[Devel] [PATCH 2/6] nfsd: swap fs root in NFSd kthreads

[Stanislav Kinsbursky]

Thanks!

11.01.2013 21:20, J. Bruce Fields пишет:
> On Fri, Jan 11, 2013 at 12:03:12PM -0500, J. Bruce Fields wrote:
>> On Fri, Jan 11, 2013 at 06:56:58PM +0400, Stanislav Kinsbursky wrote:
>>> 11.12.2012 19:35, J. Bruce Fields пишет:
>>>> On Tue, Dec 11, 2012 at 10:20:36AM -0500, J. Bruce Fields wrote:
>>>>> On Tue, Dec 11, 2012 at 07:07:00PM +0400, Stanislav Kinsbursky wrote:
>>>>>> I don't really understand, how  mountd's root can be wrong. I.e.
>>>>>> its' always right as I see it. NFSd kthreads have to swap/use
>>>>>> relative path/whatever to communicate with proper mountd.
>>>>>> Or I'm missing something?
>>>>>
>>>>> Ugh, I see the problem: I thought svc_export_request was called at the
>>>>> time mountd does the read, but instead its done at the time nfsd does
>>>>> the upcall.
>>>>>
>>>>> I suspect that's wrong, and we really want this done in the context of
>>>>> the mountd process when it does the read call.  If d_path is called
>>>>> there then we have no problem.
>>>>
>>>> Right, so I'd be happier if we could modify sunrpc_cache_pipe_upcall to
>>>> skip calling cache_request and instead delay that until cache_read().  I
>>>> think that should be possible.
>>>>
>>>
>>> So, Bruce, what we going to do (or what you want me to do) with the rest of NFSd changes?
>>> I.e. how I should solve this d_path() problem?
>>> I.e. I don't understand what did you mean by "I'd be happier if we could modify sunrpc_cache_pipe_upcall to
>>> skip calling cache_request and instead delay that until cache_read()".
>>> Could you give me a hint?
>>
>> Definitely.  So normally the way these upcalls happen are:
>>
>> 	1. the kernel does a cache lookup, finds no matching item, and
>> 	   calls sunrpc_cache_pipe_upcall().
>> 	2. sunrpc_cache_pipe_upcall() formats the upcall: it allocates a
>> 	   struct cache_request crq and fills crq->buf with the upcall
>> 	   data by calling the cache's ->cache_request() method.
>> 	3. Then rpc.mountd realizes there's data available in
>> 	   /proc/net/rpc/nfsd.fh/content, so it does a read on that file.
>> 	4. cache_read copies the formatted upcall from crq->buf to
>> 	   to userspace.
>>
>> So all I'm suggesting is that instead of calling ->cache_request() at
>> step 2, we do it at step 4.
>>
>> Then cache_request will be called from rpc.mountd's read.  So we'll know
>> which container rpc.mountd is in.
>>
>> Does that make sense?
>
> The following is untested, ugly, and almost certainly insufficient and
> wrong, but maybe it's a starting point:
>
> --b.
>
> diff --git a/net/sunrpc/cache.c b/net/sunrpc/cache.c
> index 9f84703..f15e4c1 100644
> --- a/net/sunrpc/cache.c
> +++ b/net/sunrpc/cache.c
> @@ -744,6 +744,7 @@ struct cache_request {
>   	char			* buf;
>   	int			len;
>   	int			readers;
> +	void (*cache_request)(struct cache_detail *, struct cache_head *, char **, int *);
>   };
>   struct cache_reader {
>   	struct cache_queue	q;
> @@ -785,10 +786,19 @@ static ssize_t cache_read(struct file *filp, char __user *buf, size_t count,
>   	spin_unlock(&queue_lock);
>
>   	if (rp->offset == 0 && !test_bit(CACHE_PENDING, &rq->item->flags)) {
> +		char *bp;
> +		int len = PAGE_SIZE;
> +
>   		err = -EAGAIN;
>   		spin_lock(&queue_lock);
>   		list_move(&rp->q.list, &rq->q.list);
>   		spin_unlock(&queue_lock);
> +
> +		bp = rq->buf;
> +		rq->cache_request(cd, rq->item, &bp, &len);
> +		if (rq->len < 0)
> +			goto out;
> +		rq->len = PAGE_SIZE - len;
>   	} else {
>   		if (rp->offset + count > rq->len)
>   			count = rq->len - rp->offset;
> @@ -1149,8 +1159,6 @@ int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
>
>   	char *buf;
>   	struct cache_request *crq;
> -	char *bp;
> -	int len;
>
>   	if (!cache_listeners_exist(detail)) {
>   		warn_no_listener(detail);
> @@ -1167,19 +1175,10 @@ int sunrpc_cache_pipe_upcall(struct cache_detail *detail, struct cache_head *h,
>   		return -EAGAIN;
>   	}
>
> -	bp = buf; len = PAGE_SIZE;
> -
> -	cache_request(detail, h, &bp, &len);
> -
> -	if (len < 0) {
> -		kfree(buf);
> -		kfree(crq);
> -		return -EAGAIN;
> -	}
> +	crq->cache_request = cache_request;
>   	crq->q.reader = 0;
>   	crq->item = cache_get(h);
>   	crq->buf = buf;
> -	crq->len = PAGE_SIZE - len;
>   	crq->readers = 0;
>   	spin_lock(&queue_lock);
>   	list_add_tail(&crq->q.list, &detail->queue);
>


-- 
Best regards,
Stanislav Kinsbursky