Attempts to install a Windows driver from WinPE
by Matthew Booth
One of the things we would really like to be able to do for V2V is to
install a new driver in a Windows guest. There are a couple of reasons
for this:
* The guest may not be bootable without the driver installed, for
example because the underlying virtual hardware has changed from vmscsi
to virtio.
* If the guest can boot, the alternative is to modify the guest to run a
script on next boot. This requires making assumptions about supporting
software being installed and working correctly on the guest. Certain
environments, particularly heavily locked-down environments, make this
an unsafe assumption.
The Windows PE environment looks perfect for this task. It gives you a
very lightweight Windows OS which can be customised with additional
tools. It is specifically for doing installations. I spent Friday trying
to use it to install a driver in a guest. Here's what I tried and why it
didn't work.
Installing a driver in Windows is 'driven' by a .inf file. From my
(admittedly limited) understanding, this broadly describes:
* The files which need to be installed
* The hardware the driver is compatible with
The files, including the .inf file itself must be copied in to the
correct places. In addition, information from the .inf file must be
written to the registry. It is this last part which causes problems.
From reading documentation, it appears that a driver would normally be
installed using the SetupCopyOEMInf() library call. I wrote a simple
wrapper round this and installed it in the Windows PE image, along with
the VirtIO drivers. I booted into Windows PE and attempted to install
the driver. As you might expect, the drivers were installed into the
Windows PE image rather than the guest. I then tried setting
%systemdrive% and %systemroot% to the guest image. This appears to have
no effect. This is what makes me suspect that the process is primarily
registry driven.
I started looking around for ways of using a different registry. I
discovered the Registry Editor PE plugin to BartPE
(https://sourceforge.net/projects/regeditpe/) which allows editing the
registry of a guest. Looking at how it does this, it uses reg.exe to
load the guests's hives. I confirmed that you can do this. Unfortunately
you don't seem to be able to replace the default hives. The new hives
are loaded in a different part of the tree, and are therefore ignored.
This is as far as I've got. Still on my list are:
* Ask on various Windows mailing lists how to do this
* Investigate if the packaged .msi containing the drivers is more flexible.
* Look for other, possibly lower level, ways of replacing making a
process use a different registry.
Any and all suggestions are gratefully received,
Matt
--
Matthew Booth, RHCA, RHCSS
Red Hat Engineering, Virtualisation Team
M: +44 (0)7977 267231
GPG ID: D33C3490
GPG FPR: 3733 612D 2D05 5458 8A8A 1600 3441 EA19 D33C 3490
14 years, 5 months
Re: Attempts to install a Windows driver from WinPE
by Ayal Baron
----- "Matthew Booth" <mbooth(a)redhat.com> wrote:
> One of the things we would really like to be able to do for V2V is to
>
> install a new driver in a Windows guest. There are a couple of reasons
>
> for this:
>
> * The guest may not be bootable without the driver installed, for
> example because the underlying virtual hardware has changed from
> vmscsi
> to virtio.
>
> * If the guest can boot, the alternative is to modify the guest to run
> a
> script on next boot. This requires making assumptions about supporting
>
> software being installed and working correctly on the guest. Certain
> environments, particularly heavily locked-down environments, make this
>
> an unsafe assumption.
>
> The Windows PE environment looks perfect for this task. It gives you a
>
> very lightweight Windows OS which can be customised with additional
> tools. It is specifically for doing installations. I spent Friday
> trying
> to use it to install a driver in a guest. Here's what I tried and why
> it
> didn't work.
>
> Installing a driver in Windows is 'driven' by a .inf file. From my
> (admittedly limited) understanding, this broadly describes:
>
> * The files which need to be installed
> * The hardware the driver is compatible with
>
> The files, including the .inf file itself must be copied in to the
> correct places. In addition, information from the .inf file must be
> written to the registry. It is this last part which causes problems.
