From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751291AbdJDJYq (ORCPT ); Wed, 4 Oct 2017 05:24:46 -0400 Received: from mail-lf0-f43.google.com ([209.85.215.43]:49893 "EHLO mail-lf0-f43.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751161AbdJDJYo (ORCPT ); Wed, 4 Oct 2017 05:24:44 -0400 X-Google-Smtp-Source: AOwi7QAJROeg1mfgkdDmfouJl4Pq88Cruko8ht7UY5ZEZghQ4Oq8N0Cumb9lDKxwATZFxqoRsN55zg== Date: Wed, 4 Oct 2017 11:24:42 +0200 From: Johan Hovold To: Takashi Iwai Cc: Greg Kroah-Hartman , Alan Stern , Andrey Konovalov , alsa-devel@alsa-project.org, Arvind Yadav , Jaroslav Kysela , Takashi Sakamoto , LKML , Dmitry Vyukov , Kostya Serebryany , syzkaller , linux-usb@vger.kernel.org Subject: Re: usb/sound/bcd2000: warning in bcd2000_init_device Message-ID: <20171004092442.GF3404@localhost> References: <20171003174221.GA13006@kroah.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: User-Agent: Mutt/1.7.2 (2016-11-26) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, Oct 04, 2017 at 08:10:59AM +0200, Takashi Iwai wrote: > On Tue, 03 Oct 2017 19:42:21 +0200, > Greg Kroah-Hartman wrote: > > > > On Tue, Oct 03, 2017 at 12:50:08PM -0400, Alan Stern wrote: > > > On Tue, 3 Oct 2017, Takashi Iwai wrote: > > > > > > > > It's a dev_WARN because it indicates a potentially serious error in the > > > > > driver: The driver has submitted an interrupt URB to a bulk endpoint. > > > > > That may not sound bad, but the same check gets triggered if a driver > > > > > submits a bulk URB to an isochronous endpoint, or any other invalid > > > > > combination. > > > > > > > > > > Most likely the explanation here is that the driver doesn't bother to > > > > > check the endpoint type because it expects the endpoint will always be > > > > > interrupt. But that is not a safe strategy. USB devices and their > > > > > firmware should not be trusted unnecessarily. > > > > > > > > > > The best fix is, like you said, to add a sanity check in the caller. > > > > > > > > OK, but then do we have some handy helper for the check? > > > > As other bug reports by syzkaller suggest, there are a few other > > > > drivers that do the same, submitting a urb with naive assumption of > > > > the fixed EP for specific devices. In the end we'll need to put the > > > > very same checks there in multiple places. > > > > > > Perhaps we could add a helper routine that would take a list of > > > expected endpoint types and check that the actual endpoints match the > > > types. But of course, all the drivers you're talking about would have > > > to add a call to this helper routine. > > > > We have almost this type of function, usb_find_common_endpoints(), > > what's wrong with using that? Johan has already swept the tree and > > added a lot of these checks, odds are no one looked at the sound/ > > subdir... Yeah, I only swept the tree for instances were a missing endpoint could lead to a NULL-deref. This is not the case here were the endpoint addresses are hardcoded in the driver. I also never got around to applying the new helper outside of drivers/usb. > Well, what I had in my mind is just a snippet from usb_submit_urb(), > something like: > > bool usb_sanity_check_urb_pipe(struct urb *urb) > { > struct usb_host_endpoint *ep; > int xfertype; > static const int pipetypes[4] = { > PIPE_CONTROL, PIPE_ISOCHRONOUS, PIPE_BULK, PIPE_INTERRUPT > }; > > ep = usb_pipe_endpoint(urb->dev, urb->pipe); > xfertype = usb_endpoint_type(&ep->desc); > return usb_pipetype(urb->pipe) != pipetypes[xfertype]; > } > > And calling this before usb_submit_urb() in each place that assigns > the fixed EP as device-specific quirks. > Does it make sense? Not really. Your driver should not even bind to an interface which lacks the expected endpoints (rather than check this at a potentially later point in time when URBs are submitted). The new helper which Greg mentioned would allow this to implemented with just a few lines of code. Just add it to bcd2000_init_midi() or similar. Thanks, Johan