Imaging Guide

From Physiki
Revision as of 02:04, 15 February 2010 by Ahubl (Talk | contribs)
Jump to: navigation, search

Decprecated incarnations of tablet imaging walkthroughs are located at the Deprecated Imaging Guides.

Contents

CloneZilla Implementation

There is some patchy original documentation at the Clonezilla Guide and DRBL homepage.

Import the repo key to your keyring

wget http://drbl.sourceforge.net/GPG-KEY-DRBL -O - | sudo apt-key add -

Edit sources.list

Add the deb line (below) using System > Software Sources, or create a new file /etc/apt/sources.list.d/clonezilla.list and add the following entry:

(This also works if you add this line to /etc/apt/sources.list)

Now you can install clonezilla and drbl:

sudo apt-get update && sudo apt-get install clonezilla drbl

Install DRBL, step 1

There are two commands you use to install and activate CloneZilla. The first is

sudo /opt/drbl/sbin/drblsrv -i

Note: There is a space before the "-i". The first time you run it, this will also connect to the Internet and download a bunch of packages. Decline installing packages for “Diskless Linux”, as we don't run any thin clients. Other than that, the questions are pretty straightforward. Make sure you have two network interfaces up when you run this command – it will detect one as “Internet” and one as “private”. If it doesn't detect two interfaces, the clients won't get a DHCP lease (for a centralized system like the new TICC server setup this is fine). Don't bother to configure your clients – letting the range in dhcp.conf control leases is fine. It warns you about this, but we don't care.

These are the answers I used:

  1. Install the network installation boot images
  2. I did not use the serial console output on the client computer. Use this only if you know what you are doing.
  3. I used the same architecture as my server, but I knew that the clients and my server were all x86's.
  4. I upgraded the OS to make sure my server wass up to date (you don't have to do this).
  5. I used the same kernel as was installed on my server.

Install DRBL, step 2

The second step creates the configuration for all the clients:

sudo /opt/drbl/sbin/drblpush -i

The default options are again mostly acceptable. We want “no diskless Linux” and “Clonezilla box mode”. Some options give information about “increased requirements on server for more clients”. These options are never what we want – we don't care about persistent client configuration.

These are the options I used:

