[Devel] Re: [PATCH v2] SUNRPC: check current nsproxy before set of node name on client creation

[Stanislav Kinsbursky]

10.09.2012 19:27, Myklebust, Trond пишет:
> On Mon, 2012-09-10 at 12:43 +0400, Stanislav Kinsbursky wrote:
>> 08.09.2012 18:33, Myklebust, Trond пишет:
>>> On Sat, 2012-09-08 at 08:59 +0300, Stanislav Kinsbursky wrote:
>>>> 08.09.2012 01:32, Myklebust, Trond пишет:
>>>>> On Mon, 2012-08-13 at 08:10 -0400, Jeff Layton wrote:
>>>>>> On Mon, 13 Aug 2012 15:37:31 +0400
>>>>>> Stanislav Kinsbursky <skinsbursky at parallels.com> wrote:
>>>>>>
>>>>>>> v2:
>>>>>>> 1) rpc_clnt_set_nodename() prototype updated.
>>>>>>> 2) fixed errors in comment.
>>>>>>>
>>>>>>> When child reaper exits, it can destroy mount namespace it belongs to, and if
>>>>>>> there are NFS mounts inside, then it will try to umount them. But in this
>>>>>>> point current->nsproxy is set to NULL and all namespaces will be destroyed one
>>>>>>> by one. I.e. we can't dereference current->nsproxy to obtain uts namespace.
>>>>>>>
>>>>>>> Signed-off-by: Stanislav Kinsbursky <skinsbursky at parallels.com>
>>>>>>> ---
>>>>>>>     net/sunrpc/clnt.c |   16 +++++++++++++---
>>>>>>>     1 files changed, 13 insertions(+), 3 deletions(-)
>>>>>>>
>>>>>>> diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
>>>>>>> index 9a9676e..8fbcbc8 100644
>>>>>>> --- a/net/sunrpc/clnt.c
>>>>>>> +++ b/net/sunrpc/clnt.c
>>>>>>> @@ -277,8 +277,18 @@ void rpc_clients_notifier_unregister(void)
>>>>>>>     	return rpc_pipefs_notifier_unregister(&rpc_clients_block);
>>>>>>>     }
>>>>>>>
>>>>>>> -static void rpc_clnt_set_nodename(struct rpc_clnt *clnt, const char *nodename)
>>>>>>> +static void rpc_clnt_set_nodename(struct rpc_clnt *clnt)
>>>>>>>     {
>>>>>>> +	const char *nodename;
>>>>>>> +
>>>>>>> +	/*
>>>>>>> +	 * We have to protect against dying child reaper, which has released
>>>>>>> +	 * its nsproxy already and is trying to destroy mount namespace.
>>>>>>> +	 */
>>>>>>> +	if (current->nsproxy == NULL)
>>>>>>> +		return;
>>>>>>> +
>>>>>>> +	nodename = utsname()->nodename;
>>>>>>>     	clnt->cl_nodelen = strlen(nodename);
>>>>>>>     	if (clnt->cl_nodelen > UNX_MAXNODENAME)
>>>>>>>     		clnt->cl_nodelen = UNX_MAXNODENAME;
>>>>>>> @@ -365,7 +375,7 @@ static struct rpc_clnt * rpc_new_client(const struct rpc_create_args *args, stru
>>>>>>>     	}
>>>>>>>
>>>>>>>     	/* save the nodename */
>>>>>>> -	rpc_clnt_set_nodename(clnt, utsname()->nodename);
>>>>>>> +	rpc_clnt_set_nodename(clnt);
>>>>>>>     	rpc_register_client(clnt);
>>>>>>>     	return clnt;
>>>>>>>
>>>>>>> @@ -524,7 +534,7 @@ rpc_clone_client(struct rpc_clnt *clnt)
>>>>>>>     	err = rpc_setup_pipedir(new, clnt->cl_program->pipe_dir_name);
>>>>>>>     	if (err != 0)
>>>>>>>     		goto out_no_path;
>>>>>>> -	rpc_clnt_set_nodename(new, utsname()->nodename);
>>>>>>> +	rpc_clnt_set_nodename(new);
>>>>>>>     	if (new->cl_auth)
>>>>>>>     		atomic_inc(&new->cl_auth->au_count);
>>>>>>>     	atomic_inc(&clnt->cl_count);
>>>>>>>
>>>>>>> --
>>>>>>> To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
>>>>>>> the body of a message to majordomo at vger.kernel.org
>>>>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>>> Acked-by: Jeff Layton <jlayton at redhat.com>
>>>>> OK, colour me confused (again).
>>>>
>>>> What color?
>>>>
>>>>> Why should a umount trigger an
>>>>> rpc_create() or rpc_clone_client()?
>>>>
>>>> It calls nsm_create().
>>>> Here is the trace (https://bugzilla.redhat.com/show_bug.cgi?id=830862,
>>>> comment 68):
>>>
>>> Right, but if we're using NFSv3 lock monitoring, we know in advance that
>>> we're going to need an nsm call to localhost. Why can't we just cache
>>> the one that we used to start lock monitoring in the first place?
>>>
>>
>> Do you suggest to cache the call or the client for the call?
>
> Hi Stanislav,
>
> Sorry, I agree that the above was unclear. My intention was to suggest
> that we should cache a reference to the rpc client that we used to
> connect to rpc.statd when initiating lock monitoring.
>
> Basically, I'm suggesting that we should do something similar to the
> rpcbind rpc_client caching scheme in net/sunrpc/rpcb_clnt.c.
>

Hi, Trond.
So, if I understand you right, we can create rpc client (or increase usage 
counter) on NSMPROC_MON call and destroy (or decrease usage counter) on 
NSMPROC_UNMON call.
Will this solution works?

> Cheers
>    Trond
>


-- 
Best regards,
Stanislav Kinsbursky