Setting sev->fh to NULL causes problems for the del op added in the next
patch of this series, since this op needs a way to get to its own data
structures, and typically this will be done by using container_of on an
embedded v4l2_fh struct.
The reason the original code is setting sev->fh to NULL is to signal
to users of the event framework that the unsubscription has happened,
but since their is no shared lock between the event framework and users
of it, this is inherently racy, and it also turns out to be unnecessary
as long as both the event framework and the user of the framework do their
own locking properly and the user guarantees that it holds no references
to the subcribed_event structure after its del operation has been called.
This is best explained by looking at the only code currently checking for
sev->fh being set to NULL on unsubscribe, which is the v4l2-ctrls.c send_event
function. Here is the relevant code from v4l2-ctrls: send_event():
if (sev->fh && (sev->fh != fh ||
(sev->flags & V4L2_EVENT_SUB_FL_ALLOW_FEEDBACK)))
v4l2_event_queue_fh(sev->fh, &ev);
Now lets say that v4l2_event_unsubscribe and v4l2-ctrls: send_event() race
on the same sev, then the following could happens:
1) send_event checks sev->fh, finds it is not NULL
<thread switch>
2) v4l2_event_unsubscribe sets sev->fh NULL
3) v4l2_event_unsubscribe calls v4l2_ctrls del_event function, this blocks
as the thread calling send_event holds the ctrl_lock
<thread switch>
4) send_event calls v4l2_event_queue_fh(sev->fh, &ev) which not is equivalent
to calling: v4l2_event_queue_fh(NULL, &ev)
5) oops, NULL pointer deref.
Now again without setting sev->fh to NULL in v4l2_event_unsubscribe and
without the (now senseless since always true) sev->fh != NULL check in
1) send_event is about to call v4l2_event_queue_fh(sev->fh, &ev)
<thread switch>
2) v4l2_event_unsubscribe removes sev->list from the fh->subscribed list
<thread switch>
3) send_event calls v4l2_event_queue_fh(sev->fh, &ev)
4) v4l2_event_queue_fh blocks on the fh_lock spinlock
<thread switch>
5) v4l2_event_unsubscribe unlocks the fh_lock spinlock
6) v4l2_event_unsubscribe calls v4l2_ctrls del_event function, this blocks
as the thread calling send_event holds the ctrl_lock
<thread switch>
8) v4l2_event_queue_fh takes the fh_lock
7) v4l2_event_queue_fh calls v4l2_event_subscribed, does not find it since
sev->list has been removed from fh->subscribed already -> does nothing
9) v4l2_event_queue_fh releases the fh_lock
10) the caller of send_event releases the ctrl lock (mutex)
<thread switch>
11) v4l2_ctrls del_event takes the ctrl lock
12) v4l2_ctrls del_event removes sev->node from the ev_subs list
13) v4l2_ctrls del_event releases the ctrl lock
14) v4l2_event_unsubscribe frees the sev, to which no references are being
held anymore
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Hans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
fill_event(&ev, ctrl, changes);
list_for_each_entry(sev, &ctrl->ev_subs, node)
- if (sev->fh && (sev->fh != fh ||
- (sev->flags & V4L2_EVENT_SUB_FL_ALLOW_FEEDBACK)))
+ if (sev->fh != fh ||
+ (sev->flags & V4L2_EVENT_SUB_FL_ALLOW_FEEDBACK))
v4l2_event_queue_fh(sev->fh, &ev);
}
fh->navailable--;
}
list_del(&sev->list);
- sev->fh = NULL;
}
spin_unlock_irqrestore(&fh->vdev->fh_lock, flags);