Re: [PATCH] Bluetooth: L2CAP: Fix use-after-free in l2cap_unregister_user

[Pauli Virtanen]

ke, 2025-11-05 kello 19:52 +0530, ssrane_b23@xxxxxxxxxxxxx kirjoitti:
> From: Shaurya Rane <ssrane_b23@xxxxxxxxxxxxx>
> 
> Syzbot reported a use-after-free in l2cap_unregister_user(), caused by
> missing reference counting on the associated hci_dev. If the device is
> unregistered while L2CAP users are still active, l2cap_unregister_user()
> may access a freed hci_dev when taking its lock.
> 
> Fix this by taking a device reference in l2cap_register_user() using
> hci_dev_hold(), and releasing it in l2cap_unregister_user() via
> hci_dev_put(). This ensures the hci_dev remains valid for the lifetime
> of registered L2CAP users.
> 
> Reported-by: syzbot+14b6d57fb728e27ce23c@xxxxxxxxxxxxxxxxxxxxxxxxx
> Closes: https://syzkaller.appspot.com/bug?extid=14b6d57fb728e27ce23c
> Fixes: c8992cffbe74 ("Bluetooth: hci_event: Use of a function table to handle Command Complete")
> Signed-off-by: Shaurya Rane <ssrane_b23@xxxxxxxxxxxxx>
> ---
>  net/bluetooth/l2cap_core.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/net/bluetooth/l2cap_core.c b/net/bluetooth/l2cap_core.c
> index 805c752ac0a9..6a880f8ab6c2 100644
> --- a/net/bluetooth/l2cap_core.c
> +++ b/net/bluetooth/l2cap_core.c
> @@ -1688,6 +1688,11 @@ int l2cap_register_user(struct l2cap_conn *conn, struct l2cap_user *user)
>  	struct hci_dev *hdev = conn->hcon->hdev;
>  	int ret;
>  
> +	/* Hold a reference to hdev to prevent it from being freed while
> +	 * we have registered users.
> +	 */
> +	hci_dev_hold(hdev);
> +
>  	/* We need to check whether l2cap_conn is registered. If it is not, we
>  	 * must not register the l2cap_user. l2cap_conn_del() is unregisters
>  	 * l2cap_conn objects, but doesn't provide its own locking. Instead, it

The old comment here seems out of date since commit ab4eedb790cae,
currently l2cap_conn_del() appears to be using conn->lock to do

	mutex_lock(&conn->lock);
	...
	l2cap_unregister_all_users(conn);
	...
        hci_chan_del(conn->hchan);
        conn->hchan = NULL;
	...
	mutex_unlock(&conn->lock);

so it looks likely also taking conn->lock could avoid the races with
conn->users and conn->hchan.

> @@ -1717,6 +1722,10 @@ int l2cap_register_user(struct l2cap_conn *conn, struct l2cap_user *user)
>  
>  out_unlock:
>  	hci_dev_unlock(hdev);
> +
> +	if (ret)
> +		hci_dev_put(hdev);
> +
>  	return ret;
>  }
>  EXPORT_SYMBOL(l2cap_register_user);
> @@ -1735,6 +1744,9 @@ void l2cap_unregister_user(struct l2cap_conn *conn, struct l2cap_user *user)
>  
>  out_unlock:
>  	hci_dev_unlock(hdev);
> +
> +	/* Release the reference we took in l2cap_register_user */
> +	hci_dev_put(hdev);
>  }
>  EXPORT_SYMBOL(l2cap_unregister_user);
>  

-- 
Pauli Virtanen