[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CABBYNZLBfH+0=yhgcAK4XzizUKqpmAxjyxGpBACiFZpPsr0CEQ@mail.gmail.com>
Date: Wed, 14 Jul 2021 12:12:33 -0700
From: Luiz Augusto von Dentz <luiz.dentz@...il.com>
To: Desmond Cheong Zhi Xi <desmondcheongzx@...il.com>
Cc: Marcel Holtmann <marcel@...tmann.org>,
Johan Hedberg <johan.hedberg@...il.com>,
David Miller <davem@...emloft.net>,
Jakub Kicinski <kuba@...nel.org>, stefan@...enfreihafen.org,
"linux-bluetooth@...r.kernel.org" <linux-bluetooth@...r.kernel.org>,
"open list:NETWORKING [GENERAL]" <netdev@...r.kernel.org>,
Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
skhan@...uxfoundation.org,
Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
linux-kernel-mentees@...ts.linuxfoundation.org,
syzbot+2f6d7c28bb4bf7e82060@...kaller.appspotmail.com
Subject: Re: [PATCH v2] Bluetooth: fix inconsistent lock state in sco
Hi Desmond,
On Tue, Jul 13, 2021 at 9:29 AM Desmond Cheong Zhi Xi
<desmondcheongzx@...il.com> wrote:
>
> Syzbot reported an inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} lock
> usage in sco_conn_del and sco_sock_timeout that could lead to
> deadlocks.
>
> This inconsistent lock state can also happen in sco_conn_ready,
> rfcomm_connect_ind, and bt_accept_enqueue.
>
> The issue is that these functions take a spin lock on the socket with
> interrupts enabled, but sco_sock_timeout takes the lock in an IRQ
> context. This could lead to deadlocks:
>
> CPU0
> ----
> lock(slock-AF_BLUETOOTH-BTPROTO_SCO);
> <Interrupt>
> lock(slock-AF_BLUETOOTH-BTPROTO_SCO);
>
> *** DEADLOCK ***
>
> We fix this by ensuring that local bh is disabled before calling
> bh_lock_sock.
>
> After doing this, we additionally need to protect sco_conn_lock by
> disabling local bh.
>
> This is necessary because sco_conn_del makes a call to sco_chan_del
> while holding on to the sock lock, and sco_chan_del itself makes a
> call to sco_conn_lock. If sco_conn_lock is held elsewhere with
> interrupts enabled, there could still be a
> slock-AF_BLUETOOTH-BTPROTO_SCO --> &conn->lock#2 lock inversion as
> follows:
>
> CPU0 CPU1
> ---- ----
> lock(&conn->lock#2);
> local_irq_disable();
> lock(slock-AF_BLUETOOTH-BTPROTO_SCO);
> lock(&conn->lock#2);
> <Interrupt>
> lock(slock-AF_BLUETOOTH-BTPROTO_SCO);
>
> *** DEADLOCK ***
>
> As sco_conn_del now disables local bh before calling sco_chan_del,
> instead of disabling local bh for the calls to sco_conn_lock in
> sco_chan_del, we instead wrap other calls to sco_chan_del with
> local_bh_disable/enable.
>
> Reported-by: syzbot+2f6d7c28bb4bf7e82060@...kaller.appspotmail.com
> Tested-by: syzbot+2f6d7c28bb4bf7e82060@...kaller.appspotmail.com
> Signed-off-by: Desmond Cheong Zhi Xi <desmondcheongzx@...il.com>
> ---
>
> Hi,
>
> The previous version of this patch was a bit of a mess, so I made the
> following changes.
>
> v1 -> v2:
> - Instead of pulling out the clean-up code out from sco_chan_del and
> using it directly in sco_conn_del, disable local irqs for relevant
> sections.
> - Disable local irqs more thoroughly for instances of
> bh_lock_sock/bh_lock_sock_nested in the bluetooth subsystem.
> Specifically, the calls in af_bluetooth.c and rfcomm/sock.c are now made
> with local irqs disabled as well.
>
> Best wishes,
> Desmond
>
> net/bluetooth/rfcomm/sock.c | 2 ++
> net/bluetooth/sco.c | 26 +++++++++++++++++++++++++-
> 2 files changed, 27 insertions(+), 1 deletion(-)
>
> diff --git a/net/bluetooth/rfcomm/sock.c b/net/bluetooth/rfcomm/sock.c
> index ae6f80730561..d8734abb2df4 100644
> --- a/net/bluetooth/rfcomm/sock.c
> +++ b/net/bluetooth/rfcomm/sock.c
> @@ -974,6 +974,7 @@ int rfcomm_connect_ind(struct rfcomm_session *s, u8 channel, struct rfcomm_dlc *
> if (!parent)
> return 0;
>
> + local_bh_disable();
> bh_lock_sock(parent);
>
> /* Check for backlog size */
> @@ -1002,6 +1003,7 @@ int rfcomm_connect_ind(struct rfcomm_session *s, u8 channel, struct rfcomm_dlc *
>
> done:
> bh_unlock_sock(parent);
> + local_bh_enable();
Looks like you are touching RFCOMM as well, perhaps you should have it
split, also how about other sockets like L2CAP and HCI are they
affected? There seems to be a lot of problem with the likes of
bh_lock_sock I wonder if going with local_bh_disable is overall a
better way to handle.
> if (test_bit(BT_SK_DEFER_SETUP, &bt_sk(parent)->flags))
> parent->sk_state_change(parent);
> diff --git a/net/bluetooth/sco.c b/net/bluetooth/sco.c
> index 3bd41563f118..2548b8f81473 100644
> --- a/net/bluetooth/sco.c
> +++ b/net/bluetooth/sco.c
> @@ -167,16 +167,22 @@ static void sco_conn_del(struct hci_conn *hcon, int err)
> BT_DBG("hcon %p conn %p, err %d", hcon, conn, err);
>
> /* Kill socket */
> + local_bh_disable();
> sco_conn_lock(conn);
> sk = conn->sk;
> sco_conn_unlock(conn);
> + local_bh_enable();
>
> if (sk) {
> sock_hold(sk);
> +
> + local_bh_disable();
> bh_lock_sock(sk);
> sco_sock_clear_timer(sk);
> sco_chan_del(sk, err);
> bh_unlock_sock(sk);
> + local_bh_enable();
> +
> sco_sock_kill(sk);
> sock_put(sk);
> }
> @@ -202,6 +208,7 @@ static int sco_chan_add(struct sco_conn *conn, struct sock *sk,
> {
> int err = 0;
>
> + local_bh_disable();
> sco_conn_lock(conn);
> if (conn->sk)
> err = -EBUSY;
> @@ -209,6 +216,7 @@ static int sco_chan_add(struct sco_conn *conn, struct sock *sk,
> __sco_chan_add(conn, sk, parent);
>
> sco_conn_unlock(conn);
> + local_bh_enable();
> return err;
> }
>
> @@ -303,9 +311,11 @@ static void sco_recv_frame(struct sco_conn *conn, struct sk_buff *skb)
> {
> struct sock *sk;
>
> + local_bh_disable();
> sco_conn_lock(conn);
> sk = conn->sk;
> sco_conn_unlock(conn);
> + local_bh_enable();
>
> if (!sk)
> goto drop;
> @@ -420,18 +430,25 @@ static void __sco_sock_close(struct sock *sk)
> if (sco_pi(sk)->conn->hcon) {
> sk->sk_state = BT_DISCONN;
> sco_sock_set_timer(sk, SCO_DISCONN_TIMEOUT);
> + local_bh_disable();
> sco_conn_lock(sco_pi(sk)->conn);
> hci_conn_drop(sco_pi(sk)->conn->hcon);
> sco_pi(sk)->conn->hcon = NULL;
> sco_conn_unlock(sco_pi(sk)->conn);
> - } else
> + local_bh_enable();
> + } else {
> + local_bh_disable();
> sco_chan_del(sk, ECONNRESET);
> + local_bh_enable();
> + }
> break;
>
> case BT_CONNECT2:
> case BT_CONNECT:
> case BT_DISCONN:
> + local_bh_disable();
> sco_chan_del(sk, ECONNRESET);
> + local_bh_enable();
> break;
>
> default:
> @@ -1084,21 +1101,26 @@ static void sco_conn_ready(struct sco_conn *conn)
>
> if (sk) {
> sco_sock_clear_timer(sk);
> + local_bh_disable();
> bh_lock_sock(sk);
> sk->sk_state = BT_CONNECTED;
> sk->sk_state_change(sk);
> bh_unlock_sock(sk);
> + local_bh_enable();
> } else {
> + local_bh_disable();
> sco_conn_lock(conn);
>
> if (!conn->hcon) {
> sco_conn_unlock(conn);
> + local_bh_enable();
> return;
> }
>
> parent = sco_get_sock_listen(&conn->hcon->src);
> if (!parent) {
> sco_conn_unlock(conn);
> + local_bh_enable();
> return;
> }
>
> @@ -1109,6 +1131,7 @@ static void sco_conn_ready(struct sco_conn *conn)
> if (!sk) {
> bh_unlock_sock(parent);
> sco_conn_unlock(conn);
> + local_bh_enable();
> return;
> }
>
> @@ -1131,6 +1154,7 @@ static void sco_conn_ready(struct sco_conn *conn)
> bh_unlock_sock(parent);
>
> sco_conn_unlock(conn);
> + local_bh_enable();
> }
> }
>
> --
> 2.25.1
>
--
Luiz Augusto von Dentz
Powered by blists - more mailing lists