Re: [PATCH] audio/control: Fix invalid read on exit

[Bastien Nocera <hadess@hadess.net>]

On Thu, 2017-11-23 at 14:51 +0200, Luiz Augusto von Dentz wrote:
> Hi Bastien,
> 
> On Thu, Nov 23, 2017 at 12:23 PM, Luiz Augusto von Dentz
> <luiz.dentz@gmail.com> wrote:
> > Hi Bastien,
> > 
> > On Wed, Nov 22, 2017 at 3:50 PM, Bastien Nocera <hadess@hadess.net>
> > wrote:
> > > Make control_disconnect() a no-op if control was already
> > > disconnected and freed.
> > > 
> > > bluetoothd[8436]: profiles/audio/avrcp.c:avrcp_disconnect() path
> > > /org/bluez/hci0/dev_38_71_DE_C0_FC_26
> > > ==8436== Invalid read of size 8
> > > ==8436==    at 0x41FFB9: control_disconnect (control.c:130)
> > > ==8436==    by 0x469E0A: service_remove (service.c:177)
> > > ==8436==    by 0x475869: device_remove (device.c:4071)
> > > ==8436==    by 0x45DEF9: adapter_remove (adapter.c:5485)
> > > ==8436==    by 0x466B2A: adapter_cleanup (adapter.c:8679)
> > > ==8436==    by 0x40BE16: main (main.c:782)
> > > ==8436==  Address 0x84b8ac8 is 8 bytes inside a block of size 48
> > > free'd
> > > ==8436==    at 0x4C30D18: free (vg_replace_malloc.c:530)
> > > ==8436==    by 0x50D44AD: g_free (in /usr/lib64/libglib-
> > > 2.0.so.0.5400.2)
> > > ==8436==    by 0x488AED: remove_interface (object.c:667)
> > > ==8436==    by 0x488FE9: g_dbus_unregister_interface
> > > (object.c:1391)
> > > ==8436==    by 0x469E20: service_remove (service.c:179)
> > > ==8436==    by 0x475869: device_remove (device.c:4071)
> > > ==8436==    by 0x45DEF9: adapter_remove (adapter.c:5485)
> > > ==8436==    by 0x466B2A: adapter_cleanup (adapter.c:8679)
> > > ==8436==    by 0x40BE16: main (main.c:782)
> > > ==8436==  Block was alloc'd at
> > > ==8436==    at 0x4C31A1E: calloc (vg_replace_malloc.c:711)
> > > ==8436==    by 0x50D43F0: g_malloc0 (in /usr/lib64/libglib-
> > > 2.0.so.0.5400.2)
> > > ==8436==    by 0x41FC09: control_init (control.c:320)
> > > ==8436==    by 0x42006D: control_init_remote (control.c:361)
> > > ==8436==    by 0x469D59: service_probe (service.c:160)
> > > ==8436==    by 0x46E464: probe_service (device.c:4207)
> > > ==8436==    by 0x46E4C2: dev_probe (device.c:4226)
> > > ==8436==    by 0x46989B: btd_profile_foreach (profile.c:708)
> > > ==8436==    by 0x47239D: device_probe_profiles (device.c:4285)
> > > ==8436==    by 0x50ED51C: g_slist_foreach (in /usr/lib64/libglib-
> > > 2.0.so.0.5400.2)
> > > ==8436==    by 0x50ED54A: g_slist_free_full (in
> > > /usr/lib64/libglib-2.0.so.0.5400.2)
> > > ==8436==    by 0x45F1EE: load_devices (adapter.c:3864)
> > > ==8436==    by 0x465E59: adapter_register (adapter.c:7741)
> > > ==8436==    by 0x465E59: read_info_complete (adapter.c:8284)
> > > ==8436==    by 0x48D277: request_complete (mgmt.c:261)
> > > ==8436==    by 0x48DD44: can_read_data (mgmt.c:353)
> > > ==8436==    by 0x4999A2: watch_callback (io-glib.c:170)
> > > ==8436==    by 0x50CEBB6: g_main_context_dispatch (in
> > > /usr/lib64/libglib-2.0.so.0.5400.2)
> > > ==8436==    by 0x50CEF5F: ??? (in /usr/lib64/libglib-
> > > 2.0.so.0.5400.2)
> > > ==8436==    by 0x50CF271: g_main_loop_run (in /usr/lib64/libglib-
> > > 2.0.so.0.5400.2)
> > > ==8436==    by 0x40BDE8: main (main.c:770)
> > > ---
> > >  profiles/audio/control.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/profiles/audio/control.c b/profiles/audio/control.c
> > > index edc4a98ce..707276d29 100644
> > > --- a/profiles/audio/control.c
> > > +++ b/profiles/audio/control.c
> > > @@ -127,7 +127,7 @@ int control_disconnect(struct btd_service
> > > *service)
> > >  {
> > >         struct control *control =
> > > btd_service_get_user_data(service);
> > > 
> > > -       if (!control->session)
> > > +       if (!control || !control->session)
> > >                 return -ENOTCONN;
> > > 
> > >         avctp_disconnect(control->session);
> > > --
> > > 2.14.3
> > 
> > Applied, thanks.
> 
> Johan just pointed out that this is probably not fixing the problem
> since it is not a NULL pointer deference as the pointer is valid,
> instead what is probably happening is that control_unregister is
> being
> which frees the control but as that is pointed by a second service
> (probably both target and remote services are supported) that then
> cause the crash you are experiencing.

I only compile tested it as I could not reproduce the crash afterwards.
It only happened occasionally when Ctrl+C'ing a running bluetoothd.

I'd either add leave that code as-is (with the patch), or add an
assert, so it would be clear that the condition is not supposed to
happen.

Cheers