[Linux-kernel-mentees] [PATCH] Bluetooth: fix "list_add double add" in hci_conn_complete_evt

[Marcel Holtmann]

Hi Coiby,

> When two HCI_EV_CONN_COMPLETE event packets with status=0 of the same
> HCI connection are received, device_add would be called twice which
> leads to kobject_add being called twice. Thus duplicate
> (struct hci_conn *conn)->dev.kobj.entry would be inserted into
> (struct hci_conn *conn)->dev.kobj.kset->list.
> 
> This issue can be fixed by checking (struct hci_conn *conn)->debugfs.
> If it's not NULL, it means the HCI connection has been completed and we
> won't duplicate the work as for processing the first
> HCI_EV_CONN_COMPLETE event.

do you have a btmon trace for this happening?

> Reported-and-tested-by: syzbot+dd768a260f7358adbaf9 at syzkaller.appspotmail.com
> Link: https://syzkaller.appspot.com/bug?extid=dd768a260f7358adbaf9
> Signed-off-by: Coiby Xu <coiby.xu at gmail.com>
> ---
> net/bluetooth/hci_event.c | 5 +++++
> 1 file changed, 5 insertions(+)
> 
> diff --git a/net/bluetooth/hci_event.c b/net/bluetooth/hci_event.c
> index 4b7fc430793c..1233739ce760 100644
> --- a/net/bluetooth/hci_event.c
> +++ b/net/bluetooth/hci_event.c
> @@ -2605,6 +2605,11 @@ static void hci_conn_complete_evt(struct hci_dev *hdev, struct sk_buff *skb)
> 	}
> 
> 	if (!ev->status) {
> +		if (conn->debugfs) {
> +			bt_dev_err(hdev, "The connection has been completed");
> +			goto unlock;
> +		}
> +

And instead of doing papering over a hole, I would rather detect that the HCI event is not valid since we already received one for this connection.

Regards

Marcel