Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     linux-kernel@vger.kernel.org
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        stable@vger.kernel.org, Guillaume Nault <gnault@redhat.com>,
        "David S. Miller" <davem@davemloft.net>,
        =?UTF-8?q?H=C3=A5kon=20Bugge?= <haakon.bugge@oracle.com>
Subject: [PATCH 5.4 062/144] netns: protect netns ID lookups with RCU
Date:   Mon, 13 Sep 2021 15:14:03 +0200
Message-Id: <20210913131050.045276595@linuxfoundation.org>
In-Reply-To: <20210913131047.974309396@linuxfoundation.org>
References: <20210913131047.974309396@linuxfoundation.org>
User-Agent: quilt/0.66
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Precedence: bulk

From: Guillaume Nault <gnault@redhat.com>

commit 2dce224f469f060b9998a5a869151ef83c08ce77 upstream.

__peernet2id() can be protected by RCU as it only calls idr_for_each(),
which is RCU-safe, and never modifies the nsid table.

rtnl_net_dumpid() can also do lockless lookups. It does two nested
idr_for_each() calls on nsid tables (one direct call and one indirect
call because of rtnl_net_dumpid_one() calling __peernet2id()). The
netnsid tables are never updated. Therefore it is safe to not take the
nsid_lock and run within an RCU-critical section instead.

Signed-off-by: Guillaume Nault <gnault@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com>

---
 net/core/net_namespace.c |   28 ++++++++++------------------
 1 file changed, 10 insertions(+), 18 deletions(-)

--- a/net/core/net_namespace.c
+++ b/net/core/net_namespace.c
@@ -211,9 +211,9 @@ static int net_eq_idr(int id, void *net,
 	return 0;
 }
 
-/* Should be called with nsid_lock held. If a new id is assigned, the bool alloc
- * is set to true, thus the caller knows that the new id must be notified via
- * rtnl.
+/* Must be called from RCU-critical section or with nsid_lock held. If
+ * a new id is assigned, the bool alloc is set to true, thus the
+ * caller knows that the new id must be notified via rtnl.
  */
 static int __peernet2id_alloc(struct net *net, struct net *peer, bool *alloc)
 {
@@ -237,7 +237,7 @@ static int __peernet2id_alloc(struct net
 	return NETNSA_NSID_NOT_ASSIGNED;
 }
 
-/* should be called with nsid_lock held */
+/* Must be called from RCU-critical section or with nsid_lock held */
 static int __peernet2id(struct net *net, struct net *peer)
 {
 	bool no = false;
@@ -281,9 +281,10 @@ int peernet2id(struct net *net, struct n
 {
 	int id;
 
-	spin_lock_bh(&net->nsid_lock);
+	rcu_read_lock();
 	id = __peernet2id(net, peer);
-	spin_unlock_bh(&net->nsid_lock);
+	rcu_read_unlock();
+
 	return id;
 }
 EXPORT_SYMBOL(peernet2id);
@@ -962,6 +963,7 @@ struct rtnl_net_dump_cb {
 	int s_idx;
 };
 
+/* Runs in RCU-critical section. */
 static int rtnl_net_dumpid_one(int id, void *peer, void *data)
 {
 	struct rtnl_net_dump_cb *net_cb = (struct rtnl_net_dump_cb *)data;
@@ -1046,19 +1048,9 @@ static int rtnl_net_dumpid(struct sk_buf
 			goto end;
 	}
 
-	spin_lock_bh(&net_cb.tgt_net->nsid_lock);
-	if (net_cb.fillargs.add_ref &&
-	    !net_eq(net_cb.ref_net, net_cb.tgt_net) &&
-	    !spin_trylock_bh(&net_cb.ref_net->nsid_lock)) {
-		spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
-		err = -EAGAIN;
-		goto end;
-	}
+	rcu_read_lock();
 	idr_for_each(&net_cb.tgt_net->netns_ids, rtnl_net_dumpid_one, &net_cb);
-	if (net_cb.fillargs.add_ref &&
-	    !net_eq(net_cb.ref_net, net_cb.tgt_net))
-		spin_unlock_bh(&net_cb.ref_net->nsid_lock);
-	spin_unlock_bh(&net_cb.tgt_net->nsid_lock);
+	rcu_read_unlock();
 
 	cb->args[0] = net_cb.idx;
 end: