Re: Crash in VirtualBox virtual machines running kernel 6.5

[Larry Finger]

On 7/19/23 17:37, Kees Cook wrote:
> On July 19, 2023 2:24:52 PM PDT, Larry Finger <Larry.Finger@xxxxxxxxxxxx> wrote:
> > On 7/19/23 10:36, Kees Cook wrote:
> > > Okay, please try:
> > >
> > > diff --git a/fs/vboxsf/shfl_hostintf.h b/fs/vboxsf/shfl_hostintf.h
> > > index aca829062c12..902fe3224453 100644
> > > --- a/fs/vboxsf/shfl_hostintf.h
> > > +++ b/fs/vboxsf/shfl_hostintf.h
> > > @@ -68,12 +68,11 @@ struct shfl_string {
> > >      	/** UTF-8 or UTF-16 string. Nul terminated. */
> > >    	union {
> > > -		u8 utf8[2];
> > > -		u16 utf16[1];
> > > -		u16 ucs2[1]; /* misnomer, use utf16. */
> > > +		DECLARE_FLEX_ARRAY(u8, utf8);
> > > +		DECLARE_FLEX_ARRAY(u16, utf16);
> > >    	} string;
> > >    };
> > > -VMMDEV_ASSERT_SIZE(shfl_string, 6);
> > > +VMMDEV_ASSERT_SIZE(shfl_string, 4);
> > >      /* The size of shfl_string w/o the string part. */
> > >    #define SHFLSTRING_HEADER_SIZE  4
> > >
> > >
> > > The size assert doesn't seem to be used anywhere else, but I can do a
> > > more careful binary analysis later today...''
> >
> > Kees,
> >
> > The testing was harder than I expected. My standard kernel would not load the system disk on the VM, thus I had to build one using my distros configuration. It is really painful to wait for all those drivers to build, but I figured that might be faster than trying to find the incorrect parameter.
> >
> > I can finally report that vboxsf no longer generated a BUG. It is getting farther, but I am not done yet. It now generates a line that says "Unknown parameter tag" and then hangs. I have not tracked that down yet.
> >
> > I was not aware of the DECLARE_FLEX_ARRAY() macro. I had considered removing the union and creating a simple string[] declaration, but I do not mess with file systems, and decided to let the experts handle it.
> >
> > I will let you know what I find about that unknown parameter. It probably is coming from VirtualBox.
>
> It's possible the size really needs to stay 6 bytes. In that case, try adding a "u8 legacy_padding[2]" to the union and restore the VMMDEV_ASSERT_SIZE?

Kees,

You win the prize. Adding the 2 bytes of padding restored operations.

Thanks,

Larry