Re: [PATCH RESEND] Bluetooth: Fix deadlock in vhci_send_frame

[Ying Hsu]

Hi Luiz and Arkadiusz,

I appreciate the effort you put into ensuring the quality of the
kernel, if it looks like the review might need more time, could we
consider reverting commit 92d4abd66f70 ("Bluetooth: vhci: Fix race
when opening vhci device") in the interim? This would help maintain
stability until the new patch is approved.

Best regards,
Ying


On Fri, Nov 10, 2023 at 9:46 AM Ying Hsu <yinghsu@xxxxxxxxxxxx> wrote:
>
> syzbot found a potential circular dependency leading to a deadlock:
>     -> #3 (&hdev->req_lock){+.+.}-{3:3}:
>     __mutex_lock_common+0x1b6/0x1bc2 kernel/locking/mutex.c:599
>     __mutex_lock kernel/locking/mutex.c:732 [inline]
>     mutex_lock_nested+0x17/0x1c kernel/locking/mutex.c:784
>     hci_dev_do_close+0x3f/0x9f net/bluetooth/hci_core.c:551
>     hci_rfkill_set_block+0x130/0x1ac net/bluetooth/hci_core.c:935
>     rfkill_set_block+0x1e6/0x3b8 net/rfkill/core.c:345
>     rfkill_fop_write+0x2d8/0x672 net/rfkill/core.c:1274
>     vfs_write+0x277/0xcf5 fs/read_write.c:594
>     ksys_write+0x19b/0x2bd fs/read_write.c:650
>     do_syscall_x64 arch/x86/entry/common.c:55 [inline]
>     do_syscall_64+0x51/0xba arch/x86/entry/common.c:93
>     entry_SYSCALL_64_after_hwframe+0x61/0xcb
>
>     -> #2 (rfkill_global_mutex){+.+.}-{3:3}:
>     __mutex_lock_common+0x1b6/0x1bc2 kernel/locking/mutex.c:599
>     __mutex_lock kernel/locking/mutex.c:732 [inline]
>     mutex_lock_nested+0x17/0x1c kernel/locking/mutex.c:784
>     rfkill_register+0x30/0x7e3 net/rfkill/core.c:1045
>     hci_register_dev+0x48f/0x96d net/bluetooth/hci_core.c:2622
>     __vhci_create_device drivers/bluetooth/hci_vhci.c:341 [inline]
>     vhci_create_device+0x3ad/0x68f drivers/bluetooth/hci_vhci.c:374
>     vhci_get_user drivers/bluetooth/hci_vhci.c:431 [inline]
>     vhci_write+0x37b/0x429 drivers/bluetooth/hci_vhci.c:511
>     call_write_iter include/linux/fs.h:2109 [inline]
>     new_sync_write fs/read_write.c:509 [inline]
>     vfs_write+0xaa8/0xcf5 fs/read_write.c:596
>     ksys_write+0x19b/0x2bd fs/read_write.c:650
>     do_syscall_x64 arch/x86/entry/common.c:55 [inline]
>     do_syscall_64+0x51/0xba arch/x86/entry/common.c:93
>     entry_SYSCALL_64_after_hwframe+0x61/0xcb
>
>     -> #1 (&data->open_mutex){+.+.}-{3:3}:
>     __mutex_lock_common+0x1b6/0x1bc2 kernel/locking/mutex.c:599
>     __mutex_lock kernel/locking/mutex.c:732 [inline]
>     mutex_lock_nested+0x17/0x1c kernel/locking/mutex.c:784
>     vhci_send_frame+0x68/0x9c drivers/bluetooth/hci_vhci.c:75
>     hci_send_frame+0x1cc/0x2ff net/bluetooth/hci_core.c:2989
>     hci_sched_acl_pkt net/bluetooth/hci_core.c:3498 [inline]
>     hci_sched_acl net/bluetooth/hci_core.c:3583 [inline]
>     hci_tx_work+0xb94/0x1a60 net/bluetooth/hci_core.c:3654
>     process_one_work+0x901/0xfb8 kernel/workqueue.c:2310
>     worker_thread+0xa67/0x1003 kernel/workqueue.c:2457
>     kthread+0x36a/0x430 kernel/kthread.c:319
>     ret_from_fork+0x1f/0x30 arch/x86/entry/entry_64.S:298
