Re: [Qemu-devel] [PATCH] monitor: fix dangling CPU pointer

[Greg Kurz <groug@kaod.org>]

On Mon, 16 Oct 2017 10:41:39 +0200
Igor Mammedov <imammedo@redhat.com> wrote:

> On Fri, 13 Oct 2017 13:47:51 +0200
> Greg Kurz <groug@kaod.org> wrote:
> 
> [...]
> 
> > Note that the resolution should really return a CPU object, otherwise
> > we have a bug. This is achieved by relying on object_resolve_path()
> > and CPU() instead of calling object_resolve_path_type(path, TYPE_CPU).  
> I'm not sure what you are trying to say here, could you explain
> a bit more why CPU(object_resolve_path()) is chosen vs
> object_resolve_path_type(path, TYPE_CPU)
> 

IIUC, if we use object_resolve_path_type(path, TYPE_CPU) and path doesn't
point to a CPU object, it will return NULL (see object_resolve_abs_path())
just like if the CPU got hot-unplugged.

My point is that the path we got from object_get_canonical_path() did
point to a CPU: if later on this path resolves to an object that isn't
a CPU, then some code somewhere used the same QOM path for some unrelated
object. I tend to think this is a bug in QEMU and we shouldn't silently
ignore it.

Makes sense ?

> 
> > Suggested-by: Igor Mammedov <imammedo@redhat.com>
> > Signed-off-by: Greg Kurz <groug@kaod.org>
> > ---
> >  monitor.c |   25 ++++++++++++++++++++-----
> >  1 file changed, 20 insertions(+), 5 deletions(-)
> > 
> > diff --git a/monitor.c b/monitor.c
> > index fe0d1bdbb461..8489b2ad99c0 100644
> > --- a/monitor.c
> > +++ b/monitor.c
> > @@ -200,7 +200,7 @@ struct Monitor {
> >  
> >      ReadLineState *rs;
> >      MonitorQMP qmp;
> > -    CPUState *mon_cpu;
> > +    gchar *mon_cpu_path;
> >      BlockCompletionFunc *password_completion_cb;
> >      void *password_opaque;
> >      mon_cmd_t *cmd_table;
> > @@ -579,6 +579,7 @@ static void monitor_data_init(Monitor *mon)
> >  
> >  static void monitor_data_destroy(Monitor *mon)
> >  {
> > +    g_free(mon->mon_cpu_path);
> >      qemu_chr_fe_deinit(&mon->chr, false);
> >      if (monitor_is_qmp(mon)) {
> >          json_message_parser_destroy(&mon->qmp.parser);
> > @@ -1047,20 +1048,34 @@ int monitor_set_cpu(int cpu_index)
> >      if (cpu == NULL) {
> >          return -1;
> >      }
> > -    cur_mon->mon_cpu = cpu;
> > +    g_free(cur_mon->mon_cpu_path);
> > +    cur_mon->mon_cpu_path = object_get_canonical_path(OBJECT(cpu));
> >      return 0;
> >  }
> >  
> >  CPUState *mon_get_cpu(void)
> >  {
> > -    if (!cur_mon->mon_cpu) {
> > +    CPUState *cpu;
> > +
> > +    if (cur_mon->mon_cpu_path) {
> > +        Object *obj = object_resolve_path(cur_mon->mon_cpu_path, NULL);
> > +
> > +        if (!obj) {
> > +            g_free(cur_mon->mon_cpu_path);
> > +            cur_mon->mon_cpu_path = NULL;
> > +        } else {
> > +            cpu = CPU(obj);  
> this potentially might abort if obj couldn't be cast to TYPE_CPU
> 

This is deliberate... is there a case where cur_mon->mon_cpu_path would
legitimately point to something that isn't of type TYPE_CPU ?

> 
> > +        }
> > +    }
> > +    if (!cur_mon->mon_cpu_path) {
> >          if (!first_cpu) {
> >              return NULL;
> >          }
> >          monitor_set_cpu(first_cpu->cpu_index);
> > +        cpu = first_cpu;
> >      }
> > -    cpu_synchronize_state(cur_mon->mon_cpu);
> > -    return cur_mon->mon_cpu;
> > +    cpu_synchronize_state(cpu);
> > +    return cpu;
> >  }
> >  
> >  CPUArchState *mon_get_cpu_env(void)
> > 
> >   
>