Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 1 Jul 2019 16:32:11 -0700
From:   Andrew Morton <akpm@linux-foundation.org>
To:     Henry Burns <henryburns@google.com>
Cc:     Vitaly Wool <vitalywool@gmail.com>,
        Vitaly Vul <vitaly.vul@sony.com>,
        Mike Rapoport <rppt@linux.vnet.ibm.com>,
        Xidong Wang <wangxidong_97@163.com>,
        Shakeel Butt <shakeelb@google.com>,
        Jonathan Adams <jwadams@google.com>, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mm/z3fold.c: Lock z3fold page before 
 __SetPageMovable()
Message-Id: <20190701163211.e9e0f2cf5332c06640e3019d@linux-foundation.org>
In-Reply-To: <20190701212303.168581-1-henryburns@google.com>
References: <20190701212303.168581-1-henryburns@google.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon,  1 Jul 2019 14:23:03 -0700 Henry Burns <henryburns@google.com> wrote:

> __SetPageMovable() expects it's page to be locked, but z3fold.c doesn't
> lock the page.

So this triggers the VM_BUG_ON_PAGE(!PageLocked(page), page) in
__SetPageMovable(), yes?

> Following zsmalloc.c's example we call trylock_page() and
> unlock_page(). Also makes z3fold_page_migrate() assert that newpage is
> passed in locked, as documentation.
> 
> ...
>
> --- a/mm/z3fold.c
> +++ b/mm/z3fold.c
> @@ -918,7 +918,9 @@ static int z3fold_alloc(struct z3fold_pool *pool, size_t size, gfp_t gfp,
>  		set_bit(PAGE_HEADLESS, &page->private);
>  		goto headless;
>  	}
> +	WARN_ON(!trylock_page(page));

If this warn triggers then someone else has locked the page.

>	__SetPageMovable(page, pool->inode->i_mapping);
> + 	unlock_page(page);

and we proceed to undo their lock.  So that other code path will then
perform an unlock of an unlocked page.  Etcetera.

It would be much much better to do a plain old lock_page() here.  If
that results in a deadlock then let's find out why and fix it without
trylock hacks.