From: "J. Bruce Fields" <bfields@fieldses.org>
Subject: Re: [PATCH 1/4] SUNRPC: spin svc_rqst initialization to its own
	function
Date: Fri, 18 Jan 2008 16:51:28 -0500
Message-ID: <20080118215128.GN15158@fieldses.org>
References: <1200319518-22422-1-git-send-email-jlayton@redhat.com> <1200319518-22422-2-git-send-email-jlayton@redhat.com> <20080118205943.GL15158@fieldses.org> <20080118164844.7f91d6ff@tleilax.poochiereds.net>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: neilb@suse.de, linux-nfs@vger.kernel.org,
	linux-kernel@vger.kernel.org
To: Jeff Layton <jlayton@redhat.com>
In-Reply-To: <20080118164844.7f91d6ff-RtJpwOs3+0O+kQycOl6kW4xkIHaj4LzF@public.gmane.org>
Sender: linux-nfs-owner@vger.kernel.org

On Fri, Jan 18, 2008 at 04:48:44PM -0500, Jeff Layton wrote:
> On Fri, 18 Jan 2008 15:59:43 -0500
> "J. Bruce Fields" <bfields@fieldses.org> wrote:
> 
> > On Mon, Jan 14, 2008 at 09:05:15AM -0500, Jeff Layton wrote:
> > > Move the initialzation in __svc_create_thread that happens prior to
> > > thread creation to a new function. Export the function to allow
> > > services to have better control over the svc_rqst structs.
> > > 
> > > Also rearrange the rqstp initialization to prevent NULL pointer
> > > dereferences in svc_exit_thread in case allocations fail.
> > 
> > Those NULL dereferences are from the
> > 
> > 	list_del(&rqstp->rq_all);
> > 
> > ?  OK, make sense.  Thanks!
> > 
> > --b.
> > 
> 
> Sorry, I didn't explain that well...
> 
> This was the problem that Neil pointed out with the existing code. If
> the rqstp kzalloc succeeds, but the later kmallocs in
> __svc_create_thread fail, we goto here:
> 
>     out_thread:
>             svc_exit_thread(rqstp);
> 
> svc_exit_thread does this:
> 
>     struct svc_pool *pool = rqstp->rq_pool;
> 
> ...and then later:
> 
>    spin_lock_bh(&pool->sp_lock);
> 
> ...but rq_pool is set after the kmallocs, so if they fail rq_pool will
> be NULL, and we'll oops in that spin_lock_bh().
> 
> The fix is to move the kmallocs closer to the bottom in the new
> svc_prepare_thread function.

OK, got it, thanks.

--b.