Syslinux - Mar 2008 - Adding memdisk or similar when booting linux

Hi

Thanx guys for the help on understanding howto read dmi info!
I now have a custom boot menu for install of different linux versions based if
they have been installed before (reinstall is ok for users, first install is
not), and installquirks like if we detect odd hardware, we add install options
to redhat installer so it can install anyhow (for instance hp dc7800 who needs
pci=nommconf as kernel append for booting).

But now I need to add a driver disk to my install enviroment, and when thinking
about it, the nice method would be having syslinux create a memdisk which linux
could see when starting install, and load its drivers from there. The reason i
need a memdisk image (or similar), is since often the driver I need is a network
driver, which offcourse is a tad difficult getting from the net _After_ the
kernel have taken over.

Embedding the initrd would be an option, but would require me to change all
install initrd's for every driver I put in my driver disk, and also it would
require me to also allways modify new releases with custom initrd. But with a
memdisk approach, the only image who would need attention would be the driver
disk image alone (which actually supports multi versions)

Is there any clean approches to achive this? like starting first memdisk, load a
image to it, and then start linux (all thru pxelinux?)

--Kjetil

Hi Kjetil,

I used ISOLINUX, MEMDISK and SYSLINUX to load drivers disk for Windows using
MEMDISK in a tool I made a few years back.

What I did was:

- ISOLINUX loads MEMDISK and a gzipped floppy image and boots the A: floppy
image.
- Floppy image has SYSLINUX that in my case load another floppy image using
MEMDISK to create a virtual B: drive.
(The reason for loading MEMDISK twice in my case is because of a special case in
how the Windows installer handles B: drives with driver disks inserted.)
- The B: floppy image boots and also run SYSLINUX wich kick of a bootloader that
let me change CD and boot from the new CD inserted, im my case this is the
Windows 2k3 installation media.

You should be able to do something similar like:

- PXELINUX loads MEMDISK and a gzipped floppy image and boots the A: floppy
image. This image contain your driver disk and SYSLINUX+MEMDISK+"Floppy
bootable ROM Image".
- SYSLINUX could then load MEMDISK that in turn loads and boots the "Floppy
bootable ROM Image".
- The "Floppy bootable ROM Image" will now do another PXE boot.

You can also configure the bootfile loaded by the Floppy bootable ROM Image when
this is created. E.g overriding the bootfile specified in your DHCP config. E.g
the NIC PXE boot get the bootfile info from DHCP wich is
tftp://server/loadddisk/pxelinux.0 which load MEMDISK and then the Floppy
bootable ROM Image will load tftp://server/installer/pxelinux.0 which in turn
start the kernel/initrd for OS installation.


I would'nt call this clean, but it should work... :)


\\
Harald Jens?s

> -----Original Message-----
> From: syslinux-bounces at zytor.com [mailto:syslinux-bounces at zytor.com]
On
> Behalf Of Kjetil.Mikkelborg at kongsberg.com
> Sent: 13 March 2008 11:13
> To: SYSLINUX at zytor.com
> Subject: [syslinux] Adding memdisk or similar when booting linux
> 
> Hi
> 
> Thanx guys for the help on understanding howto read dmi info!
> I now have a custom boot menu for install of different linux versions
> based if they have been installed before (reinstall is ok for users,
> first install is not), and installquirks like if we detect odd
> hardware, we add install options to redhat installer so it can install
> anyhow (for instance hp dc7800 who needs pci=nommconf as kernel append
> for booting).
> 
> But now I need to add a driver disk to my install enviroment, and when
> thinking about it, the nice method would be having syslinux create a
> memdisk which linux could see when starting install, and load its
> drivers from there. The reason i need a memdisk image (or similar), is
> since often the driver I need is a network driver, which offcourse is a
> tad difficult getting from the net _After_ the kernel have taken over.
> 
> Embedding the initrd would be an option, but would require me to change
> all install initrd's for every driver I put in my driver disk, and also
> it would require me to also allways modify new releases with custom
> initrd. But with a memdisk approach, the only image who would need
> attention would be the driver disk image alone (which actually supports
> multi versions)
> 
> Is there any clean approches to achive this? like starting first
> memdisk, load a image to it, and then start linux (all thru pxelinux?)
> 
> --Kjetil
> _______________________________________________
> SYSLINUX mailing list
> Submissions to SYSLINUX at zytor.com
> Unsubscribe or set options at:
> http://www.zytor.com/mailman/listinfo/syslinux
> Please do not send private replies to mailing list traffic.

