This document explains how to transfer, or migrate, an entire Linux system, including LILO, from one hard disk to another.
In the following explanation, /dev/hda (first IDE hard disk) means the old disk, and /dev/hdb (second IDE hard disk) means the new disk.
Specific partitions on the old disk are referred to as /dev/hda1, /dev/hda2, and so on. Specific partitions on the new disk are referred to as /dev/hdb1, /dev/hdb2, and so on.
The explanations in this document are based on Red Hat 6.0. They have also been tested with Debian 2.1, Slackware 3.5 and SuSE 6.2; we indicate a few differences to note if you're using those distributions.
If the commands don't work properly on your system, please let us know, mentioning what distribution you're using.
Modern systems can accept four EIDE devices on the hard disk controller, so there shouldn't be any problem installing both disks on your system at the same time, even if you also have other EIDE devices. Hard disks and CD-ROM drives are typical EIDE devices. Floppy drives and tape drives are usually connected to the floppy drive controller rather than to the hard disk controller.
SCSI adapters are even more flexible and can accept seven devices. If you're lucky (and rich) enough to have a SCSI adapter, you probably already know this, and you probably know which of your devices are SCSI devices! For more information, see the SCSI How-To.
Even the oldest systems can accept two devices on the hard disk controller, so you can still install both hard disks at the same time. However, if you already have another device installed in addition to your hard disk (for example, if you have both a hard disk and a CD-ROM drive), you'll have to remove the other device to be able to install the old hard disk and the new hard disk at the same time.
You must configure the disks as master or slave by installing the disks' jumpers as appropriate. You'll often find configuration information on the disks themselves; if not, consult the manuals or the disks' manufacturers.
You must also inform the BIOS of the disks' presence and of their geometry. Usually, you enter the BIOS setup program by pressing a key during the system boot-up. Here's what to do for some common BIOSes:
F2 key during Power-On Self-Test (POST)
Del key during Power-On Self-Test
Del, or Ctrl-Alt-Esc
F10 key after the square appears in the top right corner of the screen during boot-up
Ctrl-Alt-Enter
Esc key during Power-On Self-Test
F1 key during HP blue splash screen
F1 while the square with the wavy lines is displayed in the upper right corner during power-on [1]
Ctrl-Alt-Del, then Ctrl-Alt-Ins when the cursor is in the top right corner
Ctrl-Alt-S during Power-On Self-Test
For some models, F1 or F2 key during Power-On Self-Test
Ctrl-Alt-Esc, Ctrl-Alt-S, or Ctrl-Alt-Enter
Many older systems require an Installation or Reference Disk.
Reboot the system and login as root. If you use the su command to become the user root, use su -, with the hyphen option.
Some people like to mount partitions from other operating systems (DOS, Windows, OS/2, etc.) so they can use them under Linux. These partitions must be created and copied under their own operating system, and you should unmount them before copying your Linux partition. For example, if you have a DOS partition mounted at /dos, you must unmount it with this command:
umount /dos |
Note that the command is umount, without the first letter n in the word unmount.
You should also unmount network drives.
Use this command to partition the new disk:
fdisk /dev/hdb |
EIDE devices are identified as hda, hdb, hdc, and hdd in the /dev directory. Partitions on these disks can range from 1 to 16 and are also in the /dev directory. For example, /dev/hda4 refers to partition 4 on hard disk a (first EIDE hard disk).
SCSI devices are listed as devices sda, sdb, sdc, sdd, sde, sdf, and sdg in the /dev directory. Similarly, partitions on these disks can range from 1 to 16 and are also in the /dev directory. For example, /dev/sda3 refers to partition 3 on SCSI disk a (first SCSI hard disk).
For Linux partitions with the ext2 file system, use system ID 83. For swap partitions, use system ID 82.
For more information on partitioning, see the Installation How-To and the Partition Mini How-To.
