Skip to contentSkip to conversation

net: rtnetlink: Fix rtnl_dereference may be return NULL

PATCH In review Yajun Deng 4 participants 4 messages Jul 8, 2021

Subsystems (2)
lore
Share & tools
lore

Yajun Deng

author
#0 lore raw

PATCH v2 1/1 net: rtnetlink: Fix rtnl_dereference may be return NULL

The value 'link' may be NULL in rtnl_unregister(), this leads to kfree_rcu(NULL, xxx), so add this case handling. And modify the return value to 'void' in rtnl_unregister(). there is no case using it.

Fixes addf9b90de22 (net: rtnetlink: use rcu to free rtnl message handlers)
Fixes 51e13685bd93 (rtnetlink: RCU-annotate both dimensions of rtnl_msg_handlers)
---
 include/net/rtnetlink.h |  2 +-
 net/core/rtnetlink.c    | 18 ++++++++----------
 2 files changed, 9 insertions(+), 11 deletions(-)
include/net/rtnetlink.h +1 -1
index 384e800665f2..9d263ad9ea48 100644
@@ -17,7 +17,7 @@ void rtnl_register(int protocol, int msgtype,
 		   rtnl_doit_func, rtnl_dumpit_func, unsigned int flags);
 int rtnl_register_module(struct module *owner, int protocol, int msgtype,
 			 rtnl_doit_func, rtnl_dumpit_func, unsigned int flags);
-int rtnl_unregister(int protocol, int msgtype);
+void rtnl_unregister(int protocol, int msgtype);
 void rtnl_unregister_all(int protocol);
 
 static inline int rtnl_msg_family(const struct nlmsghdr *nlh)
net/core/rtnetlink.c +8 -10
index f6af3e74fc44..e80177c195a5 100644
@@ -281,10 +281,8 @@ void rtnl_register(int protocol, int msgtype,
  * rtnl_unregister - Unregister a rtnetlink message type
  * @protocol: Protocol family or PF_UNSPEC
  * @msgtype: rtnetlink message type
- *
- * Returns 0 on success or a negative error code.
  */
-int rtnl_unregister(int protocol, int msgtype)
+void rtnl_unregister(int protocol, int msgtype)
 {
 	struct rtnl_link __rcu **tab;
 	struct rtnl_link *link;
@@ -295,18 +293,18 @@ int rtnl_unregister(int protocol, int msgtype)
 
 	rtnl_lock();
 	tab = rtnl_dereference(rtnl_msg_handlers[protocol]);
-	if (!tab) {
-		rtnl_unlock();
-		return -ENOENT;
-	}
+	if (!tab)
+		goto unlock;
 
 	link = rtnl_dereference(tab[msgindex]);
-	rcu_assign_pointer(tab[msgindex], NULL);
-	rtnl_unlock();
+	if (!link)
+		goto unlock;
 
+	rcu_assign_pointer(tab[msgindex], NULL);
 	kfree_rcu(link, rcu);
 
-	return 0;
+unlock:
+	rtnl_unlock();
 }
 EXPORT_SYMBOL_GPL(rtnl_unregister);
 
Signature · 1 line
2.32.0
#1 lore raw
Quoted 3 lines · Yajun Deng

The value 'link' may be NULL in rtnl_unregister(), this leads to kfree_rcu(NULL, xxx), so add this case handling.

I don't see how. It would require the caller to unregister something they never registered. That would be a bug there, but I don't see that it's very useful to actually be defensive about bugs there.

Quoted 5 lines

And modify the return
value to 'void' in rtnl_unregister(). there is no case using it.

Fixes addf9b90de22 (net: rtnetlink: use rcu to free rtnl message handlers)
Fixes 51e13685bd93 (rtnetlink: RCU-annotate both dimensions of rtnl_msg_handlers)

It certainly fixes nothing in those patches.

johannes

#2 lore raw
Quoted 8 lines · Johannes Berg
Quoted 3 lines · Yajun Deng

The value 'link' may be NULL in rtnl_unregister(), this leads to kfree_rcu(NULL, xxx), so add this case handling.

I don't see how. It would require the caller to unregister something they never registered. That would be a bug there, but I don't see that it's very useful to actually be defensive about bugs there.

Besides, isn't kfree_rcu(NULL) safe anyway?

#3 lore raw
Quoted 6 lines · Vladimir Oltean
Quoted 5 lines · Johannes Berg
Quoted 3 lines · Yajun Deng

The value 'link' may be NULL in rtnl_unregister(), this leads to kfree_rcu(NULL, xxx), so add this case handling.

Quoted 6 lines
Quoted 3 lines

I don't see how. It would require the caller to unregister something they never registered. That would be a bug there, but I don't see that it's very useful to actually be defensive about bugs there.

Besides, isn't kfree_rcu(NULL) safe anyway?

Only from linux-5.3 I think.
(commit 12edff045bc6dd3ab1565cc02fa4841803c2a633 was not backported to old kernels)

But yes, this patch is not solving any bug, as I suspected.