems. It is possible to use resize2fs and fdisk to do this but GNU parted makes it much less prone to error. It may be included in your distribution, if not you can download it from ftp://ftp.gnu.org/pub/gnu/parted. Once you have parted on your system AND YOU HAVE BACKED THE SYSTEM UP: Boot into single user mode (type linux S at the LILO prompt) This is important. Booting single-user ensures that the root filesystem is mounted read-only and no programs are accessing the disk. Run parted to shrink the root partition Do this so there is room on the disk for a complete copy of it in a logical volume. In this example a 1.8 gig partition is shrunk to 1 gigabyte This displays the sizes and names of the partitions on the disk # parted /dev/hda (parted) p . . . Now resize the partition: (parted) resize 3 145 999 The first number here the partition number (hda3), the second is the same starting position that hda3 currently has. Do not change this. The last number should make the partition around half the size it currently is. Create a new partition (parted) mkpart primary ext2 1000 1999 This makes a new partition to hold the initial LVM data. It should start just beyond the newly shrunk hda3 and finish at the end of the disk. Quit parted (parted) q Reboot the system Make sure that the kernel you are currently running works with LVM and has CONFIG_BLK_DEV_RAM and CONFIG_BLK_DEV_INITRD set in the config file. Change the partition type on the newly created partition from Linux to LVM (8e). Parted doesn't understand LVM partitions so this has to be done using fdisk. # fdisk /dev/hda Command (m for help): t Partition number (1-4): 4 Hex code (type L to list codes): 8e Changed system type of partition 4 to 8e (Unknown) Command (m for help): w Initialize LVM (vgscan) # vgscan Make the new partition into a PV # pvcreate /dev/hda4 create a new volume group # vgcreate vg /dev/hda4 Create a logical volume to hold the new root. # lvcreate -L250M -n root vg Make a filesystem in the logical volume and copy the root files onto it. # mke2fs /dev/vg/root # mount /dev/vg/root /mnt/ # find / -xdev | cpio -pvmd /mnt Edit /mnt/etc/fstab on the new root so that / is mounted on /dev/vg/root. For example: /dev/hda3 / ext2 defaults 1 1 becomes: /dev/vg/root / ext2 defaults 1 1 # lvmcreate_initrd Make sure you note the name that lvmcreate_initrd calls the initrd image. It should be in /boot. Add an entry in /etc/lilo.conf for LVM. This should look similar to the following: image = /boot/KERNEL_IMAGE_NAME label = lvm root = /dev/vg/root initrd = /boot/INITRD_IMAGE_NAME ramdisk = 8192 Where KERNEL_IMAGE_NAME is the name of your LVM enabled kernel, and INITRD_IMAGE_NAME is the name of the initrd image created by lvmcreate_initrd. The ramdisk line may need to be increased if you have a large LVM configuration, but 8192 should suffice for most users. The default ramdisk size is 4096. If in doubt check the output from the lvmcreate_initrd command, the line that says: lvmcreate_initrd -- making loopback file (6189 kB) and make the ramdisk the size given in brackets. You should copy this new lilo.conf onto /etc in the new root fs as well. # cp /etc/lilo.conf /mnt/etc/ # lilo Reboot - at the LILO prompt type "lvm" The system should reboot into Linux using the newly created Logical Volume. If that worked then you should make lvm the default LILO boot destination by adding the line default=lvm in the first section of /etc/lilo.conf If it did not work then reboot normally and try to diagnose the problem. It could be a typing error in lilo.conf or LVM not being available in the initial RAM disk or its kernel. Examine the message produced at boot time carefully. Add the rest of the disk into LVM When you are happy with this setup you can then add the old root partition to LVM and spread out over the disk. First set the partition type to 8e(LVM) # fdisk /dev/hda Command (m for help): t Partition number (1-4): 3 Hex code (type L to list codes): 8e Changed system type of partition 3 to 8e (Unknown) Command (m for help): w Convert it into a PV and add it to the volume group: # pvcreate /dev/hda3 # vgextend vg /dev/hda3 Don't do this unless you're really sure of what you're doing. You'll probably lose all your data. If you've upgraded LVM from previous versions to early 0.9 and 0.9.1 versions of LVM and vgscan says vgscan -- no volume groups found, this is one way to fix it. Download the UUID fixer program from the contributor directory at Sistina. It is located at ftp://ftp.sistina.com/pub/LVM/contrib/uuid_fixer-0.3-IOP10.tar.gz" Extract uuid_fixer-0.3-IOP10.tar.gz # tar zxf uuid_fixer-0.3-IOP10.tar.gz cd to uuid_fixer # cd uuid_fixer You have one of two options at this point: Use the prebuild binary (it is build for i386 architecture). Make sure you list all the PVs in the VG you are restoring, and follow the prompts # ./uuid_fixer entLIST OF ALL PVS IN VG TO BE RESTOREDent Build the uuid_builder program from source Edit the Makefile with your favorite editor, and make sure LVMDIR points to your LVM source. Then run make. # make Now run uuid_fixer. Make sure you list all the PVs in the VG you are restoring, and follow the prompts. # ./uuid_fixer entLIST OF ALL PVS IN VG TO BE RESTOREDent Deactivate any active Volume