Re: [PATCH v3] mm: fix race between MADV_FREE reclaim and blkdev direct IO read

From: Christoph Hellwig <hch@infradead.org>
To: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Minchan Kim <minchan@kernel.org>,
	"Huang, Ying" <ying.huang@intel.com>, Yu Zhao <yuzhao@google.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Yang Shi <shy828301@gmail.com>, Miaohe Lin <linmiaohe@huawei.com>,
	linux-mm@kvack.org, linux-block@vger.kernel.org, axboe@kernel.dk
Subject: Re: [PATCH v3] mm: fix race between MADV_FREE reclaim and blkdev direct IO read
Date: Wed, 2 Feb 2022 06:03:16 -0800	[thread overview]
Message-ID: <YfqPJLTIx18DCGii@infradead.org> (raw)
In-Reply-To: <20220131230255.789059-1-mfo@canonical.com>

On Mon, Jan 31, 2022 at 08:02:55PM -0300, Mauricio Faria de Oliveira wrote:
> Well, blkdev_direct_IO() gets references for all pages, and on READ
> operations it only sets them dirty _later_.
> 
> So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for
> direct IO read from block devices, and page reclaim happens during
> __blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages()
> returns, but BEFORE the pages are set dirty, the situation happens.
> 
> The direct IO read eventually completes. Now, when userspace reads
> the buffers, the PTE is no longer there and the page fault handler
> do_anonymous_page() services that with the zero-page, NOT the data!

So why not just set the pages dirty early like the other direct I/O
implementations?  Or if this is fine with the patch should we remove
the early dirtying elsewhere?

> Reproducer:
> ==========
> 
> @ test.c (simplified, but works)

Can you add this to blktests or some other regularly run regression
test suite?

> +				smp_rmb();
> +
> +				/*
> +				 * The only page refs must be from the isolation
> +				 * plus one or more rmap's (dropped by discard:).

Overly long line.

> +				 */
> +				if ((ref_count == 1 + map_count) &&

No need for the inner braces.