If your new disk has over 1024 cylinders, see the Large Disk Mini How-To. In brief, you should install all files required to boot Linux within the first 1024 cylinders. One way to do this is to create a small partition (5 Mb or so) just for the /boot directory at the beginning of the disk. (Slackware only: The kernel is at /vmlinuz rather than /boot/vmlinuz, so you should put both the / directory and the /boot directory in this partition.)
Partitions for systems other than Linux should be created using their own fdisk or equivalent command rather than with Linux's fdisk.
Use the following command to format Linux partitions using ext2fs on the new disk:
mkfs.ext2 /dev/hdb1 |
To check the disk for bad blocks (physical defects), add the
-c
option just before
/dev/hdb1.
If the new disk will have more than one Linux partition, format the
other partitions with mkfs.ext2 /dev/hdb2,
mkfs.ext2 /dev/hdb3, and so on. Add the
-c
option if desired.
With older distributions, the command mkfs -t ext2 -c /dev/hdb1 didn't check for bad blocks under any of Red Hat, Debian or Slackware, contrary to what the man page stated. This has now been fixed. |
To format a swap partition, use this command:
mkswap /dev/hdb1 |
Again, you can add the -c
option before
/dev/hdb1 to check for bad blocks.
Create a directory where you'll mount the new disk, for example /new-disk, and mount it there:
mkdir /new-disk mount -t ext2 /dev/hdb1 /new-disk |
If the new disk will have more than one Linux partition, mount them all under /new-disk with the same organization they'll have later.
Example. The new disk will have four Linux partitions, as follows:
/dev/hdb1: / /dev/hdb2: /home /dev/hdb3: /var /dev/hdb4: /var/spool |
Mount the four partitions under /new-disk as follows:
/dev/hdb1: /new-disk /dev/hdb2: /new-disk/home /dev/hdb3: /new-disk/var /dev/hdb4: /new-disk/var/spool |
You must create the mount points for each level before you mount the partitions at that level.
Example.
mkdir /new-disk [1st level] mount -t ext2 /dev/hdb1 /new-disk mkdir /new-disk/home [2nd level] mount -t ext2 /dev/hdb2 /new-disk/home mkdir /new-disk/var [2nd level also] mount -t ext2 /dev/hdb3 /new-disk/var mkdir /new-disk/var/spool [3rd level] mount -t ext2 /dev/hdb4 /new-disk/var/spool |
If you've created a mount point at /new-disk/tmp, you'll need to correct the directory's permissions to let all users access it:
chmod 1777 /new-disk/tmp |
You might want to go to single-user mode before starting to copy the disk, in order to shut down the system daemons and preserve the state of the logs, and to prevent users from logging in:
/sbin/telinit 1 |
When copying the hard disk, you want to copy all directories and files, including links.
However, you don't want to copy the directory /new-disk, since this would copy the new disk to itself!
Furthermore, you want to create the /proc directory on the new disk, but you don't want to copy its contents: /proc is a virtual file system and doesn't have any actual files, but rather contains information on the processes running on the system.
Here are three different ways to copy the old disk to the new one. This may take quite a while, especially if you have a large disk or little memory. You can expect to be able to copy 10 Mb per minute, and possibly much more.
You can follow the copy's progress by using the command df from another terminal. Try watch df or watch ls -l /new-disk to see a report updated every two seconds; press Ctrl-C to end the display. Be aware that running the watch program itself will slow down the copying.
This is the simplest method, but will only work if your original Linux system is on a single disk partition.
The -a
option preserves the original system as
much as possible. The -x
option limits cp to a
single file system; this is necessary to avoid copying the
/new-disk
and
/proc directories.
SuSE only. With this method only, you must also create the directory /dev/pts on the new disk. Use the command mkdir /new-disk/dev/pts".
When using the |
(write this all on one line)
This goes to the root directory and then copies all files and directories except /new-disk and /proc to /new-disk. Note that the first option after ls is the number 1, not the letter L!
This command should work in all circumstances.
