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 S932315Ab2AMQgt (ORCPT <rfc822;w@1wt.eu>);
	Fri, 13 Jan 2012 11:36:49 -0500
Received: from relay3.sgi.com ([192.48.152.1]:40834 "EHLO relay.sgi.com"
	rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP
	id S1751778Ab2AMQgq (ORCPT <rfc822;linux-kernel@vger.kernel.org>);
	Fri, 13 Jan 2012 11:36:46 -0500
Date: Fri, 13 Jan 2012 10:36:42 -0600
From: Dimitri Sivanich <sivanich@sgi.com>
To: Al Viro <viro@ZenIV.linux.org.uk>
Cc: linux-kernel@vger.kernel.org, "David S. Miller" <davem@davemloft.net>,
        Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>,
        James Morris <jmorris@namei.org>,
        Hideaki YOSHIFUJI <yoshfuji@linux-ipv6.org>,
        Patrick McHardy <kaber@trash.net>,
        "Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
        Paul Gortmaker <paul.gortmaker@windriver.com>,
        Andrew Morton <akpm@linux-foundation.org>,
        Jiri Kosina <jkosina@suse.cz>, Avi Kivity <avi@redhat.com>,
        linux-fsdevel@vger.kernel.org, netdev@vger.kernel.org
Subject: Re: [PATCH] Fix panic in __d_lookup with high dentry hashtable
 counts
Message-ID: <20120113163642.GB25788@sgi.com>
References: <20120113155237.GA25103@sgi.com>
 <20120113162236.GK23916@ZenIV.linux.org.uk>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20120113162236.GK23916@ZenIV.linux.org.uk>
User-Agent: Mutt/1.5.20 (2009-06-14)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Fri, Jan 13, 2012 at 04:22:36PM +0000, Al Viro wrote:
> On Fri, Jan 13, 2012 at 09:52:37AM -0600, Dimitri Sivanich wrote:
> > When the number of dentry cache hash table entries gets too high
> > (2147483648 entries), use of a signed integer in the initialization
> > loop prevents the dentry_hashtable from getting initialized, resulting
> > in a panic in __d_lookup.  Fixing this in dcache_init and a few other
> > spots for consistency.
> 
> >  static void __init dcache_init(void)
> >  {
> > -	int loop;
> > +	long loop;
> 
> You've got to be kidding.  Note that D_HASHMASK is at most 32bit.  Use
> of long here is an overkill and so's 2^31 hash buckets (that's what,
> 16Gb in hash list heads alone?  What kind of average chain length do
> you expect, BTW?)

Yes, long might be overkill right now, but the code is all __init time code.
I don't have numbers showing average chain length at this point, I was
simply fixing this one end case

> 
> Can alloc_large_system_hash() produce the horrors that large, anyway?

On a 16TB system, alloc_large_system_hash() produces 2^31 hash buckets, yes.

Would simply capping the value in alloc_large_system_hash() be more palatable?

Something like the following?

Index: linux/mm/page_alloc.c
===================================================================
--- linux.orig/mm/page_alloc.c
+++ linux/mm/page_alloc.c
@@ -5257,6 +5257,7 @@ void *__init alloc_large_system_hash(con
        if (max == 0) {
                max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
                do_div(max, bucketsize);
+               max = min(max, 1ULL << 30);
        }
 
        if (numentries > max)