Libguestfs - Oct 2015 - Re: [PATCH] v2v: virtio-win: include *.dll too

On Tue, Oct 27, 2015 at 09:12:41AM +0000, Richard W.M. Jones
wrote:
> On Mon, Oct 26, 2015 at 09:00:03PM +0300, Roman Kagan wrote:
> > Windows QXL drivers include also qxldd.dll which used to get filtered
> > out and not copied over into the guest.  As a result QXL driver failed
> > to install due to a missing file.
> >      (* Skip files without specific extensions. *)
> > -    let extensions = ["cat"; "inf";
"pdb"; "sys"] in
> > +    let extensions = ["cat"; "dll";
"inf"; "pdb"; "sys"] in
> 
> Actually no, this part is (may be) wrong.
> 
> This causes WdfCoInstaller*.dll from the ISO to be copied in.  It's
> not clear to me whether these files are needed - I suspect not.
Yes it is, see the corresponding *.inf where it's mentioned in the
corresponding CopyFiles directive (and yes, I just verified that with
that file missing the balloon driver installation fails with 'file not
found' error).

As a matter of fact I tend to think that filtering of files by suffix is
a bad idea in general: the packager of the drivers should be free to
ship any files he wants, including docs and text notes.  It's only the
path that should match against the guest OS flavor and that really
matters.  E.g. the balloon driver directory also contains
blnsvr.exe (and its .pdb) which is a usermode service responsible for
feeding the balloon device with the memory stats; it's not a part of the
driver per se but it's currently bundled with it.  (I'm yet to figure
out who's responsible for installing it).
> It also causes the tests to fail, although only when you have
> installed the optional ocaml oUnit2 package:
That's a problem with the test then (and no I didn't have that package
installed so my make -C v2v check passed).

Roman.

On Tue, Oct 27, 2015 at 02:08:42PM +0300, Roman Kagan
wrote:
> On Tue, Oct 27, 2015 at 09:12:41AM +0000, Richard W.M. Jones wrote:
> > On Mon, Oct 26, 2015 at 09:00:03PM +0300, Roman Kagan wrote:
> > > Windows QXL drivers include also qxldd.dll which used to get
filtered
> > > out and not copied over into the guest.  As a result QXL driver
failed
> > > to install due to a missing file.
> > >      (* Skip files without specific extensions. *)
> > > -    let extensions = ["cat"; "inf";
"pdb"; "sys"] in
> > > +    let extensions = ["cat"; "dll";
"inf"; "pdb"; "sys"] in
> > 
> > Actually no, this part is (may be) wrong.
> > 
> > This causes WdfCoInstaller*.dll from the ISO to be copied in. 
It's
> > not clear to me whether these files are needed - I suspect not.
> 
> Yes it is, see the corresponding *.inf where it's mentioned in the
> corresponding CopyFiles directive (and yes, I just verified that with
> that file missing the balloon driver installation fails with 'file not
> found' error).
Maybe we should be looking at the CopyFiles directive?  (As you may be
able to guess, I know next to nothing about how Windows drivers work).

Anyhow, can you fix up the test at least.

Thanks, Rich.
> As a matter of fact I tend to think that filtering of files by suffix is
> a bad idea in general: the packager of the drivers should be free to
> ship any files he wants, including docs and text notes.  It's only the
> path that should match against the guest OS flavor and that really
> matters.  E.g. the balloon driver directory also contains
> blnsvr.exe (and its .pdb) which is a usermode service responsible for
> feeding the balloon device with the memory stats; it's not a part of
the
> driver per se but it's currently bundled with it.  (I'm yet to
figure
> out who's responsible for installing it).
> 
> > It also causes the tests to fail, although only when you have
> > installed the optional ocaml oUnit2 package:
> 
> That's a problem with the test then (and no I didn't have that
package
> installed so my make -C v2v check passed).
> 
> Roman.
-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
libguestfs lets you edit virtual machines.  Supports shell scripting,
bindings from many languages.  http://libguestfs.org

On Tue, Oct 27, 2015 at 12:02:40PM +0000, Richard W.M. Jones
wrote:
> On Tue, Oct 27, 2015 at 02:08:42PM +0300, Roman Kagan wrote:
> > On Tue, Oct 27, 2015 at 09:12:41AM +0000, Richard W.M. Jones wrote:
> > > On Mon, Oct 26, 2015 at 09:00:03PM +0300, Roman Kagan wrote:
> > > > Windows QXL drivers include also qxldd.dll which used to get
filtered
> > > > out and not copied over into the guest.  As a result QXL
driver failed
> > > > to install due to a missing file.
> > > >      (* Skip files without specific extensions. *)
> > > > -    let extensions = ["cat"; "inf";
"pdb"; "sys"] in
> > > > +    let extensions = ["cat"; "dll";
"inf"; "pdb"; "sys"] in
> > > 
> > > Actually no, this part is (may be) wrong.
> > > 
> > > This causes WdfCoInstaller*.dll from the ISO to be copied in. 
It's
> > > not clear to me whether these files are needed - I suspect not.
> > 
> > Yes it is, see the corresponding *.inf where it's mentioned in the
> > corresponding CopyFiles directive (and yes, I just verified that with
> > that file missing the balloon driver installation fails with 'file
not
> > found' error).
> 
> Maybe we should be looking at the CopyFiles directive?  (As you may be
> able to guess, I know next to nothing about how Windows drivers work).
So do I, and parsing the *.inf files is the last thing I'd like to do
here; I'm not sure it can be made reliable (I mean, not the parsing, but
making any sense out of the parsed data).

However I'm now facing a more serious problem: when I actually went
ahead and looked into the virio-win iso packaged in the rhel virtio-win
rpm, I found out that its contents differed significantly from what was
packaged in that very same rpm as a driver directory hierarchy.

Namely,

  - there are no qxl drivers on the iso, but they are present in the rpm

  - there are no balloon, qemupciserial, viorng, vioserial in the rpm,
    but they are present on the iso

  - qemupciserial is an inf-only 'driver' which matches all windows
    versions; it's just one copy at the sub-toplevel directory, so it
    won't pass the virtio_iso_path_matches_guest_os() criteria

  - there are a lot of duplicates between files for different windows
    flavors

To sum up, the packaging and naming policy of the virtio-win rpm and the
virtio-win iso therein are different and neither is clear.  Hardcoding
the policy in v2v without actually knowing it appears risky at best.

We need guidance here from someone who knows how that virtio-win stuff
is packaged and how different it is across distros.  Any idea whom to
contact on the matter?
> Anyhow, can you fix up the test at least.
Sure, but I'd like first to figure out how to fix up the code ;)

Roman.