User:Az990tony/edublog-beta-sw: Difference between revisions

From OLPC
Jump to navigation Jump to search
Line 819: Line 819:


Now that we have multi-boot and RAID1 mirroring in place, we can create Logical Volumes.
Now that we have multi-boot and RAID1 mirroring in place, we can create Logical Volumes.
HOWTOforge.com has a nice
[[http://www.howtoforge.com/linux_lvm Beginner's Guide To LVM]] if you are new to LVM.

Step 1. VolGroup00 for Library and Swap


Step 1.
Let's recreate the Logical Volumes that we had when we first installed XS school server.
Let's recreate the Logical Volumes that we had when we first installed XS school server.
First, we'll define three mirrored devices to serve as physical space. Go into "fdisk /dev/sda"
First, we'll define three mirrored devices to serve as physical space. Go into "fdisk /dev/sda"
Line 831: Line 834:
</tt>
</tt>


Use "mdadm" and "pvcreate" to indicate these can be used as physical space for a volume group.
Use "mdadm" to create the RAID1 pairs. This may take awhile as it needs to copy all of the blocks over. Then use "pvcreate" to indicate these can be used as physical space for a volume group.


<tt>
<tt>
Line 837: Line 840:
[root ~]# mdadm --create /dev/md9 --level=1 raid-disks=2 /dev/sda8 /dev/sdb9
[root ~]# mdadm --create /dev/md9 --level=1 raid-disks=2 /dev/sda8 /dev/sdb9
[root ~]# mdadm --create /dev/md10 --level=1 raid-disks=2 /dev/sda8 /dev/sdb10
[root ~]# mdadm --create /dev/md10 --level=1 raid-disks=2 /dev/sda8 /dev/sdb10
[root ~]# pvcreate /dev/md8
[root ~]# watch cat /proc/mdstat
[root ~]# pvcreate /dev/md9
[root ~]# pvcreate /dev/md8 /dev/md9 /dev/md10
[root ~]# pvcreate /dev/md10
[root ~]# pvdisplay
</tt>

Define the volume group. Since XS used "VolGroup00", we will use this also. I chose to use a 16MB extent size, instead of the default 4MB, so that we have fewer extents to work with.

<tt>
[root ~]# vgcreate -s 16M VolGroup00 /dev/md8 /dev/md9 /dev/md10
[root ~]# vgdisplay
</tt>

Now let's create the logical volumes. The XS had two, LogVol00 for the /library directory,
and LogVol01 for the swap space. This allows us to show the difference between lower case "-l" and upper case "-L". Lower case is in units of "extents" which we defined as 16MB, so here
library is a tiny 16MB volume (we can always expand it later!!!) The -L allows you to specify the volume group in M(MB) or G(GB). We make the swap space 2G to match the amount of memory we have installed on our system.

<tt>
[root ~]# lvcreate -l 1 --name LogVol00 VolGroup00
[root ~]# lvcreate -L 2G --name LogVol01 VolGroup00
[root ~]# lvdisplay
</tt>
</tt>


Now, to activate these into the system. For Library, the initial XS CD created it empty
Define the volume group. Since XS used "VolGroup00", we will use this also.
except for a lone "users" directory. We can recreate that here. Look at how /etc/mtab was updated with this information, and copy that into /etc/fstab to have /library mounted at boot time automatically. On my system, it got remapped to /dev/mapper/VolGroup00-LogVol00.


<tt>
<tt>
[root ~]# vgcreate VolGroup00 /dev/md8 /dev/md9 /dev/md10
[root ~]# mkdir /library
[root ~]# mount /dev/VolGroup00/LogVol00 /library
[root ~]# mkdir /library/users
[root ~]# vi /etc/mtab /etc/fstab
</tt>
</tt>



Revision as of 19:30, 21 June 2008

This software stack is based on This Beta Server hardware built for the Educational Blogger Project .

Devices Available

Based on the hardware, the following devices are available:

CD/DVD Burner

/dev/hda (it is in the Primary/Master position on the first EIDE cable, but can be represented on some systems as /dev/cdrom, /dev/dvd, /dev/dvdrw, and /dev/sr0)

SATA disks

/dev/sda (this is drive 1)

/dev/sdb (this is drive 2)

Note that /dev/sda1, sda2, sda3 and sda4 can be bootable primary partitions. Typically, sda4 is configured as an "extended partition" which allows subsets to be defined as logical partitions sda5, sda6, and so on.

USB keys or external drives

/dev/sdc for the drive itself /dev/sdc1 (for the VFAT file system usually on it)

Note: depending on when the USB was attached, this could be /dev/sdd1, /dev/sde1 or /dev/sdf1.

Ethernet Network Interface cards

eth0, eth1 and eth2 seemed to be assigned to the one motherboard connector and two PCI cards differently between operating systems. I will need to track down the MAC addresses to keep them straight.

RAID and Logical Volume Manager

RAID and LVM are supported on most recent Linux distributions and provide some key advantages. However, they can also be difficult to work with and require knowledge in particular commands to handle properly.

Redundant Array of Inexpensive Disks (RAID)

Configurations include:

RAID 0

Stripes data across for better performance. Two 160GB drives would look like a fast 320GB drive, but if you lose any one drive, you have lost everything

RAID 1

Mirrors the data across drives. Multiple copies means you have better protection, so if you lose any one drive, the rest of the system can still run. Two 160GB drives would look like a single 160GB drive. Performance is the same or slightly slower than a single drive.

RAID 5

Stripes with parity for added protection. One drive's worth is set aside for parity, so three 160GB drives would hold 320GB of data; four 160GB drives would hold 480GB of data. Since we have only two drives, this option does not apply to us. Performance is faster for read, but can be slower for writes, depending on how the parity methods are implemented.

