svcrdma: Handle device removal outside of the CM event handler

Synchronously wait for all disconnects to complete to ensure the
transports have divested all hardware resources before the
underlying RDMA device can safely be removed.

Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>

3 files changed, 40 insertions, 1 deletions

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index d33bab33099a..619fc0bd837a 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -48,6 +48,7 @@
 #include <linux/sunrpc/rpc_rdma.h>
 #include <linux/sunrpc/rpc_rdma_cid.h>
 #include <linux/sunrpc/svc_rdma_pcl.h>
+#include <linux/sunrpc/rdma_rn.h>
 
 #include <linux/percpu_counter.h>
 #include <rdma/ib_verbs.h>
@@ -76,6 +77,7 @@ struct svcxprt_rdma {
 	struct svc_xprt      sc_xprt;		/* SVC transport structure */
 	struct rdma_cm_id    *sc_cm_id;		/* RDMA connection id */
 	struct list_head     sc_accept_q;	/* Conn. waiting accept */
+	struct rpcrdma_notification sc_rn;	/* removal notification */
 	int		     sc_ord;		/* RDMA read limit */
 	int                  sc_max_send_sges;
 	bool		     sc_snd_w_inv;	/* OK to use Send With Invalidate */
diff --git a/include/trace/events/rpcrdma.h b/include/trace/events/rpcrdma.h
index a96a985c49b3..e6a72646c507 100644
--- a/include/trace/events/rpcrdma.h
+++ b/include/trace/events/rpcrdma.h
@@ -2172,6 +2172,29 @@ TRACE_EVENT(svcrdma_qp_error,
 	)
 );
 
+TRACE_EVENT(svcrdma_device_removal,
+	TP_PROTO(
+		const struct rdma_cm_id *id
+	),
+
+	TP_ARGS(id),
+
+	TP_STRUCT__entry(
+		__string(name, id->device->name)
+		__array(unsigned char, addr, sizeof(struct sockaddr_in6))
+	),
+
+	TP_fast_assign(
+		__assign_str(name);
+		memcpy(__entry->addr, &id->route.addr.dst_addr,
+		       sizeof(struct sockaddr_in6));
+	),
+
+	TP_printk("device %s to be removed, disconnecting %pISpc\n",
+		__get_str(name), __entry->addr
+	)
+);
+
 DECLARE_EVENT_CLASS(svcrdma_sendqueue_class,
 	TP_PROTO(
 		const struct svcxprt_rdma *rdma,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index f15750cacacf..581cc5ed7c0c 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -339,7 +339,6 @@ static int svc_rdma_cma_handler(struct rdma_cm_id *cma_id,
 		svc_xprt_enqueue(xprt);
 		break;
 	case RDMA_CM_EVENT_DISCONNECTED:
-	case RDMA_CM_EVENT_DEVICE_REMOVAL:
 		svc_xprt_deferred_close(xprt);
 		break;
 	default:
@@ -384,6 +383,16 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 	return &cma_xprt->sc_xprt;
 }
 
+static void svc_rdma_xprt_done(struct rpcrdma_notification *rn)
+{
+	struct svcxprt_rdma *rdma = container_of(rn, struct svcxprt_rdma,
+						 sc_rn);
+	struct rdma_cm_id *id = rdma->sc_cm_id;
+
+	trace_svcrdma_device_removal(id);
+	svc_xprt_close(&rdma->sc_xprt);
+}
+
 /*
  * This is the xpo_recvfrom function for listening endpoints. Its
  * purpose is to accept incoming connections. The CMA callback handler
@@ -425,6 +434,9 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	dev = newxprt->sc_cm_id->device;
 	newxprt->sc_port_num = newxprt->sc_cm_id->port_num;
 
+	if (rpcrdma_rn_register(dev, &newxprt->sc_rn, svc_rdma_xprt_done))
+		goto errout;
+
 	newxprt->sc_max_req_size = svcrdma_max_req_size;
 	newxprt->sc_max_requests = svcrdma_max_requests;
 	newxprt->sc_max_bc_requests = svcrdma_max_bc_requests;
@@ -580,6 +592,7 @@ static void __svc_rdma_free(struct work_struct *work)
 {
 	struct svcxprt_rdma *rdma =
 		container_of(work, struct svcxprt_rdma, sc_work);
+	struct ib_device *device = rdma->sc_cm_id->device;
 
 	/* This blocks until the Completion Queues are empty */
 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
@@ -608,6 +621,7 @@ static void __svc_rdma_free(struct work_struct *work)
 	/* Destroy the CM ID */
 	rdma_destroy_id(rdma->sc_cm_id);
 
+	rpcrdma_rn_unregister(device, &rdma->sc_rn);
 	kfree(rdma);
 }