From mboxrd@z Thu Jan 1 00:00:00 1970 From: Nick Piggin Subject: Re: [patch 01/28] fs: d_validate fixes Date: Fri, 19 Nov 2010 16:05:07 +1100 Message-ID: <20101119050507.GC3284@amd> References: <20101116140900.039761100@kernel.dk> <20101116142028.254946611@kernel.dk> <20101118.125123.241932424.davem@davemloft.net> <20101118.125913.13739447.davem@davemloft.net> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: npiggin@kernel.dk, linux-fsdevel@vger.kernel.org, linux-kernel@vger.kernel.org To: David Miller Return-path: Received: from ipmail07.adl2.internode.on.net ([150.101.137.131]:7330 "EHLO ipmail07.adl2.internode.on.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750730Ab0KSFFN (ORCPT ); Fri, 19 Nov 2010 00:05:13 -0500 Content-Disposition: inline In-Reply-To: <20101118.125913.13739447.davem@davemloft.net> Sender: linux-fsdevel-owner@vger.kernel.org List-ID: On Thu, Nov 18, 2010 at 12:59:13PM -0800, David Miller wrote: > From: David Miller > Date: Thu, 18 Nov 2010 12:51:23 -0800 (PST) > > > From: Nick Piggin > > Date: Wed, 17 Nov 2010 01:09:01 +1100 > > > >> d_validate has been broken for a long time. > >> > >> kmem_ptr_validate does not guarantee that a pointer can be dereferenced > >> if it can go away at any time. Even rcu_read_lock doesn't help, because > >> the pointer might be queued in RCU callbacks but not executed yet. > >> > >> So the parent cannot be checked, nor the name hashed. The dentry pointer > >> can not be touched until it can be verified under lock. Hashing simply > >> cannot be used. > >> > >> Instead, verify the parent/child relationship by traversing parent's > >> d_child list. It's slow, but only ncpfs and the destaged smbfs care > >> about it, at this point. > >> > >> Signed-off-by: Nick Piggin > > > > This won't apply because is conflicts with Christoph Hellwig's > > RCU conversion of d_validate(). > > > > Which is a change that went in more than a month ago. > > In fact the conflicts of your patch set are even more pervasive, since > all dcache hash traversals are essentially RCU protected instead of > dcache_lock protected right now. Not sure what you mean there. The patches are against upstream+revert of the last d_validate patch. dcache_lock splitup of this series is to split the lock out of all the other paths, and importantly allow d_lock to protect the complete dcache state of the dentry. Next 2 steps (that depend on this series but not on each other) are fine grained locking of the split locks, and rcu-walk. rcu-walk is what I called store-free path walking, because we extend RCU not only to the hash lookup but the entire path walk. I'll get all that out when I get a bit of time to work on it again. Thanks, Nick