On 03/18, Stanislav Fomichev wrote: > nl80211_pre_doit acquires rtnl_lock and then wiphy_lock, releasing > rtnl while keeping wiphy_lock held until post_doit. With the > introduction of rx_mode_wq and its flush in netdev_run_todo, calling > rtnl_unlock here creates a circular lock dependency: > > Chain exists of: > (wq_completion)rx_mode_wq --> rtnl_mutex --> &rdev->wiphy.mtx > > Possible unsafe locking scenario: > > CPU0 CPU1 > ---- ---- > lock(&rdev->wiphy.mtx); > lock(rtnl_mutex); > lock(&rdev->wiphy.mtx); > lock((wq_completion)rx_mode_wq); > > Switch to __rtnl_unlock to skip netdev_run_todo in nl80211_pre_doit. > This seems safe because we run before the op. > > Link: > http://lore.kernel.org/netdev/[email protected] > Signed-off-by: Stanislav Fomichev <[email protected]> > --- > net/wireless/nl80211.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c > index 2225f5d0b124..ce5f25d4c87e 100644 > --- a/net/wireless/nl80211.c > +++ b/net/wireless/nl80211.c > @@ -18192,7 +18192,7 @@ static int nl80211_pre_doit(const struct > genl_split_ops *ops, > __release(&rdev->wiphy.mtx); > } > if (!(internal_flags & NL80211_FLAG_NEED_RTNL)) > - rtnl_unlock(); > + __rtnl_unlock(); > > return 0; > out_unlock: > -- > 2.53.0 >
ERROR: modpost: "__rtnl_unlock" [net/wireless/cfg80211.ko] undefined! Gonna re-run local nipa ingest locally to see if anything else pops up. --- pw-bot: cr
