Re: [uml-user] Steps to Build FS and deploy using systemd?

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

First, please copy the mailing list.
Although I have made these filesystems,
1) I don't have all the answers
2) Others may benefit from the answers I do have

First one: if the guest spews out "Out of memory: Kill process...", then
you obviously need more memory. More info here:
http://uml.devloop.org.uk/howto.html

On 02/02/2013 02:54 AM, Tony Su wrote:
> Thsx Antoine,
> Perhaps the most helpful to me would be if you looked at the
> description of which kernels and fs I tried which did not work for me
> but worked for you.
I have better things to do with my time, but here goes:

> In fact, here are the specific files I'm working with, I've tried
> every combination of kernel and fs
> 
> # ls -l  ../../UML64/
> total 5734304
> -rw-r--r-- 1 tony users    4194304 Jan 29 19:50 BusyBox-1.13.2-amd64-root_fs
Works fine:
./kernel64-3.7.5 ubda=./BusyBox-1.13.2-amd64-root_fs
(...)
init: can't log to /dev/tty5
starting pid 977, tty '': '/sbin/getty -L ttyS0 9600 vt100'
#

> -rw-r--r-- 1 tony users 1073741824 Jan 29 20:11 Debian-Wheezy-AMD64-root_fs
This one has problems looking up ROOT by label but eventually boots ok:
 ./kernel64-3.7.5 ubda=./Debian-Wheezy-AMD64-root_fs mem=256m
(...)
Debian GNU/Linux wheezy/sid (none) tty0
(none) login:

> -rw-r--r-- 1 tony users 2147483648 Jan 29 19:50 Fedora17-AMD64-root_fs
No login shell but does boot:
./kernel64-3.7.5 ubda=./Fedora17-AMD64-root_fs mem=256m
(...)
[DEPEND] Dependency failed for Login Service.
(...)

> -rwxr-xr-x 1 tony users   97590904 Jan 29 17:44 kernel64-3.7.4
> -rwxr-xr-x 1 tony users   25459681 Jan 29 17:43 linux-2.6.24-x86_64
No idea where you got this from. It's so old it's not funny.
If you are using outdated kernels with recent distros, you are really
looking for trouble and should be ready to pick up the pieces yourself.

> -rw-r--r-- 1 tony users 2147483648 Jan 29 17:44 OpenSuse-12.1-amd64-root_fs
Works fine:
./kernel64-3.7.5 ubda=./OpenSuse-12.1-amd64-root_fs
(...)
[    0.250000] VFS: Mounted root (ext4 filesystem) readonly on device 98:0.
sh: cannot set terminal process group (-1): Inappropriate ioctl for device
sh: no job control in this shell
sh-4.2#

And here's a few more for good measure:
* Debian Squeeze works fine:
./kernel64-3.7.5 ubda=./Debian-Squeeze-AMD64-root_fs mem=256m
(...)
[   17.520000] getmaster - no usable host pty devices
Debian GNU/Linux 6.0 (none) tty0
(none) login:

* CentOS 6.x works fine:
./kernel64-3.7.5 ubda=./CentOS6.x-AMD64-root_fs mem=256m
(...)
CentOS release 6.2 (Final)
Kernel 3.7.5 on an x86_64
localhost login:

* DSL (32-bit) works fine:
./kernel32-3.7.5 ubda=./DSL-4.4-root_fs
(...)
CONTROL-D will exit from this shell and REBOOT the system.
root@console[/]#

* Ubuntu Precise: some issues running in UML as per the notes on website.

> For your inspection two of my attempts
> kernel 2.6
> openSUSE 12.1
> http://pastebin.com/cRD6pfHS
> 
> kernel 2.6
> Fedora 17
> http://pastebin.com/7jJnGn6z
Have you tried a more recent kernel? The only two logs you provide use
what looks like a custom made kernel, and a very old one at that.
Now, guess where I think the problem comes from?
ie: any kernel that is so old that it doesn't understand ext3/ext4 is
not going to work with those filesystems that use ext4.

And when you use the mailing list, don't use pastebin, or make sure the
files don't ever timeout so others can still see them in the future.

