Merge tag 'nfsd-6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

Pull nfsd updates from Chuck Lever:

 - Clean-ups in the READ path in anticipation of MSG_SPLICE_PAGES

 - Better NUMA awareness when allocating pages and other objects

 - A number of minor clean-ups to XDR encoding

 - Elimination of a race when accepting a TCP socket

 - Numerous observability enhancements

* tag 'nfsd-6.5' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux: (46 commits)
  nfsd: remove redundant assignments to variable len
  svcrdma: Fix stale comment
  NFSD: Distinguish per-net namespace initialization
  nfsd: move init of percpu reply_cache_stats counters back to nfsd_init_net
  SUNRPC: Address RCU warning in net/sunrpc/svc.c
  SUNRPC: Use sysfs_emit in place of strlcpy/sprintf
  SUNRPC: Remove transport class dprintk call sites
  SUNRPC: Fix comments for transport class registration
  svcrdma: Remove an unused argument from __svc_rdma_put_rw_ctxt()
  svcrdma: trace cc_release calls
  svcrdma: Convert "might sleep" comment into a code annotation
  NFSD: Add an nfsd4_encode_nfstime4() helper
  SUNRPC: Move initialization of rq_stime
  SUNRPC: Optimize page release in svc_rdma_sendto()
  svcrdma: Prevent page release when nothing was received
  svcrdma: Revert 2a1e4f21d841 ("svcrdma: Normalize Send page handling")
  SUNRPC: Revert 579900670ac7 ("svcrdma: Remove unused sc_pages field")
  SUNRPC: Revert cc93ce9529a6 ("svcrdma: Retain the page backing rq_res.head[0].iov_base")
  NFSD: add encoding of op_recall flag for write delegation
  NFSD: Add "official" reviewers for this subsystem
  ...

9 files changed, 143 insertions, 136 deletions

diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index 79967b6925bd..e7c101290425 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -109,15 +109,15 @@ param_get_pool_mode(char *buf, const struct kernel_param *kp)
 	switch (*ip)
 	{
 	case SVC_POOL_AUTO:
-		return strlcpy(buf, "auto\n", 20);
+		return sysfs_emit(buf, "auto\n");
 	case SVC_POOL_GLOBAL:
-		return strlcpy(buf, "global\n", 20);
+		return sysfs_emit(buf, "global\n");
 	case SVC_POOL_PERCPU:
-		return strlcpy(buf, "percpu\n", 20);
+		return sysfs_emit(buf, "percpu\n");
 	case SVC_POOL_PERNODE:
-		return strlcpy(buf, "pernode\n", 20);
+		return sysfs_emit(buf, "pernode\n");
 	default:
-		return sprintf(buf, "%d\n", *ip);
+		return sysfs_emit(buf, "%d\n", *ip);
 	}
 }
 
@@ -597,34 +597,25 @@ svc_destroy(struct kref *ref)
 }
 EXPORT_SYMBOL_GPL(svc_destroy);
 
-/*
- * Allocate an RPC server's buffer space.
- * We allocate pages and place them in rq_pages.
- */
-static int
+static bool
 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
 {
-	unsigned int pages, arghi;
+	unsigned long pages, ret;
 
 	/* bc_xprt uses fore channel allocated buffers */
 	if (svc_is_backchannel(rqstp))
-		return 1;
+		return true;
 
 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
 				       * We assume one is at most one page
 				       */
-	arghi = 0;
 	WARN_ON_ONCE(pages > RPCSVC_MAXPAGES);
 	if (pages > RPCSVC_MAXPAGES)
 		pages = RPCSVC_MAXPAGES;
-	while (pages) {
-		struct page *p = alloc_pages_node(node, GFP_KERNEL, 0);
-		if (!p)
-			break;
-		rqstp->rq_pages[arghi++] = p;
-		pages--;
-	}
-	return pages == 0;
+
+	ret = alloc_pages_bulk_array_node(GFP_KERNEL, node, pages,
+					  rqstp->rq_pages);
+	return ret == pages;
 }
 
 /*
@@ -1173,6 +1164,7 @@ static void __svc_unregister(struct net *net, const u32 program, const u32 versi
  */
 static void svc_unregister(const struct svc_serv *serv, struct net *net)
 {
+	struct sighand_struct *sighand;
 	struct svc_program *progp;
 	unsigned long flags;
 	unsigned int i;
@@ -1189,9 +1181,12 @@ static void svc_unregister(const struct svc_serv *serv, struct net *net)
 		}
 	}
 
-	spin_lock_irqsave(&current->sighand->siglock, flags);
+	rcu_read_lock();
+	sighand = rcu_dereference(current->sighand);
+	spin_lock_irqsave(&sighand->siglock, flags