Re: [PATCH v2] nfsd: Fix races between nfsd4_cb_release() and nfsd4_shutdown_callback()

["J. Bruce Fields" <bfields@redhat.com>]

On Wed, Oct 23, 2019 at 05:43:18PM -0400, Trond Myklebust wrote:
> When we're destroying the client lease, and we call
> nfsd4_shutdown_callback(), we must ensure that we do not return
> before all outstanding callbacks have terminated and have
> released their payloads.

This is great, thanks!  We've seen what I'm fairly sure is the same bug
from Red Hat users.  I think my blind spot was an assumption that
rpc tasks wouldn't outlive rpc_shutdown_client().

However, it's causing xfstests runs to hang, and I haven't worked out
why yet.

I'll spend some time on it this afternoon and let you know what I figure
out.

--b.

> Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
> ---
> v2 - Fix a leak of clp->cl_cb_inflight when running null probes
> 
>  fs/nfsd/nfs4callback.c | 97 +++++++++++++++++++++++++++++++++---------
>  fs/nfsd/state.h        |  1 +
>  2 files changed, 79 insertions(+), 19 deletions(-)
> 
> diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
> index 524111420b48..a3c9517bcc64 100644
> --- a/fs/nfsd/nfs4callback.c
> +++ b/fs/nfsd/nfs4callback.c
> @@ -826,6 +826,31 @@ static int max_cb_time(struct net *net)
>  	return max(nn->nfsd4_lease/10, (time_t)1) * HZ;
>  }
>  
> +static struct workqueue_struct *callback_wq;
> +
> +static bool nfsd4_queue_cb(struct nfsd4_callback *cb)
> +{
> +	return queue_work(callback_wq, &cb->cb_work);
> +}
> +
> +static void nfsd41_cb_inflight_begin(struct nfs4_client *clp)
> +{
> +	atomic_inc(&clp->cl_cb_inflight);
> +}
> +
> +static void nfsd41_cb_inflight_end(struct nfs4_client *clp)
> +{
> +
> +	if (atomic_dec_and_test(&clp->cl_cb_inflight))
> +		wake_up_var(&clp->cl_cb_inflight);
> +}
> +
> +static void nfsd41_cb_inflight_wait_complete(struct nfs4_client *clp)
> +{
> +	wait_var_event(&clp->cl_cb_inflight,
> +			!atomic_read(&clp->cl_cb_inflight));
> +}
> +
>  static const struct cred *get_backchannel_cred(struct nfs4_client *clp, struct rpc_clnt *client, struct nfsd4_session *ses)
>  {
>  	if (clp->cl_minorversion == 0) {
> @@ -937,14 +962,21 @@ static void nfsd4_cb_probe_done(struct rpc_task *task, void *calldata)
>  		clp->cl_cb_state = NFSD4_CB_UP;
>  }
>  
> +static void nfsd4_cb_probe_release(void *calldata)
> +{
> +	struct nfs4_client *clp = container_of(calldata, struct nfs4_client, cl_cb_null);
> +
> +	nfsd41_cb_inflight_end(clp);
> +
> +}
> +
>  static const struct rpc_call_ops nfsd4_cb_probe_ops = {
>  	/* XXX: release method to ensure we set the cb channel down if
>  	 * necessary on early failure? */
>  	.rpc_call_done = nfsd4_cb_probe_done,
> +	.rpc_release = nfsd4_cb_probe_release,
>  };
>  
> -static struct workqueue_struct *callback_wq;
> -
>  /*
>   * Poke the callback thread to process any updates to the callback
>   * parameters, and send a null probe.
> @@ -975,9 +1007,12 @@ void nfsd4_change_callback(struct nfs4_client *clp, struct nfs4_cb_conn *conn)
>   * If the slot is available, then mark it busy.  Otherwise, set the
>   * thread for sleeping on the callback RPC wait queue.
>   */
> -static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
> +static bool nfsd41_cb_get_slot(struct nfsd4_callback *cb, struct rpc_task *task)
>  {
> -	if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
> +	struct nfs4_client *clp = cb->cb_clp;
> +
> +	if (!cb->cb_holds_slot &&
> +	    test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
>  		rpc_sleep_on(&clp->cl_cb_waitq, task, NULL);
>  		/* Race breaker */
>  		if (test_and_set_bit(0, &clp->cl_cb_slot_busy) != 0) {
> @@ -985,10 +1020,32 @@ static bool nfsd41_cb_get_slot(struct nfs4_client *clp, struct rpc_task *task)
>  			return false;
>  		}
>  		rpc_wake_up_queued_task(&clp->cl_cb_waitq, task);
> +		cb->cb_holds_slot = true;
>  	}
>  	return true;
>  }
>  
> +static void nfsd41_cb_release_slot(struct nfsd4_callback *cb)
> +{
> +	struct nfs4_client *clp = cb->cb_clp;
> +
> +	if (cb->cb_holds_slot) {
> +		cb->cb_holds_slot = false;
> +		clear_bit(0, &clp->cl_cb_slot_busy);
> +		rpc_wake_up_next(&clp->cl_cb_waitq);
> +	}
> +}
> +
> +static void nfsd41_destroy_cb(struct nfsd4_callback *cb)
> +{
> +	struct nfs4_client *clp = cb->cb_clp;
> +
> +	nfsd41_cb_release_slot(cb);
> +	if (cb->cb_ops && cb->cb_ops->release)
> +		cb->cb_ops->release(cb);
> +	nfsd41_cb_inflight_end(clp);
> +}
> +
>  /*
>   * TODO: cb_sequence should support referring call lists, cachethis, multiple
>   * slots, and mark callback channel down on communication errors.
> @@ -1005,11 +1062,8 @@ static void nfsd4_cb_prepare(struct rpc_task *task, void *calldata)
>  	 */
>  	cb->cb_seq_status = 1;
>  	cb->cb_status = 0;
> -	if (minorversion) {
> -		if (!cb->cb_holds_slot && !nfsd41_cb_get_slot(clp, task))
> -			return;
> -		cb->cb_holds_slot = true;
> -	}
> +	if (minorversion && !nfsd41_cb_get_slot(cb, task))
> +		return;
>  	rpc_call_start(task);
>  }
>  
> @@ -1076,9 +1130,7 @@ static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback
>  			cb->cb_seq_status);
>  	}
>  
> -	cb->cb_holds_slot = false;
> -	clear_bit(0, &clp->cl_cb_slot_busy);
> -	rpc_wake_up_next(&clp->cl_cb_waitq);
> +	nfsd41_cb_release_slot(cb);
>  	dprintk("%s: freed slot, new seqid=%d\n", __func__,
>  		clp->cl_cb_session->se_cb_seq_nr);
>  
> @@ -1091,8 +1143,10 @@ static bool nfsd4_cb_sequence_done(struct rpc_task *task, struct nfsd4_callback
>  		ret = false;
>  	goto out;
>  need_restart:
> -	task->tk_status = 0;
> -	cb->cb_need_restart = true;
> +	if (!test_bit(NFSD4_CLIENT_CB_KILL, &clp->cl_flags)) {
> +		task->tk_status = 0;
> +		cb->cb_need_restart = true;
> +	}
>  	return false;
>  }
>  
> @@ -1134,9 +1188,9 @@ static void nfsd4_cb_release(void *calldata)
>  	struct nfsd4_callback *cb = calldata;
>  
>  	if (cb->cb_need_restart)
> -		nfsd4_run_cb(cb);
> +		nfsd4_queue_cb(cb);
>  	else
> -		cb->cb_ops->release(cb);
> +		nfsd41_destroy_cb(cb);
>  
>  }
>  
> @@ -1170,6 +1224,7 @@ void nfsd4_shutdown_callback(struct nfs4_client *clp)
>  	 */
>  	nfsd4_run_cb(&clp->cl_cb_null);
>  	flush_workqueue(callback_wq);
> +	nfsd41_cb_inflight_wait_complete(clp);
>  }
>  
>  /* requires cl_lock: */
> @@ -1255,8 +1310,7 @@ nfsd4_run_cb_work(struct work_struct *work)
>  	clnt = clp->cl_cb_client;
>  	if (!clnt) {
>  		/* Callback channel broken, or client killed; give up: */
> -		if (cb->cb_ops && cb->cb_ops->release)
> -			cb->cb_ops->release(cb);
> +		nfsd41_destroy_cb(cb);
>  		return;
>  	}
>  
> @@ -1265,6 +1319,7 @@ nfsd4_run_cb_work(struct work_struct *work)
>  	 */
>  	if (!cb->cb_ops && clp->cl_minorversion) {
>  		clp->cl_cb_state = NFSD4_CB_UP;
> +		nfsd41_destroy_cb(cb);
>  		return;
>  	}
>  
> @@ -1290,5 +1345,9 @@ void nfsd4_init_cb(struct nfsd4_callback *cb, struct nfs4_client *clp,
>  
>  void nfsd4_run_cb(struct nfsd4_callback *cb)
>  {
> -	queue_work(callback_wq, &cb->cb_work);
> +	struct nfs4_client *clp = cb->cb_clp;
> +
> +	nfsd41_cb_inflight_begin(clp);
> +	if (!nfsd4_queue_cb(cb))
> +		nfsd41_cb_inflight_end(clp);
>  }
> diff --git a/fs/nfsd/state.h b/fs/nfsd/state.h
> index 46f56afb6cb8..d61b83b9654c 100644
> --- a/fs/nfsd/state.h
> +++ b/fs/nfsd/state.h
> @@ -367,6 +367,7 @@ struct nfs4_client {
>  	struct net		*net;
>  	struct list_head	async_copies;	/* list of async copies */
>  	spinlock_t		async_lock;	/* lock for async copies */
> +	atomic_t		cl_cb_inflight;	/* Outstanding callbacks */
>  };
>  
>  /* struct nfs4_client_reset
> -- 
> 2.21.0
>