@@ -109,7 +109,7 @@
__assign_str(port, rpcrdma_portstr(r_xprt));
),
- TP_printk("peer=[%s]:%s r_xprt=%p: rc=%d connect status=%d",
+ TP_printk("peer=[%s]:%s r_xprt=%p: rc=%d connection status=%d",
__get_str(addr), __get_str(port), __entry->r_xprt,
__entry->rc, __entry->connect_status
)
@@ -409,6 +409,7 @@
DEFINE_CONN_EVENT(connect);
DEFINE_CONN_EVENT(disconnect);
+DEFINE_CONN_EVENT(flush_dct);
DEFINE_RXPRT_EVENT(xprtrdma_create);
DEFINE_RXPRT_EVENT(xprtrdma_op_destroy);
@@ -358,8 +358,8 @@ struct rpcrdma_mr_seg *frwr_map(struct rpcrdma_xprt *r_xprt,
/**
* frwr_wc_fastreg - Invoked by RDMA provider for a flushed FastReg WC
- * @cq: completion queue (ignored)
- * @wc: completed WR
+ * @cq: completion queue
+ * @wc: WCE for a completed FastReg WR
*
*/
static void frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc)
@@ -371,6 +371,8 @@ static void frwr_wc_fastreg(struct ib_cq *cq, struct ib_wc *wc)
/* WARNING: Only wr_cqe and status are reliable at this point */
trace_xprtrdma_wc_fastreg(wc, frwr);
/* The MR will get recycled when the associated req is retransmitted */
+
+ rpcrdma_flush_disconnect(cq, wc);
}
/**
@@ -441,8 +443,8 @@ static void __frwr_release_mr(struct ib_wc *wc, struct rpcrdma_mr *mr)
/**
* frwr_wc_localinv - Invoked by RDMA provider for a LOCAL_INV WC
- * @cq: completion queue (ignored)
- * @wc: completed WR
+ * @cq: completion queue
+ * @wc: WCE for a completed LocalInv WR
*
*/
static void frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc)
@@ -455,12 +457,14 @@ static void frwr_wc_localinv(struct ib_cq *cq, struct ib_wc *wc)
/* WARNING: Only wr_cqe and status are reliable at this point */
trace_xprtrdma_wc_li(wc, frwr);
__frwr_release_mr(wc, mr);
+
+ rpcrdma_flush_disconnect(cq, wc);
}
/**
* frwr_wc_localinv_wake - Invoked by RDMA provider for a LOCAL_INV WC
- * @cq: completion queue (ignored)
- * @wc: completed WR
+ * @cq: completion queue
+ * @wc: WCE for a completed LocalInv WR
*
* Awaken anyone waiting for an MR to finish being fenced.
*/
@@ -475,6 +479,8 @@ static void frwr_wc_localinv_wake(struct ib_cq *cq, struct ib_wc *wc)
trace_xprtrdma_wc_li_wake(wc, frwr);
__frwr_release_mr(wc, mr);
complete(&frwr->fr_linv_done);
+
+ rpcrdma_flush_disconnect(cq, wc);
}
/**
@@ -562,8 +568,8 @@ void frwr_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
/**
* frwr_wc_localinv_done - Invoked by RDMA provider for a signaled LOCAL_INV WC
- * @cq: completion queue (ignored)
- * @wc: completed WR
+ * @cq: completion queue
+ * @wc: WCE for a completed LocalInv WR
*
*/
static void frwr_wc_localinv_done(struct ib_cq *cq, struct ib_wc *wc)
@@ -581,6 +587,8 @@ static void frwr_wc_localinv_done(struct ib_cq *cq, struct ib_wc *wc)
/* Ensure @rep is generated before __frwr_release_mr */
smp_rmb();
rpcrdma_complete_rqst(rep);
+
+ rpcrdma_flush_disconnect(cq, wc);
}
/**
@@ -129,13 +129,31 @@ static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt)
}
/**
+ * rpcrdma_flush_disconnect - Disconnect on flushed completion
+ * @cq: completion queue
+ * @wc: work completion entry
+ *
+ * Must be called in process context.
+ */
+void rpcrdma_flush_disconnect(struct ib_cq *cq, struct ib_wc *wc)
+{
+ struct rpcrdma_xprt *r_xprt = cq->cq_context;
+ struct rpc_xprt *xprt = &r_xprt->rx_xprt;
+
+ if (wc->status != IB_WC_SUCCESS && r_xprt->rx_ep.rep_connected == 1) {
+ r_xprt->rx_ep.rep_connected = -ECONNABORTED;
+ trace_xprtrdma_flush_dct(r_xprt, wc->status);
+ xprt_force_disconnect(xprt);
+ }
+}
+
+/**
* rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC
* @cq: completion queue
- * @wc: completed WR
+ * @wc: WCE for a completed Send WR
*
*/
-static void
-rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
+static void rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
{
struct ib_cqe *cqe = wc->wr_cqe;
struct rpcrdma_sendctx *sc =
@@ -144,21 +162,21 @@ static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt)
/* WARNING: Only wr_cqe and status are reliable at this point */
trace_xprtrdma_wc_send(sc, wc);
rpcrdma_sendctx_put_locked((struct rpcrdma_xprt *)cq->cq_context, sc);
+ rpcrdma_flush_disconnect(cq, wc);
}
/**
* rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
- * @cq: completion queue (ignored)
- * @wc: completed WR
+ * @cq: completion queue
+ * @wc: WCE for a completed Receive WR
*
*/
-static void
-rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
+static void rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
{
struct ib_cqe *cqe = wc->wr_cqe;
struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep,
rr_cqe);
- struct rpcrdma_xprt *r_xprt = rep->rr_rxprt;
+ struct rpcrdma_xprt *r_xprt = cq->cq_context;
/* WARNING: Only wr_cqe and status are reliable at this point */
trace_xprtrdma_wc_receive(wc);
@@ -179,6 +197,7 @@ static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt)
return;
out_flushed:
+ rpcrdma_flush_disconnect(cq, wc);
rpcrdma_rep_destroy(rep);
}
@@ -395,7 +414,7 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
goto out_destroy;
}
- ep->rep_attr.recv_cq = ib_alloc_cq_any(id->device, NULL,
+ ep->rep_attr.recv_cq = ib_alloc_cq_any(id->device, r_xprt,
ep->rep_attr.cap.max_recv_wr,
IB_POLL_WORKQUEUE);
if (IS_ERR(ep->rep_attr.recv_cq)) {
@@ -452,6 +452,7 @@ struct rpcrdma_xprt {
/*
* Endpoint calls - xprtrdma/verbs.c
*/
+void rpcrdma_flush_disconnect(struct ib_cq *cq, struct ib_wc *wc);
int rpcrdma_xprt_connect(struct rpcrdma_xprt *r_xprt);
void rpcrdma_xprt_disconnect(struct rpcrdma_xprt *r_xprt);
Completion errors after a disconnect often occur much sooner than a CM_DISCONNECT event. Use this to try to detect connection loss more quickly. Note that other kernel ULPs do take care to disconnect explicitly when a WR is flushed. Signed-off-by: Chuck Lever <chuck.lever@oracle.com> --- include/trace/events/rpcrdma.h | 3 ++- net/sunrpc/xprtrdma/frwr_ops.c | 24 ++++++++++++++++-------- net/sunrpc/xprtrdma/verbs.c | 37 ++++++++++++++++++++++++++++--------- net/sunrpc/xprtrdma/xprt_rdma.h | 1 + 4 files changed, 47 insertions(+), 18 deletions(-)