[PATCH v2] Bluetooth: fix inconsistent lock state in sco
From: Desmond Cheong Zhi Xi <hidden>
Date: 2021-07-13 16:29:17
Also in:
linux-bluetooth, linux-kernel-mentees, lkml
Subsystem:
bluetooth subsystem, the rest · Maintainers:
Marcel Holtmann, Luiz Augusto von Dentz, Linus Torvalds
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@syzkaller.appspotmail.com
Tested-by: syzbot+2f6d7c28bb4bf7e82060@syzkaller.appspotmail.com
Signed-off-by: Desmond Cheong Zhi Xi <redacted>
---
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(); 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