1 files changed, 84 insertions, 95 deletions

diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index 3d3673ba9e1e..63262ef0c2e3 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -49,46 +49,36 @@
  *  o buffer memory
  */
 
+#include <linux/bitops.h>
 #include <linux/interrupt.h>
 #include <linux/slab.h>
 #include <linux/sunrpc/addr.h>
 #include <linux/sunrpc/svc_rdma.h>
 #include <linux/log2.h>
 
-#include <asm-generic/barrier.h>
-#include <asm/bitops.h>
+#include <asm/barrier.h>
 
 #include <rdma/ib_cm.h>
 
 #include "xprt_rdma.h"
 #include <trace/events/rpcrdma.h>
 
-/*
- * Globals/Macros
- */
-
-#if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
-# define RPCDBG_FACILITY	RPCDBG_TRANS
-#endif
-
-/*
- * internal functions
- */
 static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt,
 				       struct rpcrdma_sendctx *sc);
 static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt);
-static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep);
 static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt);
 static void rpcrdma_ep_get(struct rpcrdma_ep *ep);
 static int rpcrdma_ep_put(struct rpcrdma_ep *ep);
 static struct rpcrdma_regbuf *
-rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
-		     gfp_t flags);
+rpcrdma_regbuf_alloc_node(size_t size, enum dma_data_direction direction,
+			  int node);
+static struct rpcrdma_regbuf *
+rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction);
 static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb);
 static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb);
 
@@ -234,7 +224,6 @@ static void rpcrdma_update_cm_private(struct rpcrdma_ep *ep,
 static int
 rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
 {
-	struct sockaddr *sap = (struct sockaddr *)&id->route.addr.dst_addr;
 	struct rpcrdma_ep *ep = id->context;
 
 	might_sleep();
@@ -253,10 +242,6 @@ rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
 		ep->re_async_rc = -ENETUNREACH;
 		complete(&ep->re_done);
 		return 0;
-	case RDMA_CM_EVENT_DEVICE_REMOVAL:
-		pr_info("rpcrdma: removing device %s for %pISpc\n",
-			ep->re_id->device->name, sap);
-		fallthrough;
 	case RDMA_CM_EVENT_ADDR_CHANGE:
 		ep->re_connect_status = -ENODEV;
 		goto disconnected;
@@ -274,8 +259,6 @@ rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
 		ep->re_connect_status = -ENETUNREACH;
 		goto wake_connect_worker;
 	case RDMA_CM_EVENT_REJECTED:
-		dprintk("rpcrdma: connection to %pISpc rejected: %s\n",
-			sap, rdma_reject_msg(id, event->status));
 		ep->re_connect_status = -ECONNREFUSED;
 		if (event->status == IB_CM_REJ_STALE_CONN)
 			ep->re_connect_status = -ENOTCONN;
@@ -291,11 +274,17 @@ disconnected:
 		break;
 	}
 
-	dprintk("RPC:       %s: %pISpc on %s/frwr: %s\n", __func__, sap,
-		ep->re_id->device->name, rdma_event_msg(event->event));
 	return 0;
 }
 
+static void rpcrdma_ep_removal_done(struct rpcrdma_notification *rn)
+{
+	struct rpcrdma_ep *ep = container_of(rn, struct rpcrdma_ep, re_rn);
+
+	trace_xprtrdma_device_removal(ep->re_id);
+	xprt_force_disconnect(ep->re_xprt);
+}
+
 static struct rdma_cm_id *rpcrdma_create_id(struct rpcrdma_xprt *r_xprt,
 					    struct rpcrdma_ep *ep)
 {
@@ -335,6 +324,10 @@ static struct rdma_cm_id *rpcrdma_create_id(struct rpcrdma_xprt *r_xprt,
 	if (rc)
 		goto out;
 
+	rc = rpcrdma_rn_register(id->device, &ep->re_rn, rpcrdma_ep_removal_done);
+	if (rc)
+		goto out;
+
 	return id;
 
 out:
@@ -362,6 +355,8 @@ static void rpcrdma_ep_destroy(struct kref *kref)
 		ib_dealloc_pd(ep->re_pd);
 	ep->re_pd = NULL;
 
+	rpcrdma_rn_unregister(ep->re_id->device, &ep->re_rn);
+
 	kfree(ep);
 	module_put(THIS_MODULE);
 }
@@ -388,7 +383,7 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
 	struct rpcrdma_ep *ep;
 	int rc;
 
-	ep = kzalloc(sizeof(*ep), GFP_NOFS);
+	ep = kzalloc(sizeof(*ep), XPRTRDMA_GFP_FLAGS);
 	if (!ep)
 		return -ENOTCONN;
 	ep->re_xprt = &r_xprt->rx_xprt;
@@ -419,14 +414,6 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
 	ep->re_attr.qp_type = IB_QPT_RC;
 	ep->re_attr.port_num = ~0;
 
