On Wed, Jun 19, 2019 at 01:23:50PM +0200, Peter Zijlstra wrote:
> On Wed, Jun 19, 2019 at 01:12:12PM +0200, Miroslav Benes wrote:
> > > @@ -3780,7 +3781,7 @@ static int load_module(struct load_info *info, 
> > > const char __user *uargs,
> > >  
> > >   err = prepare_coming_module(mod);
> > >   if (err)
> > > -         goto bug_cleanup;
> > > +         goto coming_cleanup;
> > 
> > Not good. klp_module_going() is not prepared to be called without 
> > klp_module_coming() succeeding. "Funny" things might happen.
> 
> Bah, I did look at that but failed to spot it :/
> 
> > So it calls for more fine-grained error handling.
> 
> Another approach that I considered was trying to re-iterate the notifier
> list up until the point we got, but that was fairly non-trivial and
> needed changes to the notifier crud itself.
> 
> I'll try again.

How's something like so:

diff --git a/kernel/module.c b/kernel/module.c
index 80c7c09584cf..eba6560c89da 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -3631,16 +3631,28 @@ static int complete_formation(struct module *mod, 
struct load_info *info)
 
 static int prepare_coming_module(struct module *mod)
 {
-       int err;
+       struct blocking_notifier_head *nh = &module_notify_list;
+       int err, nr;
 
        ftrace_module_enable(mod);
        err = klp_module_coming(mod);
        if (err)
                return err;
 
-       blocking_notifier_call_chain(&module_notify_list,
-                                    MODULE_STATE_COMING, mod);
-       return 0;
+       if (!rcu_access_pointer(nh->head))
+               return 0;
+
+       down_read(&nh->rwsem);
+       ret = notifier_call_chain(&nh->head, MODULE_STATE_COMING, mod, -1, &nr);
+       if (ret & NOTIFIER_STOP_MASK)
+               notifier_call_chain(&nh->head, MODULE_STATE_GOING, mod, nr, 
NULL);
+       up_read(&nh->rwsem);
+
+       err = notifier_to_err(err);
+       if (err)
+               klp_module_going(mod);
+
+       return err;
 }
 
 static int unknown_module_param_cb(char *param, char *val, const char *modname,

Reply via email to