RAID 10

Striped and Mirrored. Sometimes referred to as RAID 0+1 or RAID 1+0, this combines RAID 1 with RAID 0. Data is striped across disk 1 and disk 2, disk 3 is a mirror of disk 1 stripe, and disk 4 is mirror of disk 2 stripe. Thus four 160GB drives holds 320GB of data. If you lose any one drive, the mirrored copy is still available, so the system will be able to continue to run. Performance is same or slightly slower than a 2-drive RAID 0 configuration.

LINEAR

This just strings together a bunch of different sized disks sequentially. A 160GB drive, 320GB and 500GB drive could be combined into 980GB.

JBOD

JBOD stands for "Just a Bunch of Disks", a term meaning there is no RAID for protection or performance improvement. This refers to deployments where each disk drive is independent.

RAID-1 Considerations:

  • Can mirror the entire drive ( /dev/sda <--> /dev/sdb ) or individual same-sized partitions ( /dev/sda1 <--> /dev/sdb1 )
  • Can mirror swap space. This appears to be preferred best practice over just having swap space on separate disks.
  • Mirrored drives appear as:
    • Software RAID (/dev/md0, /dev/md1)
    • or, Hardware-assisted BIOS "fakeraid" (/dev/dm-0, /dev/dm1) which is not supported on the hardware I have for this project
  • In general, a mirrored partition can be used wherever a native partition can be used
  • Can be used as physical volumes for Logical Volume Manager
  • /boot directory can be on RAID1, but not be on RAID0, RAID5, RAID10 or any other RAID configuration.

RAID rebuild: In the event a disk fails in RAID-1 mode, the single surviving drive can be used to continue operations. When a new replacement drive is available, the following procedures can be used to rebuild.

Place the surviving drive as "drive 1", remove the bad drive from the system.

  • boot up
  • indicate that drive 2 has failed

Place the new drive as "drive 2"

  • boot up
  • copy over the disk layout from "drive 1" using "sfdisk" utility
  • copy over the MBR from "drive 1" using "dd" utility
  • add the new partitions to /dev/md0, /dev/md1, etc. (the data will be copied over)

Logical Volume Manager (LVM2)

Logical Volume Manager allows a level of storage virtualization between the physical hardware and the operating system and applications. There are two formats, LVM (older) and LVM2 (newer). We will use LVM2.

Physical Volumes

A physical volume (pv) can be an entire disk drive (/dev/sda), partition (/dev/sda1), or RAID mirror (/dev/md0)

Volume Groups

A volume group (vg) combines the space capacity of one or more physical volumes. A physical volume can only be part of a single volume group. A volume group can host one or more logical volumes.

Logical Extents

The volume group is divided into fixed-size pieces called "logical extents" or LE. These are typically 32MB in size.

Logical Volume

A logical volume is a "virtual drive" that appears like a device or partition to the operating system. A logical volume can be the entire space of a single volume group, or a subset of a single volume group. Logical volumes can be increased or decreased in size, and can be moved as needed onto or away from specific physical volumes. A logical volume is one or more logical extents, so the smallest Logical Volume is one LE (typicallly 32MB). A logical extent can only be part of one logical volume.

LVM2 Considerations:

  • A volume group can contain physical volumes of different sizes
  • Linux cannot boot from a Logical Volume, this can be solved by having a separate /boot directory that is not part of the Logical Volume, and the rest of root (/) in the logical volume.
  • Logical volumes cannot be resized using the "gparted" tool
  • Linux swap space can be defined onto logical volume

Disk Layout

Most Linux systems only require the following:

  • A Master Boot Record (MBR) that contains the boot loader and disk layout
  • A root directory (/)
    • The root directory can be split up into subdirectories for /boot, /var, /home on separate partitions
  • swap space (typically as big or bigger than installed RAM)

Possible Disk Layout configurations

Several disk layouts are possible. Depending on the Operating System installation procedure, you may have to install in JBOD, RAID or LVM, and then migrate to be a full RAID+LVM configuration.

1. Simple JBOD (no RAID, no LVM)

/dev/sda                                 /dev/sdb
MBR (boot loader / disk layout )         MBR (disk layout only)
  /dev/sda1 as /boot                         /dev/sdb1 for extra space
  /dev/sda5 as /
  /dev/sda6 as swap

2. RAID (no LVM)

/dev/sda                                                 /dev/sdb
MBR1 (boot loader/layout)  <-independent->                  MBR2 (boot loader/layout)
  /dev/sda1 as /boot       <- /dev/md0  ->                  /dev/sdb1
  /dev/sda5 as /           <- /dev/md1  ->                  /dev/sdb5
  /dev/sda6 as swap        <- /dev/md2  ->                  /dev/sdb6

2. LVM (no RAID)

/dev/sda                                 /dev/sdb
MBR (boot loader / disk layout )         MBR (disk layout only)                        
  /dev/sda1 as /boot                         /dev/sdb1 as physical volume
  /dev/sda5 as physical volume
where:
VolGroup00 --- combines /dev/sda5 and /dev/sdb1
LogVol00 -- as /
LogVol01 -- as swap

3. RAID (no LVM)

/dev/sda                                            /dev/sdb
MBR1 (boot loader/layout)      <-independent->      MBR2 (boot loader/layout)
  /dev/sda1 as /boot           <- /dev/md0  ->          /dev/sdb1
  /dev/sda5 as /               <- /dev/md1  ->          /dev/sdb5
  /dev/sda6 as swap            <- /dev/md2  ->          /dev/sdb6

