[PATCH net] net/sched: act_api: use mutex in tcf_idr_check_alloc

[Kyle Zeng <kylebot@openai.com>]

Currently, the NEWTFILTER path uses RCU to guard action idr accesses while
the DELTFILTER path uses mutex to guard action accesses. This
inconsistency leads to a race condition scenario, which can lead to
erroneous operations on refcount, eventually leading to use-after-free
situation.
In this patch, we revert the introduction of RCU back to mutex in the
NEWFILTER path, which is consistent with the DELFILTER path, avoiding
the race condition.

Fixes: 4b55e86736d5 ("net/sched: act_api: rely on rcu in tcf_idr_check_alloc")
Signed-off-by: Kyle Zeng <kylebot@openai.com>
---
 net/sched/act_api.c | 12 +++++-------
 1 file changed, 5 insertions(+), 7 deletions(-)

diff --git a/net/sched/act_api.c b/net/sched/act_api.c
index 332fd9695e54..63841c3e4310 100644
--- a/net/sched/act_api.c
+++ b/net/sched/act_api.c
@@ -874,36 +874,34 @@ int tcf_idr_check_alloc(struct tc_action_net *tn, u32 *index,
 	u32 max;
 
 	if (*index) {
-		rcu_read_lock();
+		mutex_lock(&idrinfo->lock);
 		p = idr_find(&idrinfo->action_idr, *index);
 
 		if (IS_ERR(p)) {
 			/* This means that another process allocated
 			 * index but did not assign the pointer yet.
 			 */
-			rcu_read_unlock();
+			mutex_unlock(&idrinfo->lock);
 			return -EAGAIN;
 		}
 
 		if (!p) {
 			/* Empty slot, try to allocate it */
 			max = *index;
-			rcu_read_unlock();
+			mutex_unlock(&idrinfo->lock);
 			goto new;
 		}
 
 		if (!refcount_inc_not_zero(&p->tcfa_refcnt)) {
 			/* Action was deleted in parallel */
-			rcu_read_unlock();
+			mutex_unlock(&idrinfo->lock);
 			return -EAGAIN;
 		}
 
 		if (bind)
 			atomic_inc(&p->tcfa_bindcnt);
 		*a = p;
-
-		rcu_read_unlock();
-
+		mutex_unlock(&idrinfo->lock);
 		return 1;
 	} else {
 		/* Find a slot */
-- 
2.54.0