Libguestfs - Mar 2020 - [PATCH v3] windows: delay installation of qemu-ga MSI

Instead of running firstboot script during early boot schedule a task
delayed for 2 minutes.

During the first boot, after virt-v2v conversion, Windows installs the
drivers injected by virt-v2v. When this installation is finished
Windows enforces some kind of internal reboot. This unfortunately
terminates any running firstboot scripts thus killing the installation
of qemu-ga MSI.

This is just a best-effort mitigation. It can still happen (e.g. with
slow disk drives) that the drivers are not yet installed when the
delayed installation starts. On the other hand we cannot delay it too
much otherwise we risk that the users logs in and will be doing some
work when the MSI installation starts. After MSI installation finishes
the VM needs to be rebooted which would be annoying if that would happen
under users hands. Although this is not a best fix (that may come later
as it is more complex, e.g. introducing waiting mechanism), the delay as
it is defined works in most cases. And it dramaticaly improves the
situations -- originaly I experienced more than 90% failure rate.

Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
---
 common                 |  2 +-
 v2v/convert_windows.ml | 12 ++++--------
 2 files changed, 5 insertions(+), 9 deletions(-)

diff --git a/common b/common
index ea10827b..5371257c 160000
--- a/common
+++ b/common
@@ -1 +1 @@
-Subproject commit ea10827b4cfb3cfe5f782421c01d2902e5f73f90
+Subproject commit 5371257c3cf27fb09d5f2e31ba378b0e6ccf5df6
diff --git a/v2v/convert_windows.ml b/v2v/convert_windows.ml
index 0fda1d4e..bed5989a 100644
--- a/v2v/convert_windows.ml
+++ b/v2v/convert_windows.ml
@@ -429,14 +429,10 @@ popd
    List.iter (
      fun msi_path ->
        let fb_script = "\
-echo Installing qemu-ga from " ^ msi_path ^ "
-\"\\" ^ msi_path ^ "\" /norestart /qn /l+*vx
\"%~dpn0.log\"
-set elvl=!errorlevel!
-echo Done installing qemu-ga error_level=!elvl!
-if !elvl! == 0 (
-  echo Restarting Windows...
-  shutdown /r /f /c \"rebooted by firstboot script\"
-)
+echo Removing any previously scheduled qemu-ga installation
+schtasks.exe /Delete /TN Firstboot-qemu-ga /F
+echo Scheduling delayed installation of qemu-ga from " ^ msi_path ^ "
+powershell.exe -command \"$d = (get-date).AddSeconds(120); schtasks.exe
/Create /SC ONCE /ST $d.ToString('HH:mm') /SD
$d.ToString('MM/dd/yyyy') /RU SYSTEM /TN Firstboot-qemu-ga /TR
\\\"C:\\" ^ msi_path ^ " /forcerestart /qn /l+*vx C:\\" ^
msi_path ^ ".log\\\"\"
 " in
       Firstboot.add_firstboot_script g inspect.i_root
         ("install " ^ msi_path) fb_script;
-- 
2.25.0