Groups (optional) # vgchange -an Run vgscan # vgscan Reactivate Volume Groups # vgchange -ay Be very careful doing this, LVM is not currently cluster-aware and it is very easy to lose all your data. If you have a fibre-channel or shared-SCSI environment where more than one machine has physical access to a set of disks then you can use LVM to divide these disks up into logical volumes. If you want to share data you should really be looking at GFS or other cluster filesystems. The key thing to remember when sharing volumes is that all the LVM administration must be done on one node only and that all other nodes must have LVM shut down before changing anything on the admin node. Then, when the changes have been made, it is necessary to run vgscan on the other nodes before reloading the volume groups. Also, unless you are running a cluster-aware filesystem (such as GFS) or application on the volume, only one node can mount each filesystem. It is up to you, as system administrator to enforce this, LVM will not stop you corrupting your data. The startup sequence of each node is the same as for a single-node setup with vgscan vgchange -ay in the startup scripts. If you need to do any changes to the LVM metadata (regardless of whether it affects volumes mounted on other nodes) you must go through the following sequence. In the steps below ``admin node'' is any arbirarily chosen node in the cluster. Admin node Other nodes ---------- ----------- Close all Logical volumes (umount) vgchange -an entmake changes, eg lvextendent vgscan vgchange -ay You do not need to, nor should you, unload the VGs on the admin node, so this can be the node with the highest uptime requirement. I'll say it again: Be very careful doing this Just telling us that LVM did not work does not provide us with enough information to help you. We need to know about your setup and the various components of your configuration. The first thing you should do is check the linux-lvm mailing list archives to see if someone else has already reported the same bug. If you do not find a bug report for a problem similar to yours you should collect as much of the following information as possible. The list is grouped into three categories of errors. For compilation errors: Detail the specific version of LVM you have. If you extracted LVM from a tarball give the name of the tar file and list any patches you applied. If you acquired LVM from the Public CVS server, give the date and time you checked it out. Provide the exact error message. Copy a couple of lines of output before the actual error message, as well as, a couple of lines after. These lines occasionally give hints as to why the error occurred. List the steps, in order, that produced the error. Is the error reproducible? If you start from a clean state does the same sequence of steps reproduce the error? For LVM errors: Include all of the information requested in the compilation section. Attach a short description of your hardware: types of machines and disks, disks interface (SCSI, FC, NBD). Any other tidbits about your hardware you feel is important. Include the output from pinfo -s The command line used to make LVM and the file system on top of it. The command line used to mount the file system. When LVM trips a panic trap: Include all of the information requested in two sections above. Provide the debug dump for the machine. This is best accomplished if you are watching the console output of the computer over a serial link, since you can't very well copy and paste from a panic'd machine, and it is very easy to mistype something if you try to copy the output by hand. This can be a lot of information. If you end up with more than a couple of files, tar and gzip them into a single archive. Submit this compressed archive file to lvm-devel along with a short description of the error. Before you post to any of our lists please read the all of this document and check the archives to see if your question has already been answered. Please post in text only to our lists, fancy formated messages are near impossible to read if someone else is not running a mail client that understands it. Standard mailing list etiquette applies. Incomplete questions or configuration data make it very hard for us to answer your questions. Subscription to all lists is accomplished through a web interface. linux-lvm This list is aimed at user-related questions and comments. You may be able to get the answers you need from other people who have the same issues. Open discussion is encouraged. Bug reports should be sent to this list, although technical discussion regarding the bug's fix may be moved to the lvm-devel list. lvm-devel This is the development list for LVM. It is intended to be an open discussion on bugs, desired features, and questions about the internals of LVM. Feel free to post anything relevant to LVM or logical volume managers in general. We wish this to be a fairly high volume list. lvm-commit This list gets messages automatically whenever someone commits to the cvs tree. Its main purpose is to keep up with the cvs tree. lvm-bugs This list is rarely used anymore. Bugs should be sent to the linux-lvm list. LVM Links: The Logical Volume Manager home page. The LVM ftp site.