4. RAID and LVM

/dev/sda                                            /dev/sdb
MBR1 (boot loader/layout)      <-independent->      MBR2 (boot loader/layout)
  /dev/sda1 as /boot           <- /dev/md0  ->          /dev/sdb1
  /dev/sda5 as physical vol    <- /dev/md1  ->          /dev/sdb5   
where:
VolGroup00 --- /dev/md1
LogVol00 -- as /
LogVol01 -- as swap

5. RAID and LVM with rescue partition

/dev/sda                                            /dev/sdb
MBR1 (boot loader/layout)      <-independent->      MBR2 (boot loader/layout)
  /dev/sda1 as /boot           <- /dev/md0  ->          /dev/sdb1
  /dev/sda2 as /rescue         <- /dev/md1  ->          /dev/sdb2
  /dev/sda5 as physical vol    <- /dev/md2  ->          /dev/sdb5   
where:
VolGroup00 --- /dev/md2
LogVol00 -- as /
LogVol01 -- as swap


6. RAID and LVM with rescue and scratch partitions

/dev/sda                                            /dev/sdb
MBR1 (boot loader/layout)      <-independent->      MBR2 (boot loader/layout)
  /dev/sda1 as /boot           <- /dev/md0  ->          /dev/sdb1
  /dev/sda1 as /rescue         <- /dev/md1  ->          /dev/sdb2
  /dev/sda3 as /scratch1       <-independent->          /dev/sdb3 as /scratch2
  /dev/sda5 as swap            <-independent->          /dev/sdb5   
  /dev/sda6 as physical vol    <- /dev/md2  ->          /dev/sdb6  
  /dev/sda7 as physical vol    <- /dev/md3  ->          /dev/sdb6 
where:
VolGroup00 --- combine /dev/md2 and /dev/md3
LogVol00 -- as /
LogVol01 -- as swap

Partition Sizes

The following sizes can be used:

boot -- 680 MB

This could fit on a single CD for recovery purposes. /boot will contain /boot/grub boot loader, and can contain alternative kernels to select from during the boot process. Most OS only need 100MB, so this should be plenty.

rescue -- 4GB

This would be enough to have a basic operating system with software utilities, and include manuals, ISO images of CDs, and so on. This could fit on a single DVD or several CDs for recovery purposes.

scratch -- 30GB each

This can be used to download ISO files, store "partimage" backups, and so on. These are not RAID protected, so if you lose a disk drive, these files will not be mirrored on the other drive.

swap -- 2GB

This can be swap during the initial installs.

base1, base2, base3 -- rest of the disk (40GB of so each)

This will be the /dev/sda4 extended partition carved into smaller logical partitions that can be defined as physical volumes or RAID-1 mirrored pairs.

Initial Drive Format

For some reason, with fresh disks that have no format or partition, most 32-bit LiveCD were unable to boot. To solve this, I booted a 64-bit AMD64-based LiveCD (Fedora 8 in my case), created the partitions and then could use a 32-bit LiveCD for the rest of the install.

[[ http://sysresccd.org/Main_Page | SysRescCD 1.0.3 ]] is also capable of booting in either 64-bit or 32-bit mode, use "boot: rescue64" to boot into 64-bit mode. If you received the disk drive from someone else, and are concerned about remnants of any past data or operating systems, you can use the following command to clobber all data with random data. At 35 MB/sec, this would take about 90 minutes for each 160GB drive.

dd if=/dev/urandom of=/dev/sda

Boot from the "SysRescCD" and use "fdisk" to configure /dev/sda. The final layout should look something like this:

Disk /dev/sda: 160.0 GB, 160041885696 bytes
255 heads, 63 sectors/track, 19457 cylinders
Units = cylinders of 16065 * 512 = 8225280 bytes
Disk identifier: 0x000a0a65
_                                                           _
  Device Boot      Start         End      Blocks   Id  System
/dev/sda1   *           1          80      642568+  83  Linux
/dev/sda2              81         581     4024282+  83  Linux
/dev/sda4            4230       19457   122318910    5  Extended
/dev/sda5            4230        4497     2152678+  82  Linux swap / Solaris

Note that all of the partitions are of type "83" (native Linux) except /dev/sda5 which is type "82" (Linux swap).

Operating System

The team wants to develop and test the EduBlog software on a variety of platforms. The operating systems considered were:

XS School Server

We're using the XS-163 (the latest level at this time). This is a customized version of Fedora 7 with special settings to run as a school server. The CD runs unattended, works when there is only a single SATA drive installed, and wipes away any previous partition table to form a new partition table.

Fedora 7

The XS School Server is based on Fedora 7, so if we are unable to get the XS installation kickstart file to support the disk layout, we felt a backup plan was to install Fedora 7 natively. Fedora 7 could also be used graphically by Glen on premises to help with problem determination.

Debian 4.0-r3 Etch

Ceibal (OLPC Uruguay) has chosen to use Debian 4.0-r3 Etch instead for their projects. It was a stretch goal to see if we could set up a "dual-boot" or "multi-boot" environment that would support XS-Debian, Fedora-Debian, or XS-Fedora-Debian configurations.

System Rescue

While not a complete Operating System to run EduBlog or perform development work, we might need to have an independent OS for administration or maintenance. Many utilities work best on data that is not currently active in use. This could be done with a LiveCD or as a separate partition on the disk drive.

[[ http://sysresccd.org/Main_Page | SysRescCD 1.0.3 ]] supports sshd so that with the right network setup, someone could effect software repairs remotely. It has a variety of utilities, including "partimage" that can be used to backup individual partitions or logical volumes. Installing on hard disk is a matter of copying just seven files over to /sysrcd directory on a primary or mirrored partition, but not in a Logical Volume. "GRUB" as the boot loader can be updated to run this at reboot.

Scaffolding

A simple technique to build a complicated configuration is known as "scaffolding", borrowing from the concept of building a large house, apartment or office structure.

Here is the basic sequence:

  1. Prepare the first OS (XS-163 School Server)
  2. Prepare the second OS (Fedora 7)
  3. Prepare the third OS (Debian 4 Etch)
  4. Re-define partition tables with no LVM2 or RAID
  5. Assemble the individual OS images into a "multi-boot" environment
  6. Connect the second SATA drive
  7. Create RAID pairs of each partition you want protected
  8. Devine LVM2 pv, vg and lv and move data into the appropriate logical volume

Preparation

Linux installers make a few general assumptions

  • You are a novice user willing to accept a standard disk layour (true for most people)
  • You are installing only a single OS on the system (true for most)
  • It will take up all the disk space available (true for most)
  • That you won't mind that some partitions are LVM2 Logical Volumes (true for most)
  • If you have two identical-size drives, then you might want Software RAID capability, creating crazy devices like /dev/mapper/pdc_bbbhp1 instead of /dev/sda1. Unless you specify RAID during the initial interactive setup, it will create these crazy devices, but not actually put them into RAID mode for you. The default is "No RAID, crazy device names" (what ?)

If you have a normal Linux distribution, you can go through an interactive meny and choose which options you like. However, the XS-163.iso for XS School Server is non-interactive (batch, unattended, hands off) installation that assumes "standard" device names.

To get around these assumptions, we will detach the red SATA signal cable from the motherboard of the second disk drive (you can leave the power cable attached to the drive itself is fine). We will wipe the system clean three times. Each time we will install a fresh operating system image as if it were the only operating system we are going to use, and then backup the appropriate files to external media using "partimage" utility.

Step 1. Wipe the System clean

For each OS, wipe the system clean. The following command will zero out the first cylinder of your hard disk eliminating the master boot record and partition table.

[root ~]#  dd if=/dev/zero of=/dev/sda bs=512 count=16

Step 2. Install from CD

Run the installation from CD. The XS-163 will carve up the disk as follows:

/dev/sda1     /boot -- 100MB
/dev/sda2     /     -- (root) 8GB (8192MB)
/dev/sda5     pv    -- physical volume for LVM2, rest of disk

VolGroup00  -- volume group on physical volume
LogVol00    /library
LogVol01    swap

If you are installing Fedora 7, Debian 4, etc. wipe out all partitions, and choose "create custom setup" and define the following simple setup. The rest of the drive capacity will be unused, but it is important to make the initial /(root) directory as small as possible.

/dev/sda1     /boot -- 200MB
/dev/sda2     /     -- (root) 8GB (8192MB)
/dev/sda3     swap

Step 3. Capture initial list of application packages

Before you change anything or install any packages, do the following, which will document all of the base RPM packages on your XS163 or Fedora 7 system. That way you can compare any new packages that were installed afterwards and see the differences using diff command. For Debian 4, use the apt-get utility instead.

[root ~]# cd /root
[root ~]# mkdir /boot/details
[root ~]# rpm -qa | sort > /boot/details/rpm-xs163.txt

Step 4. Create partition images

We will now backup this OS image. These partition images will then be used to re-assemble the final system. We can use this procedure for Backup-and-Restore-Using-SysRescueCD.

Here is what you end up with:

For each OS you will have:

  • master.mbr -- this is the Master Boot Record (first 446 bytes of each disk)
  • sfdisk-d.txt -- this is the disk layout, human-and-computer readable
  • grubmenu.txt -- this has the initial boot parameters (/boot/grub/menu.lst --> grub.conf)
  • etcfstab.txt -- this is the list of mount points (/etc/fstab file)
  • sda1boot.img.000 -- this is partition image of the /boot directory
  • sda2base.img.000 -- this is first piece of the partition image of the / (root) directory
  • sda2base.img.001 -- (the files are 692MB each, small enough so they can be burned onto CDrom, uploaded to an FTP site, sent over using "yousendit.com", or fit on a FAT32 file system.)

Re-Define the Partion Tables

We'll choose a configuration similar to the left half of "6. RAID and LVM with rescue and scratch partitions" list above in the section Possible_Disk_Layout_configurations.

/dev/sda                      
MBR1 (boot loader/layout)      
  /dev/sda1 as /boot           
  /dev/sda1 as /rescue (/ for Debian 4)  
  /dev/sda3 as /scratch1
  /dev/sda4 as extended partition holding sda5..sda10
  _
  /dev/sda5  as / (for fedora 7)    
  /dev/sda6  as / (for XS 163)
  /dev/sda7  as swap            
  /dev/sda8  as physical vol    
  /dev/sda9  as physical vol      
  /dev/sda10 as physical vol    

The /boot will contain the kernel images of all the operating systems, as well as the multi-boot /boot/grub directory. This will allow you to choose the different OS from a menu.

I chose 4GB to hold a set of rescue files, and then later decided to install Debian 4, it fits in 4GB! I create three equal-sized 10GB partitions, one for scratch (temp) space, one for the root directory of Fedora 7, and one for root directory of XS 163 School Server. Finally, I divided the rest of the volume into three 40GB pieces which can be used for physical volumes of an LVM Volume Group.

