Received-SPF: pass (google.com: domain of linux-bluetooth-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
From:   Ying Hsu <yinghsu@chromium.org>
To:     linux-bluetooth@vger.kernel.org, marcel@holtmann.org,
        leon@kernel.org, luiz.dentz@gmail.com
Cc:     chromeos-bluetooth-upstreaming@chromium.org,
        Ying Hsu <yinghsu@chromium.org>,
        "David S. Miller" <davem@davemloft.net>,
        Eric Dumazet <edumazet@google.com>,
        Jakub Kicinski <kuba@kernel.org>,
        Johan Hedberg <johan.hedberg@gmail.com>,
        Paolo Abeni <pabeni@redhat.com>,
        Tedd Ho-Jeong An <tedd.an@intel.com>,
        linux-kernel@vger.kernel.org, netdev@vger.kernel.org
Subject: [PATCH v3] Bluetooth: Fix possible deadlock in rfcomm_sk_state_change
Date:   Wed, 11 Jan 2023 03:16:14 +0000
Message-Id: <20230111031540.v3.1.I1f29bb547a03e9adfe2e6754212f9d14a2e39c4b@changeid>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Precedence: bulk

syzbot reports a possible deadlock in rfcomm_sk_state_change [1].
While rfcomm_sock_connect acquires the sk lock and waits for
the rfcomm lock, rfcomm_sock_release could have the rfcomm
lock and hit a deadlock for acquiring the sk lock.
Here's a simplified flow:

rfcomm_sock_connect:
  lock_sock(sk)
  rfcomm_dlc_open:
    rfcomm_lock()

rfcomm_sock_release:
  rfcomm_sock_shutdown:
    rfcomm_lock()
    __rfcomm_dlc_close:
        rfcomm_k_state_change:
	  lock_sock(sk)

This patch drops the sk lock before calling rfcomm_dlc_open to
avoid the possible deadlock and holds sk's reference count to
prevent use-after-free after rfcomm_dlc_open completes.

Reported-by: syzbot+d7ce59...@syzkaller.appspotmail.com
Fixes: 1804fdf6e494 ("Bluetooth: btintel: Combine setting up MSFT extension")
Link: https://syzkaller.appspot.com/bug?extid=d7ce59b06b3eb14fd218 [1]

Signed-off-by: Ying Hsu <yinghsu@chromium.org>
---
This commit has been tested with a C reproducer on qemu-x86_64
and a ChromeOS device.

Changes in v3:
- Revise the commit message.

Changes in v2:
- Fix potential use-after-free in rfc_comm_sock_connect.

 net/bluetooth/rfcomm/sock.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/net/bluetooth/rfcomm/sock.c b/net/bluetooth/rfcomm/sock.c
index 21e24da4847f..4397e14ff560 100644
--- a/net/bluetooth/rfcomm/sock.c
+++ b/net/bluetooth/rfcomm/sock.c
@@ -391,6 +391,7 @@ static int rfcomm_sock_connect(struct socket *sock, struct sockaddr *addr, int a
 	    addr->sa_family != AF_BLUETOOTH)
 		return -EINVAL;
 
+	sock_hold(sk);
 	lock_sock(sk);
 
 	if (sk->sk_state != BT_OPEN && sk->sk_state != BT_BOUND) {
@@ -410,14 +411,18 @@ static int rfcomm_sock_connect(struct socket *sock, struct sockaddr *addr, int a
 	d->sec_level = rfcomm_pi(sk)->sec_level;
 	d->role_switch = rfcomm_pi(sk)->role_switch;
 
+	/* Drop sock lock to avoid potential deadlock with the RFCOMM lock */
+	release_sock(sk);
 	err = rfcomm_dlc_open(d, &rfcomm_pi(sk)->src, &sa->rc_bdaddr,
 			      sa->rc_channel);
-	if (!err)
+	lock_sock(sk);
+	if (!err && !sock_flag(sk, SOCK_ZAPPED))
 		err = bt_sock_wait_state(sk, BT_CONNECTED,
 				sock_sndtimeo(sk, flags & O_NONBLOCK));
 
 done:
 	release_sock(sk);
+	sock_put(sk);
 	return err;
 }
 
-- 
2.39.0.314.g84b9a713c41-goog