Re: [RFC PATCH 04/13] vfio/mdev: remove the usage of the list iterator after the loop

From: Jakob <jakobkoschel@gmail.com>
To: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	linux-arch <linux-arch@vger.kernel.org>,
	Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Arnd Bergman <arnd@arndb.de>,
	Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Kees Cook <keescook@chromium.org>,
	Mike Rapoport <rppt@kernel.org>,
	"Gustavo A. R. Silva" <gustavo@embeddedor.com>,
	Brian Johannesmeyer <bjohannesmeyer@gmail.com>,
	Cristiano Giuffrida <c.giuffrida@vu.nl>,
	"Bos, H.J." <h.j.bos@vu.nl>
Subject: Re: [RFC PATCH 04/13] vfio/mdev: remove the usage of the list iterator after the loop
Date: Wed, 23 Feb 2022 15:18:50 +0100	[thread overview]
Message-ID: <6BA40980-554F-45E2-914D-5E4CD02FF21C@gmail.com> (raw)
In-Reply-To: <20220218151216.GE1037534@ziepe.ca>

> On 18. Feb 2022, at 16:12, Jason Gunthorpe <jgg@ziepe.ca> wrote:
> 
> On Thu, Feb 17, 2022 at 07:48:20PM +0100, Jakob Koschel wrote:
>> It is unsafe to assume that tmp != mdev can only evaluate to false
>> if the break within the list iterator is hit.
>> 
>> When the break is not hit, tmp is set to an address derived from the
>> head element. If mdev would match with that value of tmp it would allow
>> continuing beyond the safety check even if mdev was never found within
>> the list
> 
> I think due to other construction this is not actually possible, but I
> guess it is technically correct
> 
> This seems like just a straight up style fix with nothing to do with
> speculative execution though. Why not just send it as a proper patch?
> 
> Jason

Thank you for your feedback.

I've raised some confusion here, I'm sorry about that.
The idea was to change list_for_each_entry() to set 'pos' to NULL
when the list terminates to avoid invalid usage in speculation.

This will break this code and I therefore included the suggested change
in this RFC. This RFC was not intended to be merged as is.

However, in this example, 'tmp' will be a out-of-bounds pointer
if the loop did finish without hitting the break, so the check past the
loop *could* match 'mdev' even though no break was ever met.

I've now realized that this is probably not realistic iff mdev always
points to a valid struct mdev_device.
(It's a slightly different scenario on PATCH 03/13).

Jakob