Argh, you mean I had a huge money spinner and I threw it away! Oh well.
Anyway, back to reality.
I agree that splitting the UML boot process in two will allow the UML
owner to upgrade the UML kernel from inside the UML node. The boot
loader (e.g. grub) would live in host space, the UML kernel would be
stored in the UML filesystem).
However is such a scheme also workable from a UML sandbox point of view.
It's the job of the UML application (in host space) to stop processes
inside the UML node from breaking out into host space.
If a UML user is allowed to recompile the UML kernel (and hence a host
space application), then they could change it so that it no longer
protected the host from dangerous UML processes.
Also, Christian mentioned letting UML access the host file system where
the "linux" application is stored (via hostfs). If a UML user
recompiles UML linux and updates the "linux" file stored on the host and
then reboots, will UML pick up the new application?
Unless written explicitly to exec() the new program (stored in the same
place as the original), would UML not reboot to the same state as
before? When you update a program file, existing processes exec()ed
from that file don't change until stopped and reloaded.
I suppose that could easily be achieved by wrapping "linux" in a shell.
The UML user could issue "halt", allowing UML to completely finish,
then the shell script could start the new instance of UML.
Ulf Bartelt wrote:
> Am Mit, 2003-09-24 um 10.59 schrieb Paul Fee:
>>Yes the Linux kernel + UML patch can be compiled inside the UML node,
>>however how can a user trapped inside UML boot the new UML kernel?
> Do it like the BIOS does... launch a loader...
> Use an UMLkernel as grub substitute to grab the kernel from the UML's FS
> if it is partitioned or use debugfs if it is partitionless...
>>You could argue the case that a hosting company might want to deliver
>>the complete virtual linux box experience, including upgrading your own
> Mega big point!
> I would not spend a cent for an uml system somewhere when I cannot
> compile my own kernel...
> And from the me-as-uml-provider-view: The more problems the user can
> solve on his own, the longer will my phone stay quiet!
>>I don't know if there's any desire for anyone to do this (I don't need
>>it), but that's how I understood the statement.
> I strongly believe that having a bootloader for UMLs making this
> possible is the best way for the uml provider and the user and am
> building such a solution here.
> My thoughts even go into the direction of having kinda uml-netboot-rom
> too... (to be patented soon }:-) )
> I'll cry out loud here when I have some things worth to be shown...