> If any combination of the above works for you, or if you can suggest a
> better combination, it's appreciated.
> Thing is, there is rarely the same error in any kernel/fs combination,
> the only thing I've found common is that I haven't yet found a
> combination that works.
See above, you're clearly doing something wrong.
The only FS I found which misbehaves and isn't documented on the website
is Fedora 17, where the login shell does not show up as expected.

No word on your "research" of systemd vs others host setup then?

Cheers
Antoine

> 
> Thx,
> 
> Tony
> 
> On Fri, Feb 1, 2013 at 12:40 AM, Antoine Martin <an...@na...> wrote:
>> On 02/01/2013 01:32 AM, Tony Su wrote:
>>> Antoine,
>>> Thx a bunch for the responses...
>>>
>>> 1. Yeah, I had a mind-slip. Yup, chmod, not chroot. So, answered...
>>> Doesn't make sense.
>> You haven't said where you needed it.
>> The only place you need chmod for is the kernel after downloading it
>> And that's pretty obvious and covered in the UML specific instructions.
>>
>>> 2. By differentiating UML vs KVM, suggests that there are differences
>>> for those two uses, but I'll rely on your opinion that's not the case.
>>> I guess only devloop.uk.org can say why they defined the downloads
>>> that way.
>> There are differences in uses, with some documented in the notes.
>> But the root_fs is the same.
>> UML is a kernel, KVM can also be made to use a specific kernel.
>> They are very different things.
>>
>>> 3. Creating a matched pair of kernel and bootstrap fs seems to be
>>> getting problematic as distros vary and how they interact with the
>>> kernel. In fact, after testing most every possible combination of the
>>> suggested 64-bit kernels by the UML homepage at kernel.devloop.uk.org,
>>> none of them work and all of the few 32-bit don't as well, all failing
>>> for a number of reasons.
>> "none of them work"
>> You must be doing something wrong. I've re-tested a few to check (don't
>> have time to do them all), and got to the login/shell on most of them.
>> It's possible, even likely, that some have problems giving you a login
>> prompt. This is usually noted on the page (see notes 3 and 4):
>> http://fs.devloop.org.uk/
>> If that's the case and the web page is inaccurate, please provide
>> precise information of what you tried so that I can update the web page.
>> Saying "it does not work" is just not helpful at all.
>>
>>> So, aside from creating a fs beforehand,
>>> typical virt technologies like KVM, VMware, Xen, Hyper-V, VirtualBox,
>>> etc all support and primarily recommend creating an empty backing file
>>> but using standard OS installation routines to configure and populate
>>> the filesystem in the backing file by simply pointing the VM to a
>>> virtual CDROM device. Because the distros have all devoted so much
>>> time to making their installations bug free, it can be very reliable
>>> creating a complete VM instead of matching an arbitrary kernel with an
>>> arbitrary fs.
>> Feel free to go down that route.
>> I am just not interested in that at all. The filesystems I make are
>> generated from automated scripts.
>>
>>> 4. Maybe things like systemd vs systemvinit might not matter much
>>> within the VM, but from what I've researched so far can be very
>>> relevant configuring the VM environment on the Host.
>> I have no idea why that would be. Care to share your "research"?
>> Why would the host care what init system you run in the guest? Or for
>> that matter, any other system/userland app.
>>
>>> addemdum notes.
>>> no root fs found - I am led to believe that the rootfs should be
>>> automatically found if the fs file was built properly and should not
>>> need to be specified in the launch parameters(and can this even be
>>> done with launch parameters? Shouldn't the VM "just work" by simply
>>> executing the following
>>>
>>> ./<kernel> ubda=<fs>
>> It should, generally.
>> passing root=/dev/ubda may help if I've missed somehing.
>>
>>> kernel too old - I suppose that error should be clear to resolve, that
>>> kernel just can't be used with that particular fs
>>>
>>> unable to resolve LABEL=ROOT - Like "no root fs found" I again am
>>> somewhat surprised that if the fs file was created properly that this
>>> should need to be defined in the launch parameters. For both this and