On Tue, Mar 03, 2020 at 03:45:49PM +0100, Tomáš Golembiovský
wrote:
> Instead of running firstboot script during early boot schedule a task
> delayed for 2 minutes.
> 
> During the first boot, after virt-v2v conversion, Windows installs the
> drivers injected by virt-v2v. When this installation is finished
> Windows enforces some kind of internal reboot. This unfortunately
> terminates any running firstboot scripts thus killing the installation
> of qemu-ga MSI.
> 
> This is just a best-effort mitigation. It can still happen (e.g. with
> slow disk drives) that the drivers are not yet installed when the
> delayed installation starts. On the other hand we cannot delay it too
> much otherwise we risk that the users logs in and will be doing some
> work when the MSI installation starts. After MSI installation finishes
> the VM needs to be rebooted which would be annoying if that would happen
> under users hands. Although this is not a best fix (that may come later
> as it is more complex, e.g. introducing waiting mechanism), the delay as
> it is defined works in most cases. And it dramaticaly improves the
> situations -- originaly I experienced more than 90% failure rate.
> 
> Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
> ---
>  common                 |  2 +-
>  v2v/convert_windows.ml | 12 ++++--------
>  2 files changed, 5 insertions(+), 9 deletions(-)
> 
> diff --git a/common b/common
> index ea10827b..5371257c 160000
> --- a/common
> +++ b/common
> @@ -1 +1 @@
> -Subproject commit ea10827b4cfb3cfe5f782421c01d2902e5f73f90
> +Subproject commit 5371257c3cf27fb09d5f2e31ba378b0e6ccf5df6
> diff --git a/v2v/convert_windows.ml b/v2v/convert_windows.ml
> index 0fda1d4e..bed5989a 100644
> --- a/v2v/convert_windows.ml
> +++ b/v2v/convert_windows.ml
> @@ -429,14 +429,10 @@ popd
>     List.iter (
>       fun msi_path ->
>         let fb_script = "\
> -echo Installing qemu-ga from " ^ msi_path ^ "
> -\"\\" ^ msi_path ^ "\" /norestart /qn /l+*vx
\"%~dpn0.log\"
> -set elvl=!errorlevel!
> -echo Done installing qemu-ga error_level=!elvl!
> -if !elvl! == 0 (
> -  echo Restarting Windows...
> -  shutdown /r /f /c \"rebooted by firstboot script\"
> -)
> +echo Removing any previously scheduled qemu-ga installation
> +schtasks.exe /Delete /TN Firstboot-qemu-ga /F
> +echo Scheduling delayed installation of qemu-ga from " ^ msi_path ^
"
> +powershell.exe -command \"$d = (get-date).AddSeconds(120);
schtasks.exe /Create /SC ONCE /ST $d.ToString('HH:mm') /SD
$d.ToString('MM/dd/yyyy') /RU SYSTEM /TN Firstboot-qemu-ga /TR
\\\"C:\\" ^ msi_path ^ " /forcerestart /qn /l+*vx C:\\" ^
msi_path ^ ".log\\\"\"
>  " in
>        Firstboot.add_firstboot_script g inspect.i_root
>          ("install " ^ msi_path) fb_script;
It could be easier to follow if you use sprintf here, something like:

let fb_script = sprintf "\
echo Removing any previously scheduled qemu-ga installation
schtasks.exe /Delete /TN Firstboot-qemu-ga /F
echo Scheduling delayed installation of qemu-ga from %s
powershell.exe -command \"$d = (get-date).AddSeconds(120); schtasks.exe
/Create /SC ONCE /ST $d.ToString('HH:mm') /SD
$d.ToString('MM/dd/yyyy') /RU SYSTEM /TN Firstboot-qemu-ga /TR
\\\"C:\\%s /forcerestart /qn /l+*vx C:\\%s.log\\\"\"
  msi_path msi_path msi_path in

The other possible problem is this seems to install a firstboot script
for every element of the list qemu_ga_files.  How long is this list?

Do filenames on this list need some kind of quoting?  The filenames
don't, but they seem to contain a path that comes from the virtio-win
ISO.