(write this all on one line)
The last directory, /new-disk, is the destination for the cp command. All the other directories are the sources. Therefore, we're copying all the directories we're listing to /new-disk.
With this method, you simply list yourself the directories you want to copy. Here we listed all the directories except /new-disk and /proc. If you can't use the other methods for any reason, you can always use this command to manually specify the directories you want to copy.
With this method only, if there are any files in the root directory itself, you need another command to copy them. In particular, this is required with Debian and Slackware, since these distributions put files in the root directory:
cp -dp /* /.* /new-disk |
Previous versions of the Mini How-To stated that you could also use tar to copy the disk, but this method was found to have a bug. There are of course many other ways to copy the disks, but these three are the simplest, quickest, and most reliable.
After using any of these three methods, you must also create the /proc directory on the new disk, if it doesn't already exist:
mkdir /new-disk/proc |
At this point, you may verify the file structure on the new disk, if you wish:
umount /new-disk fsck.ext2 -f /dev/hdb1 mount -t ext2 /dev/hdb1 /new-disk |
If the new disk has more than one partition, you must unmount them from the bottom up before running fsck.ext2: in the example mentioned above, you'd first unmount the 3rd level partitions, then the 2nd level partitions, and then the 1st level partition.
You may also compare the two disks, to ensure that the files were copied properly:
find / -path /proc -prune -o -path /new-disk -prune -o -xtype f -exec cmp {} /new-disk{} \; |
(write this all on one line)
Slackware only. A basic Slackware installation ("A" series only) doesn't include the cmp command, so you won't be able to run this command if you have only installed the basic files. The cmp command is in the "AP1" series.)
This will only compare regular files, not character or block special files (in the /dev directory), sockets, etc., since the cmp command doesn't work properly with these. We would welcome suggestions on how to verify these "special" files.
If your new disk doesn't have the same partitions or organization as the old disk, modify the file /new-disk/etc/fstab on the new disk as appropriate.
Make sure that the disk partitions (first column) correspond to the organization you'll have with the new disk, once the old disk has been removed, and that you're only mounting one partition at / as shown in the second column.
For more information on the format of the file /etc/fstab, see the Linux System Administrator's Guide, section 4, under Mounting and unmounting.
For swap partitions, use a line similar to this one:
/dev/hda1 swap swap defaults 0 0 |
(Thanks to Rick Masters for helping with this.) |
We're assuming that LILO is installed on the hard disk's Master Boot Record (MBR); this seems to be the most common configuration. You want to install LILO on what's presently the second hard disk but will become the first hard disk.
Edit the file /new-disk/etc/lilo.conf as follows:
disk=/dev/hdb bios=0x80 # Tell LILO to treat the second # disk as if it were the first # disk (BIOS ID 0x80). boot=/dev/hdb # Install LILO on second hard # disk. map=/new-disk/boot/map # Location of "map file". install=/new-disk/boot/boot.b # File to copy to hard disk's # boot sector. prompt # Have LILO show "LILO boot:" # prompt. timeout=50 # Boot default system after 5 # seconds. (Value is in tenths of # seconds.) image=/new-disk/boot/vmlinuz # Location of Linux kernel. The # actual name may include a version # number, for example # "vmlinuz-2.0.35". label=linux # Label for Linux system. root=/dev/hda1 # Location of root partition on # new hard disk. Modify this as # appropriate for your system. # Note that you must use the name # of the future location, once the # old disk has been removed. read-only # Mount partition read-only at # first, to run fsck. |
Slackware only. Use image=/new-disk/vmlinuz.
If you're using a SCSI hard disk, you may have to add a line with initrd. See your existing file /etc/lilo.conf.
Install LILO on the new disk:
/sbin/lilo -C /new-disk/etc/lilo.conf |
The -C
option tells LILO what
configuration file to use.
If you wish, you can make a boot diskette, in case you run into problems when trying to boot the new disk.
