Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752272AbdHPQf1 (ORCPT ); Wed, 16 Aug 2017 12:35:27 -0400 Received: from iolanthe.rowland.org ([192.131.102.54]:50666 "HELO iolanthe.rowland.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S1752026AbdHPQfZ (ORCPT ); Wed, 16 Aug 2017 12:35:25 -0400 Date: Wed, 16 Aug 2017 12:35:24 -0400 (EDT) From: Alan Stern X-X-Sender: stern@iolanthe.rowland.org To: Anton Vasilyev cc: Felipe Balbi , Greg Kroah-Hartman , Jussi Kivilinna , Peter Senna Tschudin , Raz Manor , Romain Perier , , , Subject: Re: [PATCH] udc: Memory leak on error path and use after free In-Reply-To: <682044ba-e4c2-a1b9-3652-a01e81a40ac2@ispras.ru> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 1626 Lines: 46 On Wed, 16 Aug 2017, Anton Vasilyev wrote: > On 16.08.2017 18:29, Alan Stern wrote: > > On Wed, 16 Aug 2017, Anton Vasilyev wrote: > > > >> gadget_release() is responsible for cleanup dev memory. > >> But if net2280_probe() fails after dev allocation, then > >> gadget_release() become unregistered and dev memory leaks. > > > > This isn't needed if usb_add_gadget_udc_release() is fixed, right? > > > > No, this situation could appear before call > usb_add_gadget_udc_release(). > > >> Also net2280_remove() calls usb_del_gadget_udc() which > >> perform schedule_delayed_work() with gadget_release(), so > >> it is possible that dev will be deallocated exactly after > >> this call and leads to use after free. > > > > Where is there a possible use after free? > > > > net2280_remove() continue work with struct net2280 *dev after call > usb_del_gadget_udc(&dev->gadget), but this net2280 *dev could be > deallocated by gadget_release() > > >> The patch moves deallocation from gadget_release() to > >> net2280_remove(). > > > > Alan Stern Okay, now I understand what you were saying. Yes, I agree, the existing code isn't right. But a better solution would be to move the usb_del_gadget_udc() call from the beginning of net2280_remove() to the end. And make the call conditional, depending on whether usb_add_gadget_udc_release() has already been called successfully. The point is that the device core does not allow drivers to deallocate memory containing a struct device before the ->release callback has been invoked. Your patch might do that, if the release was delayed for some reason. Alan Stern