Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Ziyang Xuan <william.xuanziyang@huawei.com>
To:     <mani@kernel.org>, <davem@davemloft.net>, <edumazet@google.com>,
        <kuba@kernel.org>, <pabeni@redhat.com>,
        <linux-arm-msm@vger.kernel.org>, <netdev@vger.kernel.org>
CC:     <andersson@kernel.org>, <linux-kernel@vger.kernel.org>
Subject: [PATCH net] net: qrtr: Fix a refcount bug in qrtr_recvmsg()
Date:   Thu, 30 Mar 2023 09:25:32 +0800
Message-ID: <20230330012532.1065159-1-william.xuanziyang@huawei.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 7BIT
Content-Type:   text/plain; charset=US-ASCII
Precedence: bulk

Syzbot reported a bug as following:

refcount_t: addition on 0; use-after-free.
...
RIP: 0010:refcount_warn_saturate+0x17c/0x1f0 lib/refcount.c:25
...
Call Trace:
 <TASK>
 __refcount_add include/linux/refcount.h:199 [inline]
 __refcount_inc include/linux/refcount.h:250 [inline]
 refcount_inc include/linux/refcount.h:267 [inline]
 kref_get include/linux/kref.h:45 [inline]
 qrtr_node_acquire net/qrtr/af_qrtr.c:202 [inline]
 qrtr_node_lookup net/qrtr/af_qrtr.c:398 [inline]
 qrtr_send_resume_tx net/qrtr/af_qrtr.c:1003 [inline]
 qrtr_recvmsg+0x85f/0x990 net/qrtr/af_qrtr.c:1070
 sock_recvmsg_nosec net/socket.c:1017 [inline]
 sock_recvmsg+0xe2/0x160 net/socket.c:1038
 qrtr_ns_worker+0x170/0x1700 net/qrtr/ns.c:688
 process_one_work+0x991/0x15c0 kernel/workqueue.c:2390
 worker_thread+0x669/0x1090 kernel/workqueue.c:2537

It occurs in the concurrent scenario of qrtr_recvmsg() and
qrtr_endpoint_unregister() as following:

	cpu0					cpu1
qrtr_recvmsg				qrtr_endpoint_unregister
qrtr_send_resume_tx			qrtr_node_release
qrtr_node_lookup			mutex_lock(&qrtr_node_lock)
spin_lock_irqsave(&qrtr_nodes_lock, )	refcount_dec_and_test(&node->ref) [node->ref == 0]
radix_tree_lookup [node != NULL]	__qrtr_node_release
qrtr_node_acquire			spin_lock_irqsave(&qrtr_nodes_lock, )
kref_get(&node->ref) [WARNING]		...
					mutex_unlock(&qrtr_node_lock)

Use qrtr_node_lock to protect qrtr_node_lookup() implementation, this
is actually improving the protection of node reference.

Fixes: 0a7e0d0ef054 ("net: qrtr: Migrate node lookup tree to spinlock")
Reported-by: syzbot+a7492efaa5d61b51db23@syzkaller.appspotmail.com
Link: https://syzkaller.appspot.com/bug?extid=a7492efaa5d61b51db23
Signed-off-by: Ziyang Xuan <william.xuanziyang@huawei.com>
---
 net/qrtr/af_qrtr.c | 2 ++
 1 file changed, 2 insertions(+)

diff --git a/net/qrtr/af_qrtr.c b/net/qrtr/af_qrtr.c
index 5c2fb992803b..3a70255c8d02 100644
--- a/net/qrtr/af_qrtr.c
+++ b/net/qrtr/af_qrtr.c
@@ -393,10 +393,12 @@ static struct qrtr_node *qrtr_node_lookup(unsigned int nid)
 	struct qrtr_node *node;
 	unsigned long flags;
 
+	mutex_lock(&qrtr_node_lock);
 	spin_lock_irqsave(&qrtr_nodes_lock, flags);
 	node = radix_tree_lookup(&qrtr_nodes, nid);
 	node = qrtr_node_acquire(node);
 	spin_unlock_irqrestore(&qrtr_nodes_lock, flags);
+	mutex_unlock(&qrtr_node_lock);
 
 	return node;
 }
-- 
2.25.1