  1. I don't have a DNS name (I'm doing this at home), but if you do, feel free to input it; I used the default "drbl.name".
  2. I don't have an NIS/YP domain name either, so I used the default "penguinzilla".
  3. I used the default hostname prefix (my server hostname) as I didn't want to specify a unique hostname for all of the IPs for all of my tablets in /opt/drbl/conf/client-ip-hostname.
  4. My server is also my desktop (and is connected wirelessly), so the connection I use for the internet is wlan0 (DRBL wanted to use eth0). If you are running Gnome, you can tell which connection to use here by right clicking on the NetworkManager icon, selecting "Connection Information" and looking at the Interface line (mine is "802.11 WiFi (wlan0)", you want whatever is in the parentheses).
  5. I do not care to give my tablets the same IP addresses each time, so I did not collect them. Doing this is probably a labor-intensive task, but it is probably worth it for larger organizations.
  6. Again, I had to tell DRBL that I didn't want to give the same IP address to each MAC address it connected to.
  7. My server's IP is 192.168.0.1, so I want to start the clients IP addresses at 192.168.0.2, so I entered in a 2.
  8. I may have a maximum of three (3) tablets to image for this walkthrough, so I entered in 3. For a larger organization, this may change as new tablets are brought in and older tablets are broken/stolen/lost (but hopefully not).
  9. I accepted the IP settings, as it wanted to give out the IP addresses 192.168.0.2 through 192.168.0.4.
  10. DRBL then printed out the layout for my environment, which looked like:
              NIC     NIC IP                           Clients
+----------------------------------------+
|             DRBL SERVER                |
|      +-- [wlan0] 192.168.11.2          +- to WAN
|                                        |
|      +-- [eth0] 192.168.0.1            +- to clients group 0 [ 3 clients, their IP
|                                        |              from 192.168.0.2 - 192.168.0.4]
+----------------------------------------+
  1. I hit enter to confirm the network layout.
  2. I chose 0 to give each client its own NFS-based /etc and /var.
  3. I chose 1 (Clonezilla box mode) so the image loaded by the clients will not be saved after shutdown.
  4. I use /images to store the hard drive images for clients.
  5. Using a swap partition or file is useful so the client has more memory to use. With my computers, I don't think this makes a difference, but it couldn't hurt, as it only writes to swap partitions or writable filesystems.
  6. I used the 128MB as the size for the swap space.
  7. I like GUIs, so I used the Graphic mode (X window system) option.
  8. I used the auto login so I don't have to manually login to every client to image.
  9. I did not use a random password for auto login accounts.
  10. I entered in the password I wanted to use "password".
  11. Retype the password you want to use.
  12. I just let root's password be copied from the server to the clients (not the best security, but I'm on my own LAN), so I answered no.
  13. I don't want a bootup password for PXE, although this would be useful for larger organizations.
  14. I do want to set the boot prompt for the clients.
  15. I used the default 70 for the next option. I think the option is in tenths of a second, so 70 deciseconds is 0.70 seconds. The prompt is a little confusing (and I could have interpreted it wrong, we will see).
  16. I don't want to use the graphic background for the PXE menu, as that adds to the complexity of the booting.
  17. I don't need audio, cdrom, floppy, video, and plugdev used for the clients, as it will just be for installing an image.
  18. My clients don't need two IPs.
  19. My clients also do not need to run in terminal mode.
  20. My clients don't need internet access.
  21. I don't need to keep the old settings of existing DRBL clients.
  22. My server supported NFS over TCP, so I just had to hit enter.
  23. And hit enter to start the process. DRBL/Clonezilla will run for awhile, and then prompt you to PXE-boot your clients.
  24. PXE-boot clients into Gnome environment. They can now do whatever they like.

Setup NFS for All Clients

DRBL will configure NFS "incorrectly" for booting from USB clients controlled by a different DHCP server. Provided that your NFS server doesn't do anything but serve CloneZilla, you simply replace your /etc/exports with the following:

/tftpboot/node_root *(ro,sync,async,no_root_squash,subtree_check)
/usr *(ro,sync,async,no_root_squash,subtree_check)
/opt *(ro,sync,async,no_root_squash,subtree_check)
/home *(rw,sync,async,no_root_squash,no_subtree_check)
/var/spool/mail *(rw,sync,async,root_squash,no_subtree_check)
/images *(rw,sync,async,no_root_squash,no_subtree_check)

NFS Configuration Fix

CloneZilla will run into extreme difficulty if you boot too many systems at once unless you change the configuration settings for NFS. These settings are stored in the file /etc/default/nfs-kernel-server, the important ones to change are the number of simultaneous NFS servers:

# Number of servers to start up
RPCNFSDCOUNT=1000

and the number of simultaneous RPC mounts:

RPCMOUNTDOPTS="--num-threads=50"

A word

It is important to test out all configurations on a small subset (at least one machine of every model you have) before applying the changes to the rest of the machines you will be imaging. Testing is always a good idea, and if you screw up it means you only have to fix a small number of tablets instead of a large number. With that said, read on.

Get image from source tablet

All the imaging parameters are accessed by running the command:

sudo /opt/drbl/sbin/dcs
  1. Select all clients (unless you have a list of the individual tablet IPs or MAC addresses and only want to image a few) and choose the clonezilla-start mode.
  2. I then chose to use the expert interface, and the "save-disk" mode to save the entire disk of a tablet.
  3. Setting the image name now is useful, but you can also choose to set it later (when you boot the tablet from using PXE or a pre-loaded USB drive). Whatever you choose, pick a good name, I suggest adding a small description onto the end of its default name (which is the current date).
  4. Next, choose the disk/partition *device name* you want to save, eg: "sda" (without the quotes) for /dev/sda (the first disk on the tablet).
  5. You can tell Clonezilla to use a priority for cloning programs. TICC does not have a Windows partition on our tablets (we don't run Windows on the tablets), so I chose the option "-q2 Priority: partclone > partimage > dd" which (I think) means that partclone has the highest priority, then partimage, and finally dd. dd is the safest option if you have exotic types of partitions, but it is inefficient as it copies the entire disk, even if there is no data. It makes an exact bit-for-bit copy of drives (which can be very useful).
    If you have a Windows NTFS partition (eg: if you run Windows XP or greater), you may want to consider using the "-q Priority: ntfsclone > partimage > dd" option, which means that it will use ntfsclone for ntfs partitions, then partimage for the rest, and dd for anything it doesn't recognize (I think). If you don't know, this is probably the safest option, but it might not be the fastest (I honestly don't know).
  6. The next step was choosing options for cloning. The only option that was enabled by default for me was -j2 (Clone the hidden data between MBR and 1st partition), but I also turned on -c (Client waits for confirmation before cloning) so I would have to start the cloning process on the client. The defaults look reasonable, I suggest leaving them.
  7. I then chose to poweroff the clients when the cloning process was over, just to give myself (and the switch and my server) a little breathing room between copying the source image to my server and copying the image to the rest of the client machines.
  8. I chose the default compression scheme (-z1p Use parallel gzip compression...). Choose otherwise if you know what you're doing (and read the warnings).
  9. I chose the "1000000" option so the image wouldn't be split up into multiple files.

Clonezilla pestered me to setup the IP addresses of the machines so it could give out the same address each time (it looks at their MAC addresses), but since I'm on my own LAN with only my server and my clients, I don't care. If you have other machines on the network though, it might be a good idea to get the MAC addresses of the tablets you want to image and add them to dhcpd.conf as it suggests.

The command finished running, so I booted the source client tablet with PXE-boot and I was presented with an option screen. I could boot:

  • Ubuntu 9.10 Linux (DRBL mode, mostly local resources)
  • Clonezilla: save disk sda as image <image-name> (image-name from server setup above)
  • Local operating system (if available)
  • Memory test using Memtest86+

I booted into the Clonezilla mode and watched it boot up. It did run into a page table error, but that didn't seem to matter. It prompted me to continue, and I did, and it seemed to work but came up with "Something went wrong!!!", I hit enter to continue and the machine shutdown as I had set it up to do. Oh well, I guess I won't be investigating that error. On the up side, the image found its way to my server and it looked okay.

Broadcast image

After you receive the image, you have to put it back out to the others. Rerun the imaging mode selection program:

sudo /opt/drbl/sbin/dcs

Now select the option “clonezilla-start” then “clonezilla-restore-disk”. When you are broadcasting an image, you have two options. You can wait for a certain amount of time or a certain number of clients. I suggest the “number of clients” option, as it should prevent the problem we've experienced with our dd/udpcast script of missing one client and having to reboot all the clients.

Da-da!

That's it! Our test run showed a sustained transfer speed of about 1GB of data/minute, which may or may not happen on a regular basis.

Misc. Comments

Each interactive command ends with a statement like “Next time you want to do this, run the following command”, giving you a method to bypass the interactive part by using a huge command line instead. I was never paying enough attention to capture these, but we should get them and put them up on the wiki to save some time.

Good luck!


CloneZilla Netboot without DHCP

In the past TICC used CloneZilla from a local computer, requiring us to setup some computer to run CloneZilla (above process) every time we imaged. The reason we had to do this is that we do not have control over the DHCP server on the network, so we cannot use PXE to boot our clients if we connect them to the network. To work around this we have been developing a DHCP-server-less configuration of CloneZilla that allows us to perform our imaging. This build has 2 components: a special netbooting USB flash disk and modifications to our CloneZilla installation.