Insert an empty diskette, format it, create a file system on it and mount it:
fdformat /dev/fd0H1440 mkfs.ext2 /dev/fd0 mount -t ext2 /dev/fd0 /mnt |
Debian only. With Debian 2.x, use /dev/fd0u1440 instead of /dev/fd0H1440. With Debian 1.x, use /dev/fd0h1440, with a lower case h.
Debian only. With Debian 2.x, use superformat instead of
fdformat.
You can ignore the error
mformat: command not found.
With Debian 1.x, if you don't have the command
fdformat, you can omit it if the floppy
is already formatted. In this case, you should check the diskette for
bad blocks by adding -c
after the
mkfs.ext2 command.
Slackware only. Use /dev/fd0u1440 instead of /dev/fd0H1440. With older versions, try /dev/fd0h1440, with a lower case h.
SuSE only. Use /dev/fd0u1440 instead of /dev/fd0H1440.
Copy all files in /boot to the diskette:
cp -dp /boot/* /mnt |
Red Hat only. If the /boot directory contains both vmlinux and vmlinuz files (note the difference in the last letter), you only need to copy the vmlinuz files to the boot diskette. They are the same as the vmlinux files, except that they're compressed to save space.
Slackware only. Copy the file /vmlinuz to the boot diskette; use the command cp /vmlinuz /mnt.
Create a new file /mnt/lilo.conf as follows:
boot=/dev/fd0 # Install LILO on floppy disk. map=/mnt/map # Location of "map file". install=/mnt/boot.b # File to copy to floppy's # boot sector. prompt # Have LILO show "LILO boot:" # prompt. timeout=50 # Boot default system after 5 # seconds. (Value is in tenths of # seconds.) image=/mnt/vmlinuz # Location of Linux kernel on # floppy. The actual name may # include a version number, for # example "vmlinuz-2.0.35". label=linux # Label for Linux system. root=/dev/hda1 # Location of root partition on # new hard disk. Modify this as # appropriate for your system. # Note that you must use the name # of the future location, once the # old disk has been removed. read-only # Mount partition read-only at # first, to run fsck. |
Install LILO on the boot diskette:
/sbin/lilo -C /mnt/lilo.conf |
The -C
option tells LILO what
configuration file to use.
Unmount the diskette:
umount /mnt |
Shut down the system and remove the old disk. Remember to modify the disk jumpers and the BIOS information to reflect the changes.
Reboot the system. If you have problems, you can use the boot diskette you just make. To do so, you may have to modify your BIOS's boot-up sequence to A:, C:.
You should modify the file /etc/lilo.conf in case you later want to run LILO again. Here's an example of how the file can look:
boot=/dev/hda # Install LILO on first hard # disk. map=/boot/map # Location of "map file". install=/boot/boot.b # File to copy to floppy's # boot sector. prompt # Have LILO show "LILO boot:" # prompt. timeout=50 # Boot default system after 5 # seconds. (Value is in tenths of # seconds.) image=/boot/vmlinuz # Location of Linux kernel. The # actual name may include a version # number, for example # "vmlinuz-2.0.35". label=linux # Label for Linux system. root=/dev/hda1 # Location of root partition on # new hard disk. Modify this as # appropriate for your system. read-only # Mount partition read-only at # first, to run fsck. |
Slackware only. Use image=/vmlinuz.
Thanks to Scott Christensen, Frank Damgaard, Alexandre Fornieles, David Fullerton, Igor Furlan, Jerry Gaines, Chris Gillespie, Nicola Girardi, Per Gunnar Hansoe, Richard Hawes, Ralph Heimueller, Gerald Hermant, Andy Heynderickx, Paul Koning, Hannu Liljemark, Claes Maansson, Rick Masters, Jason Priebe, Josh Rabinowitz, Douglas Rhodes, Valentijn Sessink, Kragen Sitaker, Stephen Thomas, and Gerald Willman.
This document may be translated into any language. If you do so, please
send a copy of the translation to Konrad Hinsen
<hinsen@cnrs-orleans.fr>
.
[1] | Does this also apply to other Aptiva models? |