-	dprintk("RPC:       %s: requested max: dtos: send %d recv %d; "
-		"iovs: send %d recv %d\n",
-		__func__,
-		ep->re_attr.cap.max_send_wr,
-		ep->re_attr.cap.max_recv_wr,
-		ep->re_attr.cap.max_send_sge,
-		ep->re_attr.cap.max_recv_sge);
-
 	ep->re_send_batch = ep->re_max_requests >> 3;
 	ep->re_send_count = ep->re_send_batch;
 	init_waitqueue_head(&ep->re_connect_wait);
@@ -436,6 +423,7 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
 					      IB_POLL_WORKQUEUE);
 	if (IS_ERR(ep->re_attr.send_cq)) {
 		rc = PTR_ERR(ep->re_attr.send_cq);
+		ep->re_attr.send_cq = NULL;
 		goto out_destroy;
 	}
 
@@ -444,6 +432,7 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
 					      IB_POLL_WORKQUEUE);
 	if (IS_ERR(ep->re_attr.recv_cq)) {
 		rc = PTR_ERR(ep->re_attr.recv_cq);
+		ep->re_attr.recv_cq = NULL;
 		goto out_destroy;
 	}
 	ep->re_receive_count = 0;
@@ -482,6 +471,7 @@ static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt)
 	ep->re_pd = ib_alloc_pd(device, 0);
 	if (IS_ERR(ep->re_pd)) {
 		rc = PTR_ERR(ep->re_pd);
+		ep->re_pd = NULL;
 		goto out_destroy;
 	}
 
@@ -522,7 +512,7 @@ int rpcrdma_xprt_connect(struct rpcrdma_xprt *r_xprt)
 	 * outstanding Receives.
 	 */
 	rpcrdma_ep_get(ep);
-	rpcrdma_post_recvs(r_xprt, 1, true);
+	rpcrdma_post_recvs(r_xprt, 1);
 
 	rc = rdma_connect(ep->re_id, &ep->re_remote_cma);
 	if (rc)
@@ -626,7 +616,7 @@ static struct rpcrdma_sendctx *rpcrdma_sendctx_create(struct rpcrdma_ep *ep)
 	struct rpcrdma_sendctx *sc;
 
 	sc = kzalloc(struct_size(sc, sc_sges, ep->re_attr.cap.max_send_sge),
-		     GFP_KERNEL);
+		     XPRTRDMA_GFP_FLAGS);
 	if (!sc)
 		return NULL;
 
@@ -649,7 +639,7 @@ static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt)
 	 * Sends are posted.
 	 */
 	i = r_xprt->rx_ep->re_max_requests + RPCRDMA_MAX_BC_REQUESTS;
-	buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), GFP_KERNEL);
+	buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), XPRTRDMA_GFP_FLAGS);
 	if (!buf->rb_sc_ctxs)
 		return -ENOMEM;
 
@@ -760,13 +750,16 @@ rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt)
 {
 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
 	struct rpcrdma_ep *ep = r_xprt->rx_ep;
+	struct ib_device *device = ep->re_id->device;
 	unsigned int count;
 
+	/* Try to allocate enough to perform one full-sized I/O */
 	for (count = 0; count < ep->re_max_rdma_segs; count++) {
 		struct rpcrdma_mr *mr;
 		int rc;
 
-		mr = kzalloc(sizeof(*mr), GFP_NOFS);
+		mr = kzalloc_node(sizeof(*mr), XPRTRDMA_GFP_FLAGS,
+				  ibdev_to_node(device));
 		if (!mr)
 			break;
 
@@ -811,38 +804,33 @@ void rpcrdma_mrs_refresh(struct rpcrdma_xprt *r_xprt)
 	/* If there is no underlying connection, it's no use
 	 * to wake the refresh worker.
 	 */
-	if (ep->re_connect_status == 1) {
-		/* The work is scheduled on a WQ_MEM_RECLAIM
-		 * workqueue in order to prevent MR allocation
-		 * from recursing into NFS during direct reclaim.
-		 */
-		queue_work(xprtiod_workqueue, &buf->rb_refresh_worker);
-	}
+	if (ep->re_connect_status != 1)
+		return;
+	queue_work(system_highpri_wq, &buf->rb_refresh_worker);
 }
 
 /**
  * rpcrdma_req_create - Allocate an rpcrdma_req object
  * @r_xprt: controlling r_xprt
  * @size: initial size, in bytes, of send and receive buffers
- * @flags: GFP flags passed to memory allocators
  *
  * Returns an allocated and fully initialized rpcrdma_req or NULL.
  */
-struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, size_t size,
-				       gfp_t flags)
+struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt,
+				       size_t size)
 {
 	struct rpcrdma_buffer *buffer = &r_xprt->rx_buf;
 	struct rpcrdma_req *req;
 