Netboot USB Flash Disk

  • started with hardy live cd
  • modified cd image for usb (find link)
  • added netboot to usb (find link)
  • tweaked the syslinux menu to have special clonezilla options
  • added "ifconfig lo up" so DRBL doesn't puke

CloneZilla Modifications

changes to /tftpboot/node_root/sbin/init:

  • changed "nfsserver" to not "ip match"
  • added early mount for /opt
  • added ssh routine to copy the template for our IP

changes to /tftpboot/node_root/:

  • placed ssh in /usr/bin
  • placed a bunch of libraries need by ssh in /lib (find list)

added a user and group "clonezilla"

changes to /tftpboot/nodes/:

  • copied the original IP to "template"
  • applied "chmod -R g+rw template"
  • applied g+x to all directories (find command)
  • applied "chown -R "root:clonezilla template"

The Grand Unified Tablet Build

TICC is now using a single build to image all of our tablets. This build requires a script to handle the differences between the tablets, as of Intrepid Ibex we only require separate Xorg.conf files and a separate "pen button" configuration for each of the supported models. The script currently supports the tc1100, the tc4200, the tc4400, and the 2710p tablets. The GUTB scripts are now available in the TICC Personal Package Archive on Launchpad. You can install this PPA by going to the System -> Administration -> Software Sources menu and adding the source "ppa:csm-ticc/csm-ticc-ppa". After installing the PPA you can install the GUTB using the command:

sudo apt-get install gutb

Once installed, the script will ensure on boot that the appropriate settings are applied (allowing the script to be installed once and imaged to all the other tablets).

Detection of Tablet Model

A simple hash of the CPU information is used to create a unique identifier for each tablet. The script '/scripts/cpuhash.sh' can be used to generate this id, this script just runs the command:

cat /proc/cpuinfo | grep -v bogomips | grep -v "cpu MHz" | md5sum | awk '{ print $1 }'

This command takes the CPU identification and strips out the part that varies from time to time (bogomips and "cpu MHz") and creates an md5sum of the remaining data. It is important to note that because of the technique used to detect the model, more than one CPU can correspond to a particular tablet model. To add a hash to the list so that it correspond to a particular model you must tack on the result of '/scripts/cpuhash.sh' to the file '/scripts/hashes.sh'. This file is formatted 'addHash <hash> <name>'; for example, the tc1100 entry appears as:

        # CPU Hash                       # Tablet Name
addHash d4c66c8352264f0fd3a70935ad43dacc tc1100

Differences in the Wacom Driver

The tc1100 and tc4200 have two major differences in how they utilize the Wacom pen: the tc1100 has no eraser and the side-buttons have different IDs on the two tablets. To resolve the eraser issue, there is a separate xorg.conf for each tablet. To handle the button codes being different there is a separate folder of scripts for each pen button for the two tablets, when the tablet starts up a symbolic link is made to the appropriate script folder. It used to be necessary to patch the wacom driver, however, on the new Ubuntu 9.10 these changes are included so it is only necessary to setup the scripts with a small profile file configuring the buttons. A ".profile" is simply added to the user's home folder containing the instructions to convert the button clicks into key presses:

xsetwacom set stylus Button30 "CORE KEY SHIFT F1"
xsetwacom set stylus Button31 "CORE KEY SHIFT F2"
xsetwacom set stylus Button32 "CORE KEY SHIFT F3"

Differences in the Video Driver

The tc1100 uses nVidia's proprietary video card driver while the tc4200 uses the open-source Intel i810 driver. Due to this issue there is a separate xorg.conf for each tablet. While the tablets will work without these changes, without xorg.conf you will need to reconfigure the graphics driver at every boot.

Important things to remember in future builds

Don't forget to install:

  • vpnc
  • kile
  • Mathematica
  • LabVIEW

Make a set of key-bindings for the "fn" keys.

Personal tools
Namespaces
Variants
Actions
Navigation
Toolbox