Re: [Patch 4/5] Network Drop Monitor: Adding drop monitor implementation & Netlink protocol

From: David Miller <davem@davemloft.net>
To: herbert@gondor.apana.org.au
Cc: kaber@trash.net, nhorman@tuxdriver.com, zbr@ioremap.net,
	netdev@vger.kernel.org, kuznet@ms2.inr.ac.ru, pekkas@netcore.fi,
	jmorris@namei.org, yoshfuji@linux-ipv6.org
Subject: Re: [Patch 4/5] Network Drop Monitor: Adding drop monitor implementation & Netlink protocol
Date: Sun, 05 Apr 2009 02:56:21 -0700 (PDT)	[thread overview]
Message-ID: <20090405.025621.262216979.davem@davemloft.net> (raw)
In-Reply-To: <20090402144549.GA764@gondor.apana.org.au>

From: Herbert Xu <herbert@gondor.apana.org.au>
Date: Thu, 2 Apr 2009 22:45:49 +0800

> On Thu, Apr 02, 2009 at 04:42:18PM +0200, Patrick McHardy wrote:
> >
> > Actually we should be fine since the current netlink helpers only do
> > bytewise copying anyways. And I think we've pretty much gotten rid of
> > all the raw attribute accesses.
> 
> What about stuff like xfrm_userpolicy_info? I suppose if *everything*
> is copied then it wouldn't matter.

You were saying?

[ 6400.609361] Kernel unaligned access at TPC[10258734] copy_to_user_state+0x54/0x9c [xfrm_user]
[ 6400.609498] Kernel unaligned access at TPC[10258734] copy_to_user_state+0x54/0x9c [xfrm_user]
[ 6400.667575] Kernel unaligned access at TPC[10258734] copy_to_user_state+0x54/0x9c [xfrm_user]
[ 6400.667707] Kernel unaligned access at TPC[10258734] copy_to_user_state+0x54/0x9c [xfrm_user]

This problem has existed for a long time.  In this specific case even
if you use memcpy() (in fact memcpy() is being used here) GCC sees the
types and says it can do aligned 64-bit loads and stores to do the
memcpy() inline.