-	req = kzalloc(sizeof(*req), flags);
+	req = kzalloc(sizeof(*req), XPRTRDMA_GFP_FLAGS);
 	if (req == NULL)
 		goto out1;
 
-	req->rl_sendbuf = rpcrdma_regbuf_alloc(size, DMA_TO_DEVICE, flags);
+	req->rl_sendbuf = rpcrdma_regbuf_alloc(size, DMA_TO_DEVICE);
 	if (!req->rl_sendbuf)
 		goto out2;
 
-	req->rl_recvbuf = rpcrdma_regbuf_alloc(size, DMA_NONE, flags);
+	req->rl_recvbuf = rpcrdma_regbuf_alloc(size, DMA_NONE);
 	if (!req->rl_recvbuf)
 		goto out3;
 
@@ -854,7 +842,7 @@ struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, size_t size,
 	return req;
 
 out3:
-	kfree(req->rl_sendbuf);
+	rpcrdma_regbuf_free(req->rl_sendbuf);
 out2:
 	kfree(req);
 out1:
@@ -878,7 +866,7 @@ int rpcrdma_req_setup(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
 		     r_xprt->rx_ep->re_max_rdma_segs * rpcrdma_readchunk_maxsz;
 	maxhdrsize *= sizeof(__be32);
 	rb = rpcrdma_regbuf_alloc(__roundup_pow_of_two(maxhdrsize),
-				  DMA_TO_DEVICE, GFP_KERNEL);
+				  DMA_TO_DEVICE);
 	if (!rb)
 		goto out;
 
@@ -916,6 +904,8 @@ static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt)
 
 static void rpcrdma_req_reset(struct rpcrdma_req *req)
 {
+	struct rpcrdma_mr *mr;
+
 	/* Credits are valid for only one connection */
 	req->rl_slot.rq_cong = 0;
 
@@ -925,7 +915,19 @@ static void rpcrdma_req_reset(struct rpcrdma_req *req)
 	rpcrdma_regbuf_dma_unmap(req->rl_sendbuf);
 	rpcrdma_regbuf_dma_unmap(req->rl_recvbuf);
 
-	frwr_reset(req);
+	/* The verbs consumer can't know the state of an MR on the
+	 * req->rl_registered list unless a successful completion
+	 * has occurred, so they cannot be re-used.
+	 */
+	while ((mr = rpcrdma_mr_pop(&req->rl_registered))) {
+		struct rpcrdma_buffer *buf = &mr->mr_xprt->rx_buf;
+
+		spin_lock(&buf->rb_lock);
+		list_del(&mr->mr_all);
+		spin_unlock(&buf->rb_lock);
+
+		frwr_mr_release(mr);
+	}
 }
 
 /* ASSUMPTION: the rb_allreqs list is stable for the duration,
@@ -943,24 +945,23 @@ static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt)
 }
 
 static noinline
-struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
-				       bool temp)
+struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt)
 {
 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
+	struct rpcrdma_ep *ep = r_xprt->rx_ep;
+	struct ib_device *device = ep->re_id->device;
 	struct rpcrdma_rep *rep;
 
-	rep = kzalloc(sizeof(*rep), GFP_KERNEL);
+	rep = kzalloc(sizeof(*rep), XPRTRDMA_GFP_FLAGS);
 	if (rep == NULL)
 		goto out;
 
-	rep->rr_rdmabuf = rpcrdma_regbuf_alloc(r_xprt->rx_ep->re_inline_recv,
-					       DMA_FROM_DEVICE, GFP_KERNEL);
+	rep->rr_rdmabuf = rpcrdma_regbuf_alloc_node(ep->re_inline_recv,
+						    DMA_FROM_DEVICE,
+						    ibdev_to_node(device));
 	if (!rep->rr_rdmabuf)
 		goto out_free;
 
-	if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf))
-		goto out_free_regbuf;
-
 	rep->rr_cid.ci_completion_id =
 		atomic_inc_return(&r_xprt->rx_ep->re_completion_ids);
 
@@ -972,15 +973,12 @@ struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt,
 	rep->rr_recv_wr.wr_cqe = &rep->rr_cqe;
 	rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov;
 	rep->rr_recv_wr.num_sge = 1;
-	rep->rr_temp = temp;
 
 	spin_lock(&buf->rb_lock);
 	list_add(&rep->rr_all, &buf->rb_all_reps);
 	spin_unlock(&buf->rb_lock);
 	return rep;
 
-out_free_regbuf:
-	rpcrdma_regbuf_free(rep->rr_rdmabuf);
 out_free:
 	kfree(rep);
 out:
@@ -993,17 +991,6 @@ static void rpcrdma_rep_free(struct rpcrdma_rep *rep)
 	kfree(rep);
 }
 