On Thu, 13 Mar 2008, Kjetil.Mikkelborg at kongsberg.com wrote:
> Thanx guys for the help on understanding howto read dmi info!
> I now have a custom boot menu for install of different linux versions 
> based if they have been installed before (reinstall is ok for users, 
> first install is not), and installquirks like if we detect odd hardware, 
> we add install options to redhat installer so it can install anyhow (for 
> instance hp dc7800 who needs pci=nommconf as kernel append for booting).
Kjetil,

You would do me a very big favor in showing me how you do this (or even 
beter, writing up a document for the wiki how you achieved this).

I am very interested to implement the same and I am confident that this is 
something Red Hat should be doing as well (or if not, I would like to see 
within the CentOS project what is possible)

Thanks in advance,
-- 
--   dag wieers,  dag at wieers.com,  http://dag.wieers.com/   --
[Any errors in spelling, tact or fact are transmission errors]

Ok,

First off, thnx guys for giving me some hints to how to solve my problem with
driver disks, I am not sure it will work with my setup (basicly because I relies
so much on PXE), but I will use some time to figure it out at least ;)

And for you Dag, first of, my approach of fixing this is probably due to lack of
support in redhat enterprise itself, so I can really see where your questions
comes from!
the first issue, about install/reinstall Is just i feature I needed because I
have to be 110% sure the hardware my users are using actually is well behaving
(and my company have dedicated personell for installing windows, so what the
heck, let them install linux and verify that too).

The whole concept relies on using the comboot complex.c menu system included
with syslinux, and modififying it to read from a register file located on the
tftp server. This approach is not extremly scalable, but will do for now in my
usage area.

This register file is modified by a rpm package installed by yum at first boot
(after install). (hence first install or reinstall, if entry in register file
exists, we are reinstalling). And everytime my clients are booting their client
after that, the same package has installed a init script which check to see if
it has any changes (like new IP, new hostname or alias), and contacts a server
with its updates (makes dns and friends allways updated in a dhcp enviroment).

This central server when asked for making changes, verify the changes, and
updates the tftp register file if any changes has been made. So this ensures
that if a client changes hostname, the pxe install menu actually knows what it
is called (using MAC address as system-id in the register file).

Ive also password protected override for install, so if first install (no entry
in register file), my install crew can punch in their password, register what
the client actually should be named, and start installing (post script in
anaconda uses HOSTNAME= variable in cmdline from kernel boot).


My second thing (installquirks), is actually just some of the same code reads
dmi info, and checks for product name string, and parses it with help from a
installquirks file found in pxelinux tftp server. If we find a entry which
matches whats inserted in installquirks file, then we append what that record
holds when installing. The nice thing about this, is that anaconda adds by
default every "quirk" the user use for install to grub when installing
grub, so this change is only needed on install.

The coolest thing, would offcourse be if someone had a own project for
maintaining a installquirks register on the net, so it could be read (with help
of a tiny comboot program) at install by syslinux as default. But that is a
pretty big task.

So my approaches is somewhat hackish, but really makes installing and
maintaining large installation of linux clients doable :) So now I just have to
get to the buttom of the driver disk problem!

