Re: [PATCH v2] Bluetooth: Fix possible deadlock in rfcomm_sk_state_change

From: Luiz Augusto von Dentz
Date: Wed Jan 04 2023 - 17:22:18 EST


Hi Ying,

On Wed, Jan 4, 2023 at 7:07 AM Ying Hsu <yinghsu@xxxxxxxxxxxx> wrote:
>
> There's a possible deadlock when two processes are connecting
> and closing a RFCOMM socket concurrently. Here's the call trace:

Are you sure it is 2 different processes? Usually that would mean 2
different sockets (sk) then so they wouldn't share the same lock, so
this sounds more like 2 different threads, perhaps it is worth
creating a testing case in our rfcomm-tester so we are able to detect
this sort of thing in the future.

> -> #2 (&d->lock){+.+.}-{3:3}:
> __mutex_lock_common kernel/locking/mutex.c:603 [inline]
> __mutex_lock0x12f/0x1360 kernel/locking/mutex.c:747
> __rfcomm_dlc_close+0x15d/0x890 net/bluetooth/rfcomm/core.c:487
> rfcomm_dlc_close+1e9/0x240 net/bluetooth/rfcomm/core.c:520
> __rfcomm_sock_close+0x13c/0x250 net/bluetooth/rfcomm/sock.c:220
> rfcomm_sock_shutdown+0xd8/0x230 net/bluetooth/rfcomm/sock.c:907
> rfcomm_sock_release+0x68/0x140 net/bluetooth/rfcomm/sock.c:928
> __sock_release+0xcd/0x280 net/socket.c:650
> sock_close+0x1c/0x20 net/socket.c:1365
> __fput+0x27c/0xa90 fs/file_table.c:320
> task_work_run+0x16f/0x270 kernel/task_work.c:179
> exit_task_work include/linux/task_work.h:38 [inline]
> do_exit+0xaa8/0x2950 kernel/exit.c:867
> do_group_exit+0xd4/0x2a0 kernel/exit.c:1012
> get_signal+0x21c3/0x2450 kernel/signal.c:2859
> arch_do_signal_or_restart+0x79/0x5c0 arch/x86/kernel/signal.c:306
> exit_to_user_mode_loop kernel/entry/common.c:168 [inline]
> exit_to_user_mode_prepare+0x15f/0x250 kernel/entry/common.c:203
> __syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
> syscall_exit_to_user_mode+0x1d/0x50 kernel/entry/common.c:296
> do_syscall_64+0x46/0xb0 arch/x86/entry/common.c:86
> entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
> -> #1 (rfcomm_mutex){+.+.}-{3:3}:
> __mutex_lock_common kernel/locking/mutex.c:603 [inline]
> __mutex_lock+0x12f/0x1360 kernel/locking/mutex.c:747
> rfcomm_dlc_open+0x93/0xa80 net/bluetooth/rfcomm/core.c:425
> rfcomm_sock_connect+0x329/0x450 net/bluetooth/rfcomm/sock.c:413
> __sys_connect_file+0x153/0x1a0 net/socket.c:1976
> __sys_connect+0x165/0x1a0 net/socket.c:1993
> __do_sys_connect net/socket.c:2003 [inline]
> __se_sys_connect net/socket.c:2000 [inline]
> __x64_sys_connect+0x73/0xb0 net/socket.c:2000
> do_syscall_x64 arch/x86/entry/common.c:50 [inline]
> do_syscall_64+0x39/0xb0 arch/x86/entry/common.c:80
> entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
> -> #0 (sk_lock-AF_BLUETOOTH-BTPROTO_RFCOMM){+.+.}-{0:0}:
> check_prev_add kernel/locking/lockdep.c:3097 [inline]
> check_prevs_add kernel/locking/lockdep.c:3216 [inline]
> validate_chain kernel/locking/lockdep.c:3831 [inline]
> __lock_acquire+0x2a43/0x56d0 kernel/locking/lockdep.c:5055
> lock_acquire kernel/locking/lockdep.c:5668 [inline]
> lock_acquire+0x1e3/0x630 kernel/locking/lockdep.c:5633
> lock_sock_nested+0x3a/0xf0 net/core/sock.c:3470
> lock_sock include/net/sock.h:1725 [inline]
> rfcomm_sk_state_change+0x6d/0x3a0 net/bluetooth/rfcomm/sock.c:73
> __rfcomm_dlc_close+0x1b1/0x890 net/bluetooth/rfcomm/core.c:489
> rfcomm_dlc_close+0x1e9/0x240 net/bluetooth/rfcomm/core.c:520
> __rfcomm_sock_close+0x13c/0x250 net/bluetooth/rfcomm/sock.c:220
> rfcomm_sock_shutdown+0xd8/0x230 net/bluetooth/rfcomm/sock.c:907
> rfcomm_sock_release+0x68/0x140 net/bluetooth/rfcomm/sock.c:928
> __sock_release+0xcd/0x280 net/socket.c:650
> sock_close+0x1c/0x20 net/socket.c:1365
> __fput+0x27c/0xa90 fs/file_table.c:320
> task_work_run+0x16f/0x270 kernel/task_work.c:179
> exit_task_work include/linux/task_work.h:38 [inline]
> do_exit+0xaa8/0x2950 kernel/exit.c:867
> do_group_exit+0xd4/0x2a0 kernel/exit.c:1012
> get_signal+0x21c3/0x2450 kernel/signal.c:2859
> arch_do_signal_or_restart+0x79/0x5c0 arch/x86/kernel/signal.c:306
> exit_to_user_mode_loop kernel/entry/common.c:168 [inline]
> exit_to_user_mode_prepare+0x15f/0x250 kernel/entry/common.c:203
> __syscall_exit_to_user_mode_work kernel/entry/common.c:285 [inline]
> syscall_exit_to_user_mode+0x1d/0x50 kernel/entry/common.c:296
> do_syscall_64+0x46/0xb0 arch/x86/entry/common.c:86
> entry_SYSCALL_64_after_hwframe+0x63/0xcd
>
> Signed-off-by: Ying Hsu <yinghsu@xxxxxxxxxxxx>
> ---
> This commit has been tested with a C reproducer on qemu-x86_64
> and a ChromeOS device.
>
> 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;
> }

This sounds like a great solution to hold the reference and then
checking if the socket has been zapped when attempting to lock_sock,
so Ive been thinking on generalize this into something like
bt_sock_connect(sock, addr, alen, callback) so we make sure the
callback is done while holding a reference but with the socket
unlocked since typically the underline procedure only needs to access
the pi(sk) information without changing it e.g. rfcomm_dlc_open,
anyway Im fine if you don't want to pursue doing it right now but I'm
afraid these type of locking problem is no restricted to RFCOMM only.

> --
> 2.39.0.314.g84b9a713c41-goog
>


--
Luiz Augusto von Dentz