From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1752086AbcFFNH1 (ORCPT <rfc822;w@1wt.eu>);
	Mon, 6 Jun 2016 09:07:27 -0400
Received: from merlin.infradead.org ([205.233.59.134]:54556 "EHLO
	merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1751684AbcFFNHZ (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Mon, 6 Jun 2016 09:07:25 -0400
Date: Mon, 6 Jun 2016 15:07:19 +0200
From: Peter Zijlstra <peterz@infradead.org>
To: Mike Galbraith <umgwanakikbuti@gmail.com>
Cc: Mel Gorman <mgorman@suse.de>, lkml <linux-kernel@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        kirill@shutemov.name
Subject: Re: [patch] futex: Fix v4.6+ ltp futex_wait04 regression
Message-ID: <20160606130719.GA30935@twins.programming.kicks-ass.net>
References: <1465206041.6397.14.camel@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <1465206041.6397.14.camel@gmail.com>
User-Agent: Mutt/1.5.23.1 (2014-03-12)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Mon, Jun 06, 2016 at 11:40:41AM +0200, Mike Galbraith wrote:
> --- a/kernel/futex.c
> +++ b/kernel/futex.c
> @@ -469,7 +469,7 @@ get_futex_key(u32 __user *uaddr, int fsh
>  {
>  	unsigned long address = (unsigned long)uaddr;
>  	struct mm_struct *mm = current->mm;
> -	struct page *page;
> +	struct page *page, *pinned = NULL;
>  	struct address_space *mapping;
>  	int err, ro = 0;
>  
> @@ -530,8 +530,18 @@ get_futex_key(u32 __user *uaddr, int fsh
>  	 * considered here and page lock forces unnecessarily serialization
>  	 * From this point on, mapping will be re-verified if necessary and
>  	 * page lock will be acquired only if it is unavoidable
> -	 */
> +	 *
> +	 * If we're dealing with a compound page, save our reference to the
> +	 * page we locked in memory above, and take a new reference on the
> +	 * page head, dropping the previously pinned page reference on retry.
> +	 */
> +	if (unlikely(pinned && page != pinned))
> +		put_page(pinned);
> +	pinned = page;
>  	page = compound_head(page);
> +	if (unlikely(pinned != page))
> +		get_page(page);

Not needed, since {get,put}_page() explicitly use compound_head() to
track the reference count on.

> +
>  	mapping = READ_ONCE(page->mapping);
>  
>  	/*
> @@ -560,12 +570,14 @@ get_futex_key(u32 __user *uaddr, int fsh
>  		lock_page(page);
>  		shmem_swizzled = PageSwapCache(page) || page->mapping;
>  		unlock_page(page);
> -		put_page(page);
>  
> -		if (shmem_swizzled)
> +		if (shmem_swizzled) {
> +			put_page(page);
>  			goto again;
> +		}
>  
> -		return -EFAULT;
> +		err = -EFAULT;
> +		goto out;

This also simplifies away

>  	}
>  
>  	/*
> @@ -654,12 +666,14 @@ get_futex_key(u32 __user *uaddr, int fsh
>  
>  		key->both.offset |= FUT_OFF_INODE; /* inode-based key */
>  		key->shared.inode = inode;
> -		key->shared.pgoff = basepage_index(page);
> +		key->shared.pgoff = basepage_index(pinned);

But this seems to be the actual fix; because while I think the compound
page cannot change from under us, it can change between futex
invocations, and then having a key change is _bad_.

But someone other than me would have to concurr, its been too long since
I looked at all this compound muck.

>  		rcu_read_unlock();
>  	}
>  
>  out:
>  	put_page(page);
> +	if (unlikely(pinned != page))
> +		put_page(pinned);

and that too can go.

>  	return err;
>  }