--Kjetil

-----Original Message-----
From: syslinux-bounces at zytor.com [mailto:syslinux-bounces at zytor.com] On
Behalf Of Dag Wieers
Sent: 14. mars 2008 02:55
To: For discussion of SYSLINUX and tftp-hpa
Subject: Re: [syslinux] Adding memdisk or similar when booting linux

On Thu, 13 Mar 2008, Kjetil.Mikkelborg at kongsberg.com wrote:
> Thanx guys for the help on understanding howto read dmi info!
> I now have a custom boot menu for install of different linux versions
> based if they have been installed before (reinstall is ok for users,
> first install is not), and installquirks like if we detect odd hardware,
> we add install options to redhat installer so it can install anyhow (for
> instance hp dc7800 who needs pci=nommconf as kernel append for booting).
Kjetil,

You would do me a very big favor in showing me how you do this (or even
beter, writing up a document for the wiki how you achieved this).

I am very interested to implement the same and I am confident that this is
something Red Hat should be doing as well (or if not, I would like to see
within the CentOS project what is possible)

Thanks in advance,
--
--   dag wieers,  dag at wieers.com,  http://dag.wieers.com/   --
[Any errors in spelling, tact or fact are transmission errors]

_______________________________________________
SYSLINUX mailing list
Submissions to SYSLINUX at zytor.com
Unsubscribe or set options at:
http://www.zytor.com/mailman/listinfo/syslinux
Please do not send private replies to mailing list traffic.

Kjetil.Mikkelborg at kongsberg.com wrote:
> Hi
> 
> Thanx guys for the help on understanding howto read dmi info!
> I now have a custom boot menu for install of different linux versions based
if they have been installed before (reinstall is ok for users, first install is
not), and installquirks like if we detect odd hardware, we add install options
to redhat installer so it can install anyhow (for instance hp dc7800 who needs
pci=nommconf as kernel append for booting).
> 
> But now I need to add a driver disk to my install enviroment, and when
thinking about it, the nice method would be having syslinux create a memdisk
which linux could see when starting install, and load its drivers from there.
The reason i need a memdisk image (or similar), is since often the driver I need
is a network driver, which offcourse is a tad difficult getting from the net
_After_ the kernel have taken over.
> 
> Embedding the initrd would be an option, but would require me to change all
install initrd's for every driver I put in my driver disk, and also it would
require me to also allways modify new releases with custom initrd. But with a
memdisk approach, the only image who would need attention would be the driver
disk image alone (which actually supports multi versions)
> 
> Is there any clean approches to achive this? like starting first memdisk,
load a image to it, and then start linux (all thru pxelinux?)
> 
Writing a COM32 module to construct an initramfs on the fly might be the 
best of all worlds.  That way the COM32 module can download the specific 
drivers it needs.

The library support for that is already there.

	-hpa

>Kjetil.Mikkelborg at kongsberg.com wrote:
>>
>> Is there anyway i can keep redhats initramfs image, and load a second
one?
>>
>> If not, I really would need to ensure that the initramfs could hold
both redhats data, and my driverdisk data. But still this is much cleaner than
jumping from memdisk >to syslinux and so on. How does actually initramfs
handle this? can it be dynamicly enlarged? with this scenario, the initramfs
could with no problem be twice and tripple >the original size.
>>
>> And if i really can have a second initramfs, how do i access it?
>>
>
>If you load multiple initramfs images, then they overlay; the resulting
>filesystem contains both files.
>
>        -hpa
Ok, thanks, but..

Ive tried using linux.c as an example, but I am in no way capable of adding
files with this simple code

size_t kernel_len, dd_len;
void *kernel_data, *dd_data;
struct initramfs *initramfs;

loadfile("path to vmlinuz on tftp
server",&kernel_data,&kernel_len);
loadfile("Path to dd.img file on tftp
server",&dd_data,&dd_len);
initramfs = initramfs_init();
initramfs_load_archive(initramfs,"path to redhat enterprise 4 initrd.img
file on tftp server");
initramfs_add_file(initramfs,dd_data,dd_len,dd_len,"/dd.img",0,0755);
syslinux_boot-linux(kernel-data,kernel_len,initramfs,"append
string",0xffff,0);

is there any more init I have to do to add files? and,  is it really right to
use dd_len,dd_len as arguments (whats the first len since the second is for how
much data we add).

--Kjetil