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html

On Thu, Mar 05, 2020 at 08:49:04AM +0000, Richard W.M. Jones
wrote:
> On Tue, Mar 03, 2020 at 03:45:49PM +0100, Tomáš Golembiovský wrote:
> > Instead of running firstboot script during early boot schedule a task
> > delayed for 2 minutes.
> > 
> > During the first boot, after virt-v2v conversion, Windows installs the
> > drivers injected by virt-v2v. When this installation is finished
> > Windows enforces some kind of internal reboot. This unfortunately
> > terminates any running firstboot scripts thus killing the installation
> > of qemu-ga MSI.
> > 
> > This is just a best-effort mitigation. It can still happen (e.g. with
> > slow disk drives) that the drivers are not yet installed when the
> > delayed installation starts. On the other hand we cannot delay it too
> > much otherwise we risk that the users logs in and will be doing some
> > work when the MSI installation starts. After MSI installation finishes
> > the VM needs to be rebooted which would be annoying if that would
happen
> > under users hands. Although this is not a best fix (that may come
later
> > as it is more complex, e.g. introducing waiting mechanism), the delay
as
> > it is defined works in most cases. And it dramaticaly improves the
> > situations -- originaly I experienced more than 90% failure rate.
> > 
> > Signed-off-by: Tomáš Golembiovský <tgolembi@redhat.com>
> > ---
> >  common                 |  2 +-
> >  v2v/convert_windows.ml | 12 ++++--------
> >  2 files changed, 5 insertions(+), 9 deletions(-)
> > 
> > diff --git a/common b/common
> > index ea10827b..5371257c 160000
> > --- a/common
> > +++ b/common
> > @@ -1 +1 @@
> > -Subproject commit ea10827b4cfb3cfe5f782421c01d2902e5f73f90
> > +Subproject commit 5371257c3cf27fb09d5f2e31ba378b0e6ccf5df6
> > diff --git a/v2v/convert_windows.ml b/v2v/convert_windows.ml
> > index 0fda1d4e..bed5989a 100644
> > --- a/v2v/convert_windows.ml
> > +++ b/v2v/convert_windows.ml
> > @@ -429,14 +429,10 @@ popd
> >     List.iter (
> >       fun msi_path ->
> >         let fb_script = "\
> > -echo Installing qemu-ga from " ^ msi_path ^ "
> > -\"\\" ^ msi_path ^ "\" /norestart /qn /l+*vx
\"%~dpn0.log\"
> > -set elvl=!errorlevel!
> > -echo Done installing qemu-ga error_level=!elvl!
> > -if !elvl! == 0 (
> > -  echo Restarting Windows...
> > -  shutdown /r /f /c \"rebooted by firstboot script\"
> > -)
> > +echo Removing any previously scheduled qemu-ga installation
> > +schtasks.exe /Delete /TN Firstboot-qemu-ga /F
> > +echo Scheduling delayed installation of qemu-ga from " ^
msi_path ^ "
> > +powershell.exe -command \"$d = (get-date).AddSeconds(120);
schtasks.exe /Create /SC ONCE /ST $d.ToString('HH:mm') /SD
$d.ToString('MM/dd/yyyy') /RU SYSTEM /TN Firstboot-qemu-ga /TR
\\\"C:\\" ^ msi_path ^ " /forcerestart /qn /l+*vx C:\\" ^
msi_path ^ ".log\\\"\"
> >  " in
> >        Firstboot.add_firstboot_script g inspect.i_root
> >          ("install " ^ msi_path) fb_script;
> 
> It could be easier to follow if you use sprintf here, something like:
> 
> let fb_script = sprintf "\
> echo Removing any previously scheduled qemu-ga installation
> schtasks.exe /Delete /TN Firstboot-qemu-ga /F
> echo Scheduling delayed installation of qemu-ga from %s
> powershell.exe -command \"$d = (get-date).AddSeconds(120);
schtasks.exe /Create /SC ONCE /ST $d.ToString('HH:mm') /SD
$d.ToString('MM/dd/yyyy') /RU SYSTEM /TN Firstboot-qemu-ga /TR
\\\"C:\\%s /forcerestart /qn /l+*vx C:\\%s.log\\\"\"
>   msi_path msi_path msi_path in
Ok, I can change that.
> 
> The other possible problem is this seems to install a firstboot script
> for every element of the list qemu_ga_files.  How long is this list?
At the moment it is just one. Until something changes in virtio-win ISO
or in our internal logic (like in virtio_iso_path_matches_qemu_ga()) it
will always be one.
> 
> Do filenames on this list need some kind of quoting?  The filenames
> don't, but they seem to contain a path that comes from the virtio-win
> ISO.
Those are really just file names without path. See 
https://github.com/libguestfs/virt-v2v/blob/master/v2v/windows_virtio.ml#L339

    Tomas
> 
> Rich.
> 
> -- 
> Richard Jones, Virtualization Group, Red Hat
http://people.redhat.com/~rjones
> Read my programming and virtualization blog: http://rwmj.wordpress.com
> virt-builder quickly builds VMs from scratch
> http://libguestfs.org/virt-builder.1.html
> 
-- 
Tomáš Golembiovský <tgolembi@redhat.com>