On Sun, 2023-01-29 at 14:28 +0800, shaozhengchao wrote:
>
> On 2023/1/18 17:45, Johannes Berg wrote:
> > On Fri, 2022-12-02 at 12:38 +0800, Zhengchao Shao wrote:
> > >
> > > --- a/net/mac80211/main.c
> > > +++ b/net/mac80211/main.c
> > > @@ -1326,6 +1326,7 @@ int ieee80211_register_hw(struct ieee80211_hw *hw)
> > > hw->rate_control_algorithm);
> > > rtnl_unlock();
> > > if (result < 0) {
> > > + ieee80211_txq_teardown_flows(local);
> > > wiphy_debug(local->hw.wiphy,
> > > "Failed to initialize rate control algorithm\n");
> > > goto fail_rate;
> > > @@ -1364,6 +1365,7 @@ int ieee80211_register_hw(struct ieee80211_hw *hw)
> > >
> > > sband = kmemdup(sband, sizeof(*sband), GFP_KERNEL);
> > > if (!sband) {
> > > + ieee80211_txq_teardown_flows(local);
> > > result = -ENOMEM;
> > > goto fail_rate;
> > > }
> >
> > I don't understand - we have a fail_rate label here where we free
> > everything.
> >
> > What if we get to fail_wiphy_register, don't we leak it in the same way?
> >
> > johannes
> Thank you for your review. Sorry it took so long to reply. The
> fail_rate label does not release the resources applied for in the
> ieee80211_txq_setup_flows(). Or maybe I missed something?
That's my point though - if we "goto fail_ifa" or "goto
fail_wiphy_register", we have the same bug, no?
So shouldn't the patch simply be this:
diff --git a/net/mac80211/main.c b/net/mac80211/main.c
index 846528850612..a42d1f0ef7a5 100644
--- a/net/mac80211/main.c
+++ b/net/mac80211/main.c
@@ -1442,6 +1442,7 @@ int ieee80211_register_hw(struct ieee80211_hw *hw)
ieee80211_remove_interfaces(local);
rtnl_unlock();
fail_rate:
+ ieee80211_txq_teardown_flows(local);
fail_flows:
ieee80211_led_exit(local);
destroy_workqueue(local->workqueue);
johannes