Here is what the start/end cylinder counts look like using "sfdisk -l /dev/sda":

Disk /dev/sda: 19457 cylinders, 255 heads, 63 sectors/track
Units = cylinders of 8225280 bytes, blocks of 1024 bytes, counting from 0
_
 Device   Boot Start     End   #cyls    #blocks   Id  System
/dev/sda1   *      0+     79      80-    642568+  83  Linux
/dev/sda2         80     638     559    4490167+  83  Linux
/dev/sda3        639    1839    1201    9647032+  83  Linux
/dev/sda4       1840   19456   17617  141508552+   5  Extended
/dev/sda5       1840+   3040    1201-   9647001   83  Linux
/dev/sda6       3041+   4241    1201-   9647001   83  Linux
/dev/sda7       4242+   4498     257-   2064321   82  Linux swap / Solaris
/dev/sda8       4499+   9484    4986-  40050013+  8e  Linux LVM
/dev/sda9       9485+  14470    4986-  40050013+  8e  Linux LVM
/dev/sda10     14471+  19456    4986-  40050013+  8e  Linux LVM

Since there is so much work setting this up, it might be good to save it to a file. This can be done using the following command:

[root ~]#   sfdisk -d /dev/sda  > sfdisk-d.txt

This is what the file looks like. Not that "size" is counting in terms of 512 byte sectors. So /dev/sda1 is 512 * 1285137 or 657990144 bytes (658 MB).

# partition table of /dev/sda
unit: sectors
/dev/sda1 : start=       63, size=  1285137, Id=83, bootable
/dev/sda2 : start=  1285200, size=  8980335, Id=83
/dev/sda3 : start= 10265535, size= 19294065, Id=83
/dev/sda4 : start= 29559600, size=283017105, Id= 5
/dev/sda5 : start= 29559663, size= 19294002, Id=83
/dev/sda6 : start= 48853728, size= 19294002, Id=83
/dev/sda7 : start= 68147793, size=  4128642, Id=82
/dev/sda8 : start= 72276498, size= 80100027, Id=8e
/dev/sda9 : start=152376588, size= 80100027, Id=8e
/dev/sda10: start=232476678, size= 80100027, Id=8e

To restore, just change the ">" (send to) into "<" (read from) character. You can cut and paste the above and try it yourself. Note, this will clobber any previous partition table and all data may be lost.

[root ~]#   sfdisk -d /dev/sda  < sfdisk-d.txt

Assemble the Images into multi-boot

All three of the Linux operating systems support the "GRUB" boot loader. Of these, I liked the graphics on Fedora 7, so I will choose to use that as the base.

Let's boot from the SysRescueCD 1.0.3 CD. I have all my partition images on an external USB drive, but this can work with CD or DVD as well.

Step 1. Restore Fedora 7

We will use sda1 for /boot, and sda5 for / (root). Use partimage to restore each partition. Note that if you have several in a series, you just need to specify the *.000 version, and it will pick up the others automatically.

[root ~]# mount /dev/sdc1 /mnt/backup
[root ~]#  cd /mnt/backup/fedora7
[root fedora7]#  dd if=master.mbr of=/dev/sda bs=446 count=1
[root fedora7]#  partimage -e -b restore /dev/sda1 sda1_boot.img.000
[root fedora7]#  partimage -e -b restore /dev/sda5 sda2_base.img.000

We will need to edit a few files. As a precaution, I always make a bacukp copy (*.bak) of any file that I am editing with "vi", in case I type something wrong and accidently save it.

[root ~]#  cd /mnt/backup/fedora7
[root fedora7]#  mkdir /mnt/fedora
[root fedora7]#  mount /dev/sda5 /mnt/fedora
[root fedora7]#  mount /dev/sda1 /mnt/fedora/boot
[root fedora7]#  cd /mnt/fedora/boot/grub
[root grub]#  cp menu.lst menu.bak
[root grub]#  vi menu.lst
[root grub]#  cd /mnt/fedora/etc
[root etc]#  cp fstab fstab.bak
[root etc]#  vi /mnt/fedora/etc/fstab
[root etc]#  cp mtab mtab.bak
[root etc]#  vi /mnt/fedora/etc/mtab

For menu.lst (which is just a symbolic link to grub.conf, by the way), I changed the following:

  • timeout value to "15" seconds
  • put a "#" in front of "hiddenmenu" (so that I could see the menu)
  • changed to "root=/dev/sda5"
  • added "panic=30" (optional)
    • If you get a kernel panic, it probably means there is a mismatch between menu.lst, fstab or mtab files. Make sure you change all three before rebooting your system. The "panic=30" allows the system to reboot after 30 seconds of a kernel panic, giving you enough time to insert your rescue CD to fix the problem.
    • Without "panic=30" the kernel panic will just sit there, and you then have to press the power button for 5 seconds to shut the system down, then hit the power button again, and then on

my system I have to hit "f1" to confirm that my system was turned off in that manner, or "Del" to go into the BIOS setup program.

  • Added "# 0-Fedora" as a comment to remind me that "default=0" will select it.(optional)

Note that the default=0 means that the first stanza will be selected if you do nothing for the timeout value (15 seconds). On the menu, you will see the 15 seconds count down to 0, and the default highlighted. Your file will look something like this:

# /boot/grub/menu.lst --> grub.conf generated by anaconda
#boot=/dev/sda
#
default=0
timeout=15
splashimage=(hd0,0)/grub/splash.xpm.gz
#hiddenmenu
# 0-Fedora7
title Fedora 7 Core (2.6.21-1.3194.fc7)
       root (hd0,0)
       kernel /vmlinuz-2.6.21-1.3194.fc7 ro root=/dev/sda5 panic=30
       initrd /initrd-2.6.21-1.3194.fc7.img

The /etc/fstab file will look something like this. sda5, sda1, and sda7 are the root, boot, and swap locations. If you have anything that says LABEL= change it to the appropriate /dev/sda* device.

