Date:   Thu, 16 Mar 2023 20:05:38 +0100
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     Valery Zabrovsky <valthebrewer@yandex.ru>,
        Guo Zhengkui <guozhengkui@vivo.com>,
        Aaro Koskinen <aaro.koskinen@iki.fi>,
        Mian Yousaf Kaukab <yousaf.kaukab@intel.com>,
        Felipe Balbi <balbi@ti.com>,
        Alexey Khoroshilov <khoroshilov@ispras.ru>,
        linux-usb@vger.kernel.org, linux-kernel@vger.kernel.org,
        lvc-project@linuxtesting.org
Subject: Re: [PATCH] usb: gadget: net2280: fix NULL pointer dereference
Message-ID: <ZBNogj/To3j5OiQK@kroah.com>
References: <20230316155356.13391-1-valthebrewer@yandex.ru>
 <ZBNBlwc9lt+pLFkd@kroah.com>
 <2be27aee-1a2d-4c5c-bb62-018e1340d2ad@rowland.harvard.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <2be27aee-1a2d-4c5c-bb62-018e1340d2ad@rowland.harvard.edu>
Precedence: bulk

On Thu, Mar 16, 2023 at 02:22:35PM -0400, Alan Stern wrote:
> On Thu, Mar 16, 2023 at 05:19:35PM +0100, Greg Kroah-Hartman wrote:
> > On Thu, Mar 16, 2023 at 06:53:55PM +0300, Valery Zabrovsky wrote:
> > > In net2280_free_request():
> > > If _ep is NULL, then ep is NULL and is dereferenced
> > > while trying to produce an error message.
> > 
> > How can that ever happen?  How did you test and hit this?
> > 
> > > The patch replaces dev_err() with pr_err() which works fine.
> > 
> > That's not a good idea for driver code to use, sorry.
> > 
> > > Found by Linux Verification Center (linuxtesting.org) with SVACE.
> > 
> > As I have said before, your testing framework needs a lot more work.
> > 
> > good luck!
> 
> In situations like this, it might be better to remove the check 
> entirely.  If a driver does pass a NULL pointer, it will lead to an 
> invalid pointer dereference which will certainly cause an oops and might 
> very well crash the system.  That would be a lot more noticeable than an 
> error message hidden in a kernel log!
> 
> Greg, is there any general policy about the need for sanity checks such 
> as this one?  Like, don't put them in whenever a failure would lead 
> to an immediate fault which would be easy to track down?

That's the policy, no need to check if a caller is abusing the code and
will find out instantly.  Otherwise we end up checking on every
function, and that way lies madness.

thanks,

greg k-h