>>> the above shouldn't these things be already defined in places like the
>>> fstab of the bootstrap fs?
>> This message is because it *is* in the fstab, and either the filesystem
>> is not labelled correctly, it is not found, or the label lookup does not
>> work, etc..
>> Again, you don't provide any details, so I can't help you.
>>
>> Antoine
>>
>>>
>>> On Wed, Jan 30, 2013 at 6:39 PM, Antoine Martin <an...@na...> wrote:
>>>> On 01/31/2013 07:53 AM, Tony Su wrote:
>>>>> Hello,
>>>>>
>>>>> Actually, 4 questions...
>>>>>
>>>>> 1. I noticed the standard procedure for preparing and launching
>>>>> including applying chroot 755 to the backing file. Is this really
>>>>> necessary? I should think that any file system that is installed into
>>>>> a file should already be jailed or otherwise restricted from accessing
>>>>> the Host computer (or any other) file system, or am I mistaken?
>>>> chroot 755? no idea what you mean. pointers please.
>>>> chmod maybe, but where?
>>>>
>>>>> 2. I was surprised to see that the downloads at fs.devloop.uk
>>>>> specified the backing file is prepared differently for UML vs KVM.
>>>> It's not.
>>>> The files are the same, it says you may need to do extra stuff to get it
>>>> to work properly. What the "stuff" is, depends on what you are trying t do.
>>>>
>>>>> I've been looking, have not yet found what those differences should be
>>>>> and I don't see anything based on the steps used to prepare a UML
>>>>> backing file.
>>>>>
>>>>> 3. I don't see any description anywhere where a CDROM virtual device
>>>>> can be attached to an empty backing file, then run a standard distro
>>>>> "install." Can this work instead of creating backing files which I
>>>>> suspect may cause problems(like whether systemd or sysvinit is used by
>>>>> the distro)?
>>>> I don't understand the question.
>>>>
>>>>> 4. As the subject line describes, because I have run into a variety of
>>>>> problems using the two 64-bit kernels on the UML homepage in
>>>>> combination with the various 64-bit fs at http://fs.devloop.org.uk/,
>>>>> I'm looking to build my own fs (if you want slightly more info on the
>>>>> problems, the description is after my closing salutation). Because I'm
>>>>> running one of the distros that is using systemd, I'd like to take
>>>>> advantage of the relevant features using systemd. I found the
>>>>> following propsed procedure
>>>> Systemd, init, openrc, whatever. Not particularly relevant here.
>>>> Not a UML issue.
>>>>
>>>> (snip)
>>>>
>>>>> Addendum:
>>>>> I may d/l the 32-bit kernels and fs for testing but so far I have not
>>>>> been able to launch the following 64-bit fs from fs.devloop.org.uk
>>>>>
>>>>> openSUSE 12.1 (No root fs found)
>>>> "No root fs found", have you tried telling UML where the root fs is?
>>>>> Fedora 17 (No console appears as I assumed should happen)
>>>> Probably should, feel free to share a solution if you find one.
>>>>> Debian Wheezy (kernel too old, other kernel "unable to resolve LABEL=ROOT)
>>>> "kernel too old" - what do you mean?
>>>> "unable to resolve" - have you tried telling it where to find it?
>>>>
>>>>> So, bottom line is that at least the 64-bit resources won't work...
>>>> Some of them don't, yes. Some work with UML, some with KVM, etc..
>>>>
>>>> Cheers
>>>> Antoine
>>>>
>>>>> Have read the "Troubleshooting" page and none of those issues apply to
>>>>> these problems
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> Everyone hates slow websites. So do we.
>>>>> Make your web apps faster with AppDynamics
>>>>> Download AppDynamics Lite for free today:
>>>>> http://p.sf.net/sfu/appdyn_d2d_jan
>>>>> _______________________________________________
>>>>> User-mode-linux-user mailing list
>>>>> Use...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>>>>>
>>>>
>>>>
>>>> ------------------------------------------------------------------------------
>>>> Everyone hates slow websites. So do we.
>>>> Make your web apps faster with AppDynamics
>>>> Download AppDynamics Lite for free today:
>>>> http://p.sf.net/sfu/appdyn_d2d_jan
>>>> _______________________________________________
>>>> User-mode-linux-user mailing list
>>>> Use...@li...
>>>> https://lists.sourceforge.net/lists/listinfo/user-mode-linux-user
>>