Date:   Tue, 7 Feb 2023 21:58:03 +0000
From:   Xin Zhao <xnzhao@google.com>
To:     John Keeping <john@metanate.com>
Cc:     Alan Stern <stern@rowland.harvard.edu>, gregkh@linuxfoundation.org,
        jakobkoschel@gmail.com, rdunlap@infradead.org, ira.weiny@intel.com,
        linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] usb: gadget: dummyhcd: Fix use-after-free in
 dummy_free_request
Message-ID: <Y+LJa8ZuBaD1DwlL@google.com>
References: <20230206225258.2302954-1-xnzhao@google.com>
 <Y+GcoFKiAkrCoAsv@rowland.harvard.edu>
 <Y+I9HcPvrm1TzUCw@donbot>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y+I9HcPvrm1TzUCw@donbot>
Precedence: bulk

On Tue, Feb 07, 2023 at 11:59:25AM +0000, John Keeping wrote:
> On Mon, Feb 06, 2023 at 07:34:40PM -0500, Alan Stern wrote:
> > On Mon, Feb 06, 2023 at 10:52:58PM +0000, Xin Zhao wrote:
> > > DummyHCD assume when dummy_free_request is called, the request
> > > is already detached from request queues. It is correct in most
> > > cases.
> > > But when DummyHCD is detached from gadget configfs with pending
> > > requests and some requests are still in pending queue,
> > > dummy_free_request would free them directly.
> > > Later on, dummy_udc_stop would iterate pending queue to release
> > > the requests again.
> > > 
> > > Stacktrace for dummy_free_reqeust
> > > ```
> > > kfree(const void * x) (slub.c:4200)
> > > dummy_free_request(struct usb_ep * _ep, struct usb_request * _req) (dummy_hcd.c:691)
> > > usb_ep_free_request(struct usb_ep * ep, struct usb_request * req) (core.c:201)
> > > functionfs_unbind(struct ffs_data * ffs) (f_fs.c:1894)
> > 
> > That's the bug right there.  The kerneldoc for usb_ep_free_request() 
> > says "Caller guarantees the request is not queued".  So it looks like 
> > the real solution is to fix functionfs_unbind().
> 
> This is commit ce405d561b02 ("usb: gadget: f_fs: Ensure ep0req is
> dequeued before free_request") IIUC.
> 
> Xin, are you able to test a version with that commit?

Yes John, the commit may fix the issue. Thanks for bring it up.