Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH v3] net: xfrm: Add '_rcu' tag for rcu protected pointer in
 netns_xfrm
To:     Steffen Klassert <steffen.klassert@secunet.com>,
        Eric Dumazet <eric.dumazet@gmail.com>
CC:     <davem@davemloft.net>, <fw@strlen.de>, <netdev@vger.kernel.org>,
        <linux-kernel@vger.kernel.org>
References: <1551923648-17656-1-git-send-email-suyj.fnst@cn.fujitsu.com>
 <20190311101006.GK3758@gauss3.secunet.de>
 <b2beff03-f789-5c8b-bd69-57ece92372f8@gmail.com>
 <20190319151503.GW6035@gauss3.secunet.de>
From:   "Su Yanjun <suyj.fnst@cn.fujitsu.com>" <suyj.fnst@cn.fujitsu.com>
Message-ID: <91cf303a-7d44-2c15-8b7a-8b2eb797a234@cn.fujitsu.com>
Date:   Wed, 20 Mar 2019 08:53:54 +0800
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64; rv:60.0) Gecko/20100101
 Thunderbird/60.3.3
MIME-Version: 1.0
In-Reply-To: <20190319151503.GW6035@gauss3.secunet.de>
Content-Type: text/plain; charset="gbk"; format=flowed
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


On 2019/3/19 23:15, Steffen Klassert wrote:
> On Mon, Mar 18, 2019 at 10:22:46AM -0700, Eric Dumazet wrote:
>>
>> On 03/11/2019 03:10 AM, Steffen Klassert wrote:
>>> On Wed, Mar 06, 2019 at 08:54:08PM -0500, Su Yanjun wrote:
>>>> For rcu protected pointers, we'd better add '__rcu' for them.
>>>>
>>>> Once added '__rcu' tag for rcu protected pointer, the sparse tool reports
>>>> warnings.
>>>>
>>>> net/xfrm/xfrm_user.c:1198:39: sparse:    expected struct sock *sk
>>>> net/xfrm/xfrm_user.c:1198:39: sparse:    got struct sock [noderef] <asn:4> *nlsk
>>>> [...]
>>>>
>>>> So introduce a new wrapper function of nlmsg_unicast  to handle type
>>>> conversions.
>>>>
>>>> This patch also fixes a direct access of a rcu protected socket.
>>>>
>>>> Fixes: be33690d8fcf("[XFRM]: Fix aevent related crash")
>>>> Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>
>>> Patch applied, thanks!
>>>
>> Has this patch ever been tested ?
> I had this on your testing system and it did
> not complain. But maybe my testcases did not
> trigger that codepath. Su, can you answer
> on Eric question?
>
Firs of all, I didn't produce it on my test machine.

Maybe we need recompile the kernel with Eric Dumazet's advise.

CONFIG_LOCKDEP=y
CONFIG_PROVE_RCU=y

The second the code path indeed doesn't do as below:
rcu_read_lock()
rcu_dereference()
...
rcu_read_unlock()

All rcu_dereference are in the follow code path:
xfrm_user_rcv_msg
  link->doit(skb, nlh, attrs)
     rcu_dereference()

I think we can add rcu protection for nlsock

xfrm_user_rcv_msg
   rcu_read_lock()
   link->doit(skb, nlh, attrs)
   rcu_read_unlock()

Any advise?