[Spice-devel] [PATCH v5 5/9] Do not access ImageEncoders internal to lock/unlock glz encoding

[Frediano Ziglio]

> 
> On Wed, 2016-06-15 at 10:37 +0100, Frediano Ziglio wrote:
> > Signed-off-by: Frediano Ziglio <fziglio at redhat.com>
> > ---
> >  server/dcc-encoders.c    | 14 ++++++++++++++
> >  server/dcc-encoders.h    |  2 ++
> >  server/display-channel.c | 18 +++++-------------
> >  3 files changed, 21 insertions(+), 13 deletions(-)
> > 
> > diff --git a/server/dcc-encoders.c b/server/dcc-encoders.c
> > index 6fccb17..a5bc328 100644
> > --- a/server/dcc-encoders.c
> > +++ b/server/dcc-encoders.c
> > @@ -548,6 +548,20 @@ static void red_glz_drawable_free(RedGlzDrawable
> > *glz_drawable)
> >      }
> >  }
> >  
> > +void image_encoders_glz_encode_lock(ImageEncoders *enc)
> > +{
> > +    if (enc->glz_dict) {
> > +        pthread_rwlock_wrlock(&enc->glz_dict->encode_lock);
> > +    }
> > +}
> > +
> > +void image_encoders_glz_encode_unlock(ImageEncoders *enc)
> > +{
> > +    if (enc->glz_dict) {
> > +        pthread_rwlock_unlock(&enc->glz_dict->encode_lock);
> > +    }
> > +}
> > +
> >  /*
> >   * Remove from the global lz dictionary some glz_drawables that have no
> > reference to
> >   * Drawable (their qxl drawables are released too).
> > diff --git a/server/dcc-encoders.h b/server/dcc-encoders.h
> > index a38de41..e13c809 100644
> > --- a/server/dcc-encoders.h
> > +++ b/server/dcc-encoders.h
> > @@ -48,6 +48,8 @@ void
> > image_encoders_free_glz_drawables_to_free(ImageEncoders* enc);
> >  gboolean image_encoders_glz_create(ImageEncoders *enc, uint8_t id);
> >  void image_encoders_glz_get_restore_data(ImageEncoders *enc,
> >                                           uint8_t *out_id,
> > GlzEncDictRestoreData *out_data);
> > +void image_encoders_glz_encode_lock(ImageEncoders *enc);
> > +void image_encoders_glz_encode_unlock(ImageEncoders *enc);
> >  void drawable_ring_free_glz_drawables(Ring *drawable_ring);
> >  void drawable_ring_detach_glz_drawables(Ring *drawable_ring);
> >  
> > diff --git a/server/display-channel.c b/server/display-channel.c
> > index 1b5a03a..15da137 100644
> > --- a/server/display-channel.c
> > +++ b/server/display-channel.c
> > @@ -1206,14 +1206,10 @@ void display_channel_free_some(DisplayChannel
> > *display)
> >      spice_debug("#draw=%d, #glz_draw=%d", display->drawable_count,
> >                  display->encoder_globals.glz_drawable_count);
> >      FOREACH_CLIENT(display, link, next, dcc) {
> > -        GlzSharedDictionary *glz_dict = dcc->encoders.glz_dict;
> > -
> > -        if (glz_dict) {
> > -            // encoding using the dictionary is prevented since the
> > following
> > operations might
> > -            // change the dictionary
> > -            pthread_rwlock_wrlock(&glz_dict->encode_lock);
> > -            n = image_encoders_free_some_independent_glz_drawables(&dcc-
> > >encoders);
> > -        }
> > +        // encoding using the dictionary is prevented since the following
> > operations might
> > +        // change the dictionary
> > +        image_encoders_glz_encode_lock(&dcc->encoders);
> > +        n = image_encoders_free_some_independent_glz_drawables(&dcc-
> > >encoders);
> 
> This is a change of behavior. Previously, we only called
> _free_some_independent_glz_drawables() if glz_dict was non-NULL. Now we call
> it
> regardless of whether it's NULL or not. If that's intentional, an explanation
> (at least in the commit log) is warranted.
> 

Fixed in following version.

> >      }
> >  
> >      while (!ring_is_empty(&display->current_list) && n++ <
> > RED_RELEASE_BUNCH_SIZE) {
> > @@ -1221,11 +1217,7 @@ void display_channel_free_some(DisplayChannel
> > *display)
> >      }
> >  
> >      FOREACH_CLIENT(display, link, next, dcc) {
> > -        GlzSharedDictionary *glz_dict = dcc->encoders.glz_dict;
> > -
> > -        if (glz_dict) {
> > -            pthread_rwlock_unlock(&glz_dict->encode_lock);
> > -        }
> > +        image_encoders_glz_encode_unlock(&dcc->encoders);
> 
> I don't think there's currently any way that encoders->glz_dict might be
> changed
> between the lock above and the unlock here, but the fact that you no longer
> cache the glz_dict value in a local variable opens us up to this risk. If
> enc-
> >glz_dict was NULL when we called image_encoders_glz_encode_lock() and then
> >was
> set to non-NULL between there and here, calling _unlock() here will attempt
> to
> unlock a lock that had not been locked. Maybe this is not possible and not
> worth
> worrying about, but I thought I'd mention it.
> 

No, this is exactly as before, glz_dict local is computed again on
the second loop.

> >      }
> >  }
> >  
> 
> Reviewed-by: Jonathon Jongsma <jjongsma at redhat.com>
> 

Frediano