-static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep)
-{
-	struct rpcrdma_buffer *buf = &rep->rr_rxprt->rx_buf;
-
-	spin_lock(&buf->rb_lock);
-	list_del(&rep->rr_all);
-	spin_unlock(&buf->rb_lock);
-
-	rpcrdma_rep_free(rep);
-}
-
 static struct rpcrdma_rep *rpcrdma_rep_get_locked(struct rpcrdma_buffer *buf)
 {
 	struct llist_node *node;
@@ -1035,10 +1022,8 @@ static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt)
 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
 	struct rpcrdma_rep *rep;
 
-	list_for_each_entry(rep, &buf->rb_all_reps, rr_all) {
+	list_for_each_entry(rep, &buf->rb_all_reps, rr_all)
 		rpcrdma_regbuf_dma_unmap(rep->rr_rdmabuf);
-		rep->rr_temp = true;	/* Mark this rep for destruction */
-	}
 }
 
 static void rpcrdma_reps_destroy(struct rpcrdma_buffer *buf)
@@ -1084,8 +1069,8 @@ int rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt)
 	for (i = 0; i < r_xprt->rx_xprt.max_reqs; i++) {
 		struct rpcrdma_req *req;
 
-		req = rpcrdma_req_create(r_xprt, RPCRDMA_V1_DEF_INLINE_SIZE * 2,
-					 GFP_KERNEL);
+		req = rpcrdma_req_create(r_xprt,
+					 RPCRDMA_V1_DEF_INLINE_SIZE * 2);
 		if (!req)
 			goto out;
 		list_add(&req->rl_list, &buf->rb_send_bufs);
@@ -1255,15 +1240,15 @@ void rpcrdma_buffer_put(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req)
  * or Replies they may be registered externally via frwr_map.
  */
 static struct rpcrdma_regbuf *
-rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
-		     gfp_t flags)
+rpcrdma_regbuf_alloc_node(size_t size, enum dma_data_direction direction,
+			  int node)
 {
 	struct rpcrdma_regbuf *rb;
 
-	rb = kmalloc(sizeof(*rb), flags);
+	rb = kmalloc_node(sizeof(*rb), XPRTRDMA_GFP_FLAGS, node);
 	if (!rb)
 		return NULL;
-	rb->rg_data = kmalloc(size, flags);
+	rb->rg_data = kmalloc_node(size, XPRTRDMA_GFP_FLAGS, node);
 	if (!rb->rg_data) {
 		kfree(rb);
 		return NULL;
@@ -1275,6 +1260,12 @@ rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction,
 	return rb;
 }
 
+static struct rpcrdma_regbuf *
+rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction)
+{
+	return rpcrdma_regbuf_alloc_node(size, direction, NUMA_NO_NODE);
+}
+
 /**
  * rpcrdma_regbuf_realloc - re-allocate a SEND/RECV buffer
  * @rb: regbuf to reallocate
@@ -1352,10 +1343,9 @@ static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb)
  * rpcrdma_post_recvs - Refill the Receive Queue
  * @r_xprt: controlling transport instance
  * @needed: current credit grant
- * @temp: mark Receive buffers to be deleted after one use
  *
  */
-void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed, bool temp)
+void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed)
 {
 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
 	struct rpcrdma_ep *ep = r_xprt->rx_ep;
@@ -1369,8 +1359,7 @@ void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed, bool temp)
 	if (likely(ep->re_receive_count > needed))
 		goto out;
 	needed -= ep->re_receive_count;
-	if (!temp)
-		needed += RPCRDMA_MAX_RECV_BATCH;
+	needed += RPCRDMA_MAX_RECV_BATCH;
 
 	if (atomic_inc_return(&ep->re_receiving) > 1)
 		goto out;
@@ -1379,17 +1368,17 @@ void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, int needed, bool temp)
 	wr = NULL;
 	while (needed) {
 		rep = rpcrdma_rep_get_locked(buf);
-		if (rep && rep->rr_temp) {
-			rpcrdma_rep_destroy(rep);
-			continue;
-		}
 		if (!rep)
-			rep = rpcrdma_rep_create(r_xprt, temp);
+			rep = rpcrdma_rep_create(r_xprt);
 		if (!rep)
 			break;
+		if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf)) {
+			rpcrdma_rep_put(buf, rep);
+			break;
+		}
 
 		rep->rr_cid.ci_queue_id = ep->re_attr.recv_cq->res.id;
-		trace_xprtrdma_post_recv(rep);
+		trace_xprtrdma_post_recv(&rep->rr_cid);
 		rep->rr_recv_wr.next = wr;
 		wr = &rep->rr_recv_wr;
 		--needed;