# /etc/fstab
/dev/sda5               /                       ext3    defaults        1 1
/dev/sda1               /boot                   ext3    defaults        1 2
tmpfs                   /dev/shm                tmpfs   defaults        0 0
devpts                  /dev/pts                devpts  gid=5,mode=620  0 0
sysfs                   /sys                    sysfs   defaults        0 0
proc                    /proc                   proc    defaults        0 0
/dev/sda7               swap                    swap    defaults        0 0

The /etc/mtab might also need to be updated to match the fstab.

Since we moved the root directory location, we may want to re-install grub. This will refresh the appropriate files into /mnt/fedora/boot/grub (which is actually on /dev/sda1), but will not over-write any menu.lst, grub.conf, or spash.xpm.gz files you already have.

[root fedora7]#  mkdir /mnt/fedora
[root fedora7]#  mount /dev/sda5 /mnt/fedora
[root fedora7]#  mount /dev/sda1 /mnt/fedora/boot
[root fedora7]#  grub-install --root-directory=/mnt/fedora --no-floppy

We can now unmount the USB drive, use "sync" to make sure the USB drive can be detached, and reboot with no CD in the CD tray.

[root fedora7]#  cd /root
[root ~]#  umount /mnt/fedora
[root ~]#  sync
[root ~]#  eject
[root ~]#  reboot

If all goes well, you should have a running Fedora system again.

Step 2. Restore Debian

For the second OS, we cannot restore "/boot" over the existing one, so we redirect this to /dev/sda3 which is our scratch space.

[root ~]# mount /dev/sdc1 /mnt/backup
[root debian4]#  cd /mnt/backup/debian4
[root debian4]#  partimage -e -b restore /dev/sda3 sda1_boot.img.000
[root debian4]#  partimage -e -b restore /dev/sda2 sda2_base.img.000

Now we need to copy parts of the sratch space over into the /boot directory. The easiest way I have found to do this is to use "vi" on two files, position your cursor on the stanza you want to carry over from scratch, "5yy" will put those five lines in memory, "Esc:n" will switch to the next file, and then position the cursor to where you want, and use "p" to paste from memory.

[root ~]#  cd /mnt/backup/debian4
[root debian4]#  mkdir /mnt/debian /mnt/scratch
[root debian4]#  mount /dev/sda5 /mnt/debian
[root debian4]#  mount /dev/sda3 /mnt/debian/boot
[root debian4]#  mount /dev/sda1 /mnt/scratch
[root debian4]#  cd /mnt/debian/boot/grub
[root grub]#  cp menu.lst menu.bak
[root grub]#  vi /mnt/scratch/menu.lst menu.lst

You're merged menu.lst will look something like this. Note that the kernel levels are different, and that the Debian stanza has "root=/dev/sda2" instead of what Fedora has.

# /boot/grub/menu.lst --> grub.conf generated by anaconda
#boot=/dev/sda
#
default=0
timeout=15
splashimage=(hd0,0)/grub/splash.xpm.gz
#hiddenmenu
# 0-Fedora7
title Fedora 7 Core (2.6.21-1.3194.fc7)
       root (hd0,0)
       kernel /vmlinuz-2.6.21-1.3194.fc7 ro root=/dev/sda5 panic=30
       initrd /initrd-2.6.21-1.3194.fc7.img
# 1-Debian4
title Debian GNU/Linux, kernel 2.6.18-6-486
       root	(hd0,0)
       kernel	/vmlinuz-2.6.18-6-486 root=/dev/sda2 ro panic=30 
       initrd	/initrd.img-2.6.18-6-486

You will also have to edit the fstab and mtab files, as we did for Fedora. The /boot will be /dev/sda1, the / (root) will be /dev/sda2, and the swap is /dev/sda7. Reboot and make sure that you can launch either operating system successfully.

Step 3. Restore XS School Server

We can add as many operating systems as we like using this approach. For XS system, we will only restore the /boot and /(root) directories. We will reuse /dev/sda3 as scratch space, and restore the root directory into /dev/sda6. We can restore or re-create the /library partition later.

[root ~]# mount /dev/sdc1 /mnt/backup
[root ~]#  cd /mnt/backup/xs163
[root xs163]#  partimage -e -b restore /dev/sda3 sda1_boot.img.000
[root xs163]#  partimage -e -b restore /dev/sda6 sda2_base.img.000

Carry over the stanza to the existing /boot/grub/menu.lst, so that you now will have three stanzas in your menu.list, #0-Fedora, #1-Debian, and #2-XS163. Here is the stanza for the XS school server (XS-163 level).

# 2-XS163
title XS163 School Server (2.6.23.1-21.fc7)
      root (hd0,0)
      kernel /vmlinuz-2.6.23.1-21.fc7 ro root=/dev/sda6 panic=30
      initrd /initrd-2.6.23.1-21.fc7.img

Edit /etc/fstab, and /etc/mtab as we did with fedora and debian above. If you just want to merge /library in the / (root) directory, use the following commands:

[root ~]# mkdir /library
[root ~]# mkdir /library/users

Take Fresh Backup

Before we continue, now is a good time to backup your multi-boot system using the Backup-and-Restore-Using-SysRescueCD procedure. Put these in a different folder, marked "multi-boot" or something, so that you can remember to keep them separate from the previous partition images. Take a copy of every /etc/fstab from /dev/sda2, /dev/sda5 and /dev/sda

Connect second disk drive and configure RAID

We installed all the OS images with a single drive, took a fresh backup, so now it is time to reconnect the SATA signal cable to the motherboard.

I found while testing out these procedures that it was easier just to leave the left side panel off the computer, so that I could connect or detach the red SATA signal cable as needed. For safety, I always did a full shutdown, disconnected the 115 volt power source from the back of the machine, and then attached/detached the red SATA signal cable.

For this, I was inspired by the excellent resources on HOWTOforge.com, especially these long, but extremely well-written tutorials by Falko Timme.