>
> From reading documentation, it appears that a driver would normally
> be
> installed using the SetupCopyOEMInf() library call. I wrote a simple
> wrapper round this and installed it in the Windows PE image, along
> with
> the VirtIO drivers. I booted into Windows PE and attempted to install
>
> the driver. As you might expect, the drivers were installed into the
> Windows PE image rather than the guest. I then tried setting
> %systemdrive% and %systemroot% to the guest image. This appears to
> have
> no effect. This is what makes me suspect that the process is primarily
>
> registry driven.
>
> I started looking around for ways of using a different registry. I
> discovered the Registry Editor PE plugin to BartPE
> (https://sourceforge.net/projects/regeditpe/) which allows editing the
>
> registry of a guest. Looking at how it does this, it uses reg.exe to
> load the guests's hives. I confirmed that you can do this.
> Unfortunately
> you don't seem to be able to replace the default hives. The new hives
>
> are loaded in a different part of the tree, and are therefore
> ignored.
This is NOT doable. The registry holds too many configurations imperative for the proper behaviour of the OS (in this case the WinPE hive). If this is at all possible (I doubt it, too many open handles and things that require booting to change as it is) Changing this would render the system unstable (if not BSOD you entirely if you could get so far).
Now that I understand what you are trying to do (and I don't mean to discourage) but I believe you are going about this the wrong way. A direct install of the driver probably does even more things which you are not aware of and are hardwired to the system drive (not through %SYSTEMDIR% and any other environment variable).
Just an example, Changing system environment variables require booting the machine (http://technet.microsoft.com/en-us/library/bb726962.aspx) this is the relevant excerpt:
Note: When you create or modify system environment variables, the changes take effect when you restart the computer. When you create or modify user environment variables, the changes take effect the next time the user logs on to the system.
In addition, if the device driver is installed under the "system" user's credentials then: you have to boot the machine after you change the environment variables (because they are loaded only once at boot and cannot be refreshed) and changing the systemdir for the entire machine and rebooting is probably not recommended.
In any event, in order to reverse engineer what is going on when the driver is installed you can probably use some "recording" applications (apps which install filter device drivers which monitor the registry and the hard drive, record changes and filter out known things which are irrelevant).
The next phase would be to inject the relevant files directly where they should be and make the relevant registry changes (for instance by loading the Hives as you did and running a script that changes the relevant entries).
The other possibility is as I mentioned before offline tools which already know how to do this (e.g. http://technet.microsoft.com/en-us/library/cc749465%28WS.10%29.aspx).
>
> This is as far as I've got. Still on my list are:
>
> * Ask on various Windows mailing lists how to do this
> * Investigate if the packaged .msi containing the drivers is more
> flexible.
> * Look for other, possibly lower level, ways of replacing making a
> process use a different registry.
>
> Any and all suggestions are gratefully received,
>
> Matt
> --
> Matthew Booth, RHCA, RHCSS
> Red Hat Engineering, Virtualisation Team
>
> M: +44 (0)7977 267231
> GPG ID: D33C3490
> GPG FPR: 3733 612D 2D05 5458 8A8A 1600 3441 EA19 D33C 3490
14 years, 5 months
[PATCH 0/9] FOR DISCUSSION ONLY: daemon error handling
by Richard W.M. Jones
The more I look at this patch, the less I like it. I would summarise
why I think it's wrong here, but it's better if you look at the
message I posted on the gnulib mailing list here first:
http://lists.gnu.org/archive/html/bug-gnulib/2009-11/msg00434.html
Directly accessing errno on Windows is wrong: you won't see the true
reasons for an error by doing that. However depending on what the
call is and whether it was replaced by gnulib, the alternatives are a
lot more complicated.
This patch series attempts to Do The Right Thing with gnulib code as
it stands at the moment (assuming my understanding of gnulib error
handling is correct -- see above link).
I feel however that it's probably better to change gnulib error
handling to be more rational.
There's also an additional complication I discovered: MinGW error
values are often wrong. So that's a MinGW bug to take to the nice
upstream folks on the MinGW project.
Rich.
--
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
libguestfs lets you edit virtual machines. Supports shell scripting,
bindings from many languages. http://et.redhat.com/~rjones/libguestfs/
See what it can do: http://et.redhat.com/~rjones/libguestfs/recipes.html
14 years, 5 months