nl80211: fix locking for wireless device netns change

Commit Message

Johannes Berg March 10, 2021, 8:58 p.m. UTC

From: Johannes Berg <[e-mail]>

We have all the network interfaces marked as netns-local
since the only reasonable thing to do right now is to set
a whole device, including all netdevs, into a different
network namespace. For this reason, we also have our own
way of changing the network namespace.

Unfortunately, the RTNL locking changes broke this, and
it now results in many RTNL assertions. The trivial fix
for those (just hold RTNL for the changes) however leads
to deadlocks in the cfg80211 netdev notifier.

Since we only need the wiphy, and that's still protected
by the RTNL, add a new NL80211_FLAG_NO_WIPHY_MTX flag to
the nl80211 ops and use it to _not_ take the wiphy mutex
but only the RTNL. This way, the notifier does all the
work necessary during unregistration/registration of the
netdevs from the old and in the new namespace.

Reported-by: Sid Hayn <[e-mail]>
Fixes: a05829a7222e ("cfg80211: avoid holding the RTNL when calling the driver")
Signed-off-by: Johannes Berg <[e-mail]>
---
 net/wireless/nl80211.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

Patch

diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index 521d36bb0803..ae6097fff133 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -14789,6 +14789,7 @@  static int nl80211_set_tid_config(struct sk_buff *skb,
 #define NL80211_FLAG_NEED_WDEV_UP	(NL80211_FLAG_NEED_WDEV |\
 					 NL80211_FLAG_CHECK_NETDEV_UP)
 #define NL80211_FLAG_CLEAR_SKB		0x20
+#define NL80211_FLAG_NO_WIPHY_MTX	0x40
 
 static int nl80211_pre_doit(const struct genl_ops *ops, struct sk_buff *skb,
 			    struct genl_info *info)
@@ -14840,7 +14841,7 @@  static int nl80211_pre_doit(const struct genl_ops *ops, struct sk_buff *skb,
 		info->user_ptr[0] = rdev;
 	}
 
-	if (rdev) {
+	if (rdev && !(ops->internal_flags & NL80211_FLAG_NO_WIPHY_MTX)) {
 		wiphy_lock(&rdev->wiphy);
 		/* we keep the mutex locked until post_doit */
 		__release(&rdev->wiphy.mtx);
@@ -14865,7 +14866,8 @@  static void nl80211_post_doit(const struct genl_ops *ops, struct sk_buff *skb,
 		}
 	}
 
-	if (info->user_ptr[0]) {
+	if (info->user_ptr[0] &&
+	    !(ops->internal_flags & NL80211_FLAG_NO_WIPHY_MTX)) {
 		struct cfg80211_registered_device *rdev = info->user_ptr[0];
 
 		/* we kept the mutex locked since pre_doit */
@@ -15329,7 +15331,9 @@  static const struct genl_small_ops nl80211_small_ops[] = {
 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
 		.doit = nl80211_wiphy_netns,
 		.flags = GENL_UNS_ADMIN_PERM,
-		.internal_flags = NL80211_FLAG_NEED_WIPHY,
+		.internal_flags = NL80211_FLAG_NEED_WIPHY |
+				  NL80211_FLAG_NEED_RTNL |
+				  NL80211_FLAG_NO_WIPHY_MTX,
 	},
 	{
 		.cmd = NL80211_CMD_GET_SURVEY,