From: Thadeu Lima de Souza Cascardo cascardo@canonical.com
commit 8da3a0b87f4f1c3a3bbc4bfb78cf68476e97d183 upstream.
When cmtp_attach_device fails, cmtp_add_connection returns the error value which leads to the caller to doing fput through sockfd_put. But cmtp_session kthread, which is stopped in this path will also call fput, leading to a potential refcount underflow or a use-after-free.
Add a refcount before we signal the kthread to stop. The kthread will try to grab the cmtp_session_sem mutex before doing the fput, which is held when get_file is called, so there should be no races there.
Reported-by: Ryota Shiga Signed-off-by: Thadeu Lima de Souza Cascardo cascardo@canonical.com Signed-off-by: Marcel Holtmann marcel@holtmann.org Signed-off-by: Greg Kroah-Hartman gregkh@linuxfoundation.org --- net/bluetooth/cmtp/core.c | 5 +++++ 1 file changed, 5 insertions(+)
--- a/net/bluetooth/cmtp/core.c +++ b/net/bluetooth/cmtp/core.c @@ -391,6 +391,11 @@ int cmtp_add_connection(struct cmtp_conn if (!(session->flags & BIT(CMTP_LOOPBACK))) { err = cmtp_attach_device(session); if (err < 0) { + /* Caller will call fput in case of failure, and so + * will cmtp_session kthread. + */ + get_file(session->sock->file); + atomic_inc(&session->terminate); wake_up_interruptible(sk_sleep(session->sock->sk)); up_write(&cmtp_session_sem);