Re: [PATCH v4 2/4] xen/console: introduce console input permission

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Fri, 30 May 2025, dmkhn@xxxxxxxxx wrote:
> On Thu, May 29, 2025 at 05:58:00PM -0700, Stefano Stabellini wrote:
> > On Thu, 29 May 2025, dmkhn@xxxxxxxxx wrote:
> > > Add new flag to domain structure for marking permission to intercept
> > > the physical console input by the domain.
> > >
> > > Update console input switch logic accordingly.
> > >
> > > No functional change intended.
> > >
> > > Signed-off-by: Denis Mukhin <dmukhin@xxxxxxxx>
> > > ---
> > > Changes since v3:
> > > - rebased
> > > ---
> > >  xen/arch/arm/vpl011.c      |  2 ++
> > >  xen/arch/x86/pv/shim.c     |  2 ++
> > >  xen/common/domain.c        |  2 ++
> > >  xen/drivers/char/console.c | 18 +++++++++++++++++-
> > >  xen/include/xen/sched.h    |  8 +++++++-
> > >  5 files changed, 30 insertions(+), 2 deletions(-)
> > >
> > > diff --git a/xen/arch/arm/vpl011.c b/xen/arch/arm/vpl011.c
> > > index 66047bf33c..147958eee8 100644
> > > --- a/xen/arch/arm/vpl011.c
> > > +++ b/xen/arch/arm/vpl011.c
> > > @@ -737,6 +737,8 @@ int domain_vpl011_init(struct domain *d, struct 
> > > vpl011_init_info *info)
> > >      register_mmio_handler(d, &vpl011_mmio_handler,
> > >                            vpl011->base_addr, GUEST_PL011_SIZE, NULL);
> > >
> > > +    d->console.input_allowed = true;
> > 
> > This should be set only when backend_in_domain = false.
> > 
> > 
> > >      return 0;
> > >
> > >  out1:
> > > diff --git a/xen/arch/x86/pv/shim.c b/xen/arch/x86/pv/shim.c
> > > index c506cc0bec..bc2a7dd5fa 100644
> > > --- a/xen/arch/x86/pv/shim.c
> > > +++ b/xen/arch/x86/pv/shim.c
> > > @@ -238,6 +238,8 @@ void __init pv_shim_setup_dom(struct domain *d, 
> > > l4_pgentry_t *l4start,
> > >       * guest from depleting the shim memory pool.
> > >       */
> > >      d->max_pages = domain_tot_pages(d);
> > > +
> > > +    d->console.input_allowed = true;
> > >  }
> > >
> > >  static void write_start_info(struct domain *d)
> > > diff --git a/xen/common/domain.c b/xen/common/domain.c
> > > index 87e5be35e5..9bc66d80c4 100644
> > > --- a/xen/common/domain.c
> > > +++ b/xen/common/domain.c
> > > @@ -835,6 +835,8 @@ struct domain *domain_create(domid_t domid,
> > >          flags |= CDF_hardware;
> > >          if ( old_hwdom )
> > >              old_hwdom->cdf &= ~CDF_hardware;
> > > +
> > > +        d->console.input_allowed = true;
> > >      }
> > >
> > >      /* Holding CDF_* internal flags. */
> > > diff --git a/xen/drivers/char/console.c b/xen/drivers/char/console.c
> > > index 30701ae0b0..8a0bcff78f 100644
> > > --- a/xen/drivers/char/console.c
> > > +++ b/xen/drivers/char/console.c
> > > @@ -512,9 +512,21 @@ static unsigned int __read_mostly console_rx = 0;
> > >
> > >  struct domain *console_get_domain(void)
> > >  {
> > > +    struct domain *d;
> > > +
> > >      if ( console_rx == 0 )
> > >              return NULL;
> > > -    return rcu_lock_domain_by_id(console_rx - 1);
> > > +
> > > +    d = rcu_lock_domain_by_id(console_rx - 1);
> > > +    if ( !d )
> > > +        return NULL;
> > > +
> > > +    if ( d->console.input_allowed )
> > > +        return d;
> > > +
> > > +    rcu_unlock_domain(d);
> > > +
> > > +    return NULL;
> > 
> > The original idea was to skip over domains that cannot have any input so
> > I don't think we should get in this situation. We could even have an
> > assert.
> > 
> > 
> > >  }
> > >
> > >  void console_put_domain(struct domain *d)
> > > @@ -551,6 +563,10 @@ static void console_switch_input(void)
> > >          if ( d )
> > >          {
> > >              rcu_unlock_domain(d);
> > > +
> > > +            if ( !d->console.input_allowed )
> > > +                break;
> > 
> > shouldn't this be continue instead of break?
> > 
> > 
> > >              console_rx = next_rx;
> > >              printk("*** Serial input to DOM%u", domid);
> > >              break;
> > > diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
> > > index 559d201e0c..e91c99a8f3 100644
> > > --- a/xen/include/xen/sched.h
> > > +++ b/xen/include/xen/sched.h
> > > @@ -512,7 +512,7 @@ struct domain
> > >      bool             auto_node_affinity;
> > >      /* Is this guest fully privileged (aka dom0)? */
> > >      bool             is_privileged;
> > > -    /* Can this guest access the Xen console? */
> > > +    /* XSM: permission to use HYPERCALL_console_io hypercall */
> > >      bool             is_console;
> > 
> > While I am in favor of this direction and we certainly need a better way
> > to distinguish domains that can use HYPERCALL_console_io hypercall from
> > others, could we simplify this and just assume that "is_console" implies
> > input_allowed and also set is_console = true in all the same places you
> > are setting input_allowed = true in this patch?
> > 
> > For clarity, I am suggesting:
> > - do not add input_allowed
> > - set is_console = true in domain_vpl011_init, pv_shim_setup_dom, etc.
> > 
> > The only side effect is that we would allow domains with vpl011 to also
> > use console hypercalls but I don't think there is any harm in that?
> > 
> > I don't feel strongly about this, I am just trying to find ways to make
> > things simple. I apologize if it was already discussed during review of
> > one of the previous versions.
> 
> There was feedback on using is_console:
> 
>   
> https://lore.kernel.org/xen-devel/e899f63b-6182-4b53-9fb4-9a821e75648b@xxxxxxxx/
> 
> AFAIU, since XSM is the existing user of is_console, there should be a new
> separate flag to avoid collision with the existing one.

OK, I suspected as much. In that case, it is fine to continue with the
new flag.