Each tutorial follows these basic steps (the sequence is slightly different between Fedora and Debian)

  1. Use "sfdisk -d /dev/sda | sfdisk /dev/sdb" to copy over the partition layout to the second drive
  2. install "initramfs/mkinitrd" and "mdadm" (Mirrored Disk Administration) program
  3. Use "modprobe" to load "raid1" module, you can load "linear", "raid0" and others, but I only did "raid1". For debian, you must also load "md" module.
  4. Use "fdisk" to change sdbx from type "83"(Linux) to type "fd"(RAID autodetect)
  5. Use "mdadm --zero-superblock /dev/sdbx" to clear any previous RAID values
  6. Use "mdadm --create /dev/mdx --level=1 raid-disks=2 missing /dev/sdbx" to create a "deprecated" half-mirror, which I will call "(missing+sdbx)" shown as "_U".
  7. Format the half-mirror with "mkfs.ext3 -L MYLABEL /dev/mdx" (putting unique labels on each one greatly helped me keep them straight!)
  8. Update "/etc/mdadm.conf" with this new RAID setting.
  9. Mount "/dev/mdx" as "/mnt/mdx" and update the /etc/mtab, /etc/fstab. Add a new stanza in your /boot/grub/menu.lst files to indicate "root (hd1,0)" and/or "ro root=/dev/mdx". The first points to the /boot directory, and the second to the / (root) directory.
  10. Update the "initrd" file based on the values in /etc/mtab and /etc/fstab, and the modules you have loaded.
  11. copy the files over from /dev/sdax to /dev/mdx with "cp -dpRX . /mnt/mdx"
  12. run "grub" against both master boot records (hd0 is /dev/sda, and hd1 is /dev/sdb, according to "/boot/grub/device.map" file)
  13. reboot from the new stanza using the new /boot and new /root.
  14. Use "fdisk /dev/sda" to change the partition type of /dev/sdax from "83" to "fd" (RAID autodetect)
  15. Use "mdadm /dev/mdx --add /dev/sdax" to make the half-mirror whole. Now it is (sdax+sdbx). This will copy what is on /dev/sdbx over to /dev/sdax block for block. Let this finish! Do not reboot until you get the "UU" on "cat /proc/mdstat" command.
  16. Update /etc/mdadm.conf from "mdadm --examine --scan >> /etc/mdadm.conf" and edit out the duplicate lines with "vi". If there are two lines for /dev/mdx, delete the first one and keep the second.
  17. Update /boot/grub/menu.lst so that the stanza boots from root=(hd0,0) again.
  18. Upadte the "initrd" again, reboot and confirm everything works

Here were a few things to watch out for.

  • Fedora and XS163 use "yum install", but Debian uses "apt-get install"
  • I found it confusing to have /dev/md0 for (sda1+sdb1), /dev/md2 for (sda5+sdb5), and so on, so I named mine all to match /dev/md1=(sda1+sdb1), /dev/md5=(sda5+sdb5), etc.
  • Fedora and XS163 expect the "mdadm.conf" to be in the /etc directory, but Debian expects it to be in the /etc/mdadm/mdadm.conf directory. (See my symbolic link workaround below.)
  • Fedora uses "mkinitrd" and Debian uses "update-initramfs -u" (see my preload=raid1 workaround for XS163 below)
  • The tutorial assumes you will convert over the /boot and / (root) over all at the same time, but then you have problems with the other two OS images. (See my conversion sequence below.)

/etc/mdadm.conf

I was following the Debian method, and put my config options and RAID settings in /etc/mdadm/mdadm.conf on Fedora systems, and failed miserably. I wanted all the OS images to have them in the same location, so I used a symbolic link on Debian. Now, all three OS images have their values in /etc/mdadm.conf location.

[root ~]#  cd /etc
[root etc]#  ln -s /etc/mdadm/mdadm.conf mdadm.conf
[root etc]#  ls -l mdadm.conf

initrd Intial Ram file system images

The tutorial updates the initrd twice for each operating system. At first I skipped the step, thinking "I already have one" or "I already did that".

  • The tutorial creates a unique backup each time, which is probably not a bad idea, but I used the "mkinitrd -f" option that overwrites the previous file so I wouldn't have that many copies.
  • It is important to update your initrd images based on the values in the /etc/fstab, /etc/mtab and modprobe loaded modules so that your system boots properly. If you don't do this right,

you may get "kernel panic" or "file system integrity checks". This can only be done from the running system, otherwise "uname -r" returns the wrong value and you have it linked to the wrong kernel.

  • I was having problems with the XS163 mkinitrd, so used "mkinitrd --preload=radi1" parameter and that seemed to solve the problem. This is needed if your /boot or / (root) directory is RAID1 mirrored from a different operating system and you did not update your /etc/mdadm.conf correctly.
  • It is important to only reboot when all the right changes have been made as a group. Rebooting too soon or in between steps is a bad idea.

RAID conversion sequence

The tutorial works fine for single OS image systems. However, if you convert one OS /boot and / (root) directories, then when you try to boot into the second OS, it can't make sense of the /boot directory, or worse, changes one drive but not the other. The trick is to convert the /boot drive last.

I will refer to the tutorials steps as "R1" through "R18" above.

Step 1. You only need to do "R1" once, when the second drive is connected. You have a fresh backup that includes "sfdisk-d.txt", so you can always re-create it if you had to.

Step 2. Convert Debian (/dev/sda2) root partition only, using steps R2-11, R13-R16, and R18. For Step R9 change "ro root=/dev/md2" but leave "root (hd0,0)" alone. You can skip steps R12 and R17, since that only applies for /boot conversion.

Step 3. Convert Fedora (/dev/sda5) root partition only, using steps R2-11, R13-R16, and R18. Step R8, copy over the "etc/mdadm.conf" from Debian, and use "mdadm --assemble --scan" to repopulate the list of active pairs. The debian file has a lot more extra configuration option descriptions. For Step R9 change "ro root=/dev/md5" but leave "root (hd0,0)" alone. Again, you can skip steps R12 and R17, since that only applies for /boot conversion.

Step 4. Convert XS School Server (/dev/sda6) root partition only, using steps R2-11, R13-R16, and R18. For Step R9 change "ro root=/dev/md6" but leave "root (hd0,0)" alone. Step R8, copy over the "etc/mdadm.conf" from Fedora, which now has all of the Debian and Fedora settings combined. Again, you can skip steps R12 and R17, since that only applies for /boot conversion.

Step 5. Convert the common /boot directory, from any of the above operating systems, using steps R4-R18. You only have to convert this once, and it doesn't matter which system you do it from.

Step 6. Take the /etc/mdadm.conf and copy the final version with all of the devices (md1, md2, md5, and md6) to /etc of each operating system. This can all be done from one OS to update the others. For example, from the XS163 image:

[root ~]#  cd /etc
[root etc]#  mkdir /debian
[root etc]#  mount /dev/md2 /debian
[root etc]#  cp mdadm,conf /debian/etc/
[root etc]#  mkdir /fedora
[root etc]#  mount /dev/md5 /fedora
[root etc]#  cp mdadm,conf /fedora/etc/

Step 7. We can either migrate our "swap" space over to a RAID pair now, or migrate to a Logical Volume where it will be part of a RAID pair as well. If you are not going to use LVM, here is the process for mirroring your swap space LABEL=SWAP1.

[root ~]#  swapoff /dev/sda7
[root ~]#  mdadm --create /dev/md7 --level=1 --raid-disks=2 /dev/sda7 /dev/sdb7
[root ~]#  mkswap -L SWAP1 /dev/md7
[root ~]#  swapoff /dev/md7

Why not just have two SWAP spaces, one on each drive? In the event of a disk failure, you want swap to be mirrored, so that the system keeps running.

Take Fresh Backup

Before we continue, now is a good time to backup your multi-boot system using the Backup-and-Restore-Using-SysRescueCD procedure. Put these in a different folder, marked "with RAID" or something, so that you can remember to keep them separate from the previous partition images. Instead of sda1, sda2, etc. you will be using md1, md2, etc.

Take a copy of every /etc/fstab from /dev/md2, /dev/md5 and /dev/md6.

Your backup.sh might contain additional lines to capture the RAID configuration settings.

cat /proc/mdstat > mdstat.txt
mdadm --assemble --scan 
mdadm --examine --scan  > md_conf.txt
mdadm --detail /dev/md* > mdadm.txt

Configure LVM2 for Logical Volumes

Now that we have multi-boot and RAID1 mirroring in place, we can create Logical Volumes. HOWTOforge.com has a nice [Beginner's Guide To LVM] if you are new to LVM.

Step 1. VolGroup00 for Library and Swap

Let's recreate the Logical Volumes that we had when we first installed XS school server. First, we'll define three mirrored devices to serve as physical space. Go into "fdisk /dev/sda" and change the types from "83" to "fd"(raid autodect) for /dev/sda8, /dev/sda9, /dev/sda10. Use "sfdisk" to copy these changes over to the second drive.

[root ~]#  fdisk /dev/sda
[root ~]#  sfdisk -d /dev/sda | /dev/sdb

Use "mdadm" to create the RAID1 pairs. This may take awhile as it needs to copy all of the blocks over. Then use "pvcreate" to indicate these can be used as physical space for a volume group.

[root ~]#  mdadm --create /dev/md8 --level=1 raid-disks=2 /dev/sda8 /dev/sdb8
[root ~]#  mdadm --create /dev/md9 --level=1 raid-disks=2 /dev/sda8 /dev/sdb9
[root ~]#  mdadm --create /dev/md10 --level=1 raid-disks=2 /dev/sda8 /dev/sdb10
[root ~]#  watch cat /proc/mdstat
[root ~]#  pvcreate /dev/md8 /dev/md9 /dev/md10
[root ~]#  pvdisplay

Define the volume group. Since XS used "VolGroup00", we will use this also. I chose to use a 16MB extent size, instead of the default 4MB, so that we have fewer extents to work with.

[root ~]#  vgcreate -s 16M VolGroup00 /dev/md8 /dev/md9 /dev/md10
[root ~]#  vgdisplay

Now let's create the logical volumes. The XS had two, LogVol00 for the /library directory, and LogVol01 for the swap space. This allows us to show the difference between lower case "-l" and upper case "-L". Lower case is in units of "extents" which we defined as 16MB, so here library is a tiny 16MB volume (we can always expand it later!!!) The -L allows you to specify the volume group in M(MB) or G(GB). We make the swap space 2G to match the amount of memory we have installed on our system.

[root ~]#  lvcreate -l 1 --name LogVol00 VolGroup00
[root ~]#  lvcreate -L 2G --name LogVol01 VolGroup00
[root ~]#  lvdisplay

Now, to activate these into the system. For Library, the initial XS CD created it empty except for a lone "users" directory. We can recreate that here. Look at how /etc/mtab was updated with this information, and copy that into /etc/fstab to have /library mounted at boot time automatically. On my system, it got remapped to /dev/mapper/VolGroup00-LogVol00.

[root ~]#  mkdir /library
[root ~]#  mount /dev/VolGroup00/LogVol00 /library
[root ~]#  mkdir /library/users
[root ~]#  vi /etc/mtab /etc/fstab

Middleware

Additional Utilities