>
>     -> #0 ((work_completion)(&hdev->tx_work)){+.+.}-{0:0}:
>     check_prev_add kernel/locking/lockdep.c:3053 [inline]
>     check_prevs_add kernel/locking/lockdep.c:3172 [inline]
>     validate_chain kernel/locking/lockdep.c:3787 [inline]
>     __lock_acquire+0x2d32/0x77fa kernel/locking/lockdep.c:5011
>     lock_acquire+0x273/0x4d5 kernel/locking/lockdep.c:5622
>     __flush_work+0xee/0x19f kernel/workqueue.c:3090
>     hci_dev_close_sync+0x32f/0x1113 net/bluetooth/hci_sync.c:4352
>     hci_dev_do_close+0x47/0x9f net/bluetooth/hci_core.c:553
>     hci_rfkill_set_block+0x130/0x1ac net/bluetooth/hci_core.c:935
>     rfkill_set_block+0x1e6/0x3b8 net/rfkill/core.c:345
>     rfkill_fop_write+0x2d8/0x672 net/rfkill/core.c:1274
>     vfs_write+0x277/0xcf5 fs/read_write.c:594
>     ksys_write+0x19b/0x2bd fs/read_write.c:650
>     do_syscall_x64 arch/x86/entry/common.c:55 [inline]
>     do_syscall_64+0x51/0xba arch/x86/entry/common.c:93
>     entry_SYSCALL_64_after_hwframe+0x61/0xcb
>
> This change removes the need for acquiring the open_mutex in
> vhci_send_frame, thus eliminating the potential deadlock while
> maintaining the required packet ordering.
>
> Fixes: 92d4abd66f70 ("Bluetooth: vhci: Fix race when opening vhci device")
> Signed-off-by: Ying Hsu <yinghsu@xxxxxxxxxxxx>
> ---
> Tested this commit using a C reproducer on qemu-x86_64.
>
>  drivers/bluetooth/hci_vhci.c | 10 ++++++----
>  1 file changed, 6 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/bluetooth/hci_vhci.c b/drivers/bluetooth/hci_vhci.c
> index f3892e9ce800..572d68d52965 100644
> --- a/drivers/bluetooth/hci_vhci.c
> +++ b/drivers/bluetooth/hci_vhci.c
> @@ -11,6 +11,7 @@
>  #include <linux/module.h>
>  #include <asm/unaligned.h>
>
> +#include <linux/atomic.h>
>  #include <linux/kernel.h>
>  #include <linux/init.h>
>  #include <linux/slab.h>
> @@ -44,6 +45,7 @@ struct vhci_data {
>         bool wakeup;
>         __u16 msft_opcode;
>         bool aosp_capable;
> +       atomic_t initialized;
>  };
>
>  static int vhci_open_dev(struct hci_dev *hdev)
> @@ -75,11 +77,10 @@ static int vhci_send_frame(struct hci_dev *hdev, struct sk_buff *skb)
>
>         memcpy(skb_push(skb, 1), &hci_skb_pkt_type(skb), 1);
>
> -       mutex_lock(&data->open_mutex);
>         skb_queue_tail(&data->readq, skb);
> -       mutex_unlock(&data->open_mutex);
>
> -       wake_up_interruptible(&data->read_wait);
> +       if (atomic_read(&data->initialized))
> +               wake_up_interruptible(&data->read_wait);
>         return 0;
>  }
>
> @@ -464,7 +465,8 @@ static int __vhci_create_device(struct vhci_data *data, __u8 opcode)
>         skb_put_u8(skb, 0xff);
>         skb_put_u8(skb, opcode);
>         put_unaligned_le16(hdev->id, skb_put(skb, 2));
> -       skb_queue_tail(&data->readq, skb);
> +       skb_queue_head(&data->readq, skb);
> +       atomic_inc(&data->initialized);
>
>         wake_up_interruptible(&data->read_wait);
>         return 0;
> --
> 2.43.0.rc0.421.g78406f8d94-goog
>