SUNRPC: Fix a backchannel race

We need to allow the server to send a new request immediately after we've
replied to the previous one. Right now, there is a window between the
send and the release of the old request in rpc_put_task(), where the
server could send us a new backchannel RPC call, and we have no
request to service it.

Signed-off-by: Trond Myklebust <trond.myklebust@primarydata.com>

2 files changed, 29 insertions, 14 deletions

diff --git a/net/sunrpc/backchannel_rqst.c b/net/sunrpc/backchannel_rqst.c
index 1baa410..9825ff0 100644
--- a/net/sunrpc/backchannel_rqst.c
+++ b/net/sunrpc/backchannel_rqst.c
@@ -37,16 +37,18 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 static inline int xprt_need_to_requeue(struct rpc_xprt *xprt)
 {
-	return xprt->bc_alloc_count > 0;
+	return xprt->bc_alloc_count < atomic_read(&xprt->bc_free_slots);
 }
 
 static inline void xprt_inc_alloc_count(struct rpc_xprt *xprt, unsigned int n)
 {
+	atomic_add(n, &xprt->bc_free_slots);
 	xprt->bc_alloc_count += n;
 }
 
 static inline int xprt_dec_alloc_count(struct rpc_xprt *xprt, unsigned int n)
 {
+	atomic_sub(n, &xprt->bc_free_slots);
 	return xprt->bc_alloc_count -= n;
 }
 
@@ -232,9 +234,15 @@ static struct rpc_rqst *xprt_alloc_bc_request(struct rpc_xprt *xprt, __be32 xid)
 	struct rpc_rqst *req = NULL;
 
 	dprintk("RPC:       allocate a backchannel request\n");
-	if (list_empty(&xprt->bc_pa_list))
+	if (atomic_read(&xprt->bc_free_slots) <= 0)
 		goto not_found;
-
+	if (list_empty(&xprt->bc_pa_list)) {
+		req = xprt_alloc_bc_req(xprt, GFP_ATOMIC);
+		if (!req)
+			goto not_found;
+		/* Note: this 'free' request adds it to xprt->bc_pa_list */
+		xprt_free_bc_request(req);
+	}
 	req = list_first_entry(&xprt->bc_pa_list, struct rpc_rqst,
 				rq_bc_pa_list);
 	req->rq_reply_bytes_recvd = 0;
@@ -260,11 +268,21 @@ void xprt_free_bc_request(struct rpc_rqst *req)
 
 	req->rq_connect_cookie = xprt->connect_cookie - 1;
 	smp_mb__before_atomic();
-	WARN_ON_ONCE(!test_bit(RPC_BC_PA_IN_USE, &req->rq_bc_pa_state));
 	clear_bit(RPC_BC_PA_IN_USE, &req->rq_bc_pa_state);
 	smp_mb__after_atomic();
 
-	if (!xprt_need_to_requeue(xprt)) {
+	/*
+	 * Return it to the list of preallocations so that it
+	 * may be reused by a new callback request.
+	 */
+	spin_lock_bh(&xprt->bc_pa_lock);
+	if (xprt_need_to_requeue(xprt)) {
+		list_add_tail(&req->rq_bc_pa_list, &xprt->bc_pa_list);
+		xprt->bc_alloc_count++;
+		req = NULL;
+	}
+	spin_unlock_bh(&xprt->bc_pa_lock);
+	if (req != NULL) {
 		/*
 		 * The last remaining session was destroyed while this
 		 * entry was in use.  Free the entry and don't attempt
@@ -275,14 +293,6 @@ void xprt_free_bc_request(struct rpc_rqst *req)
 		xprt_free_allocation(req);
 		return;
 	}
-
-	/*
-	 * Return it to the list of preallocations so that it
-	 * may be reused by a new callback request.
-	 */
-	spin_lock_bh(&xprt->bc_pa_lock);
-	list_add_tail(&req->rq_bc_pa_list, &xprt->bc_pa_list);
-	spin_unlock_bh(&xprt->bc_pa_lock);
 }
 
 /*
@@ -326,6 +336,7 @@ void xprt_complete_bc_request(struct rpc_rqst *req, uint32_t copied)
 
 	spin_lock(&xprt->bc_pa_lock);
 	list_del(&req->rq_bc_pa_list);
+	xprt->bc_alloc_count--;
 	spin_unlock(&xprt->bc_pa_lock);
 
 	req->rq_private_buf.len = copied;
diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index 51c8cad..b47f02f 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -1351,6 +1351,7 @@ bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
 	struct kvec	*argv = &rqstp->rq_arg.head[0];
 	struct kvec	*resv = &rqstp->rq_res.head[0];
 	struct rpc_task *task;
+	int proc_error;
 	int error;
 
 	dprintk("svc: %s(%p)\n", __func__, req);
@@ -1382,7 +1383,10 @@ bc_svc_process(struct svc_serv *serv, struct rpc_rqst *req,
 	svc_getnl(argv);	/* CALLDIR */
 
 	/* Parse and execute the bc call */
-	if (!svc_process_common(rqstp, argv, resv)) {
+	proc_error = svc_process_common(rqstp, argv, resv);
+
+	atomic_inc(&req->rq_xprt->bc_free_slots);
+	if (!proc_error) {
 		/* Processing error: drop the request */
 		xprt_free_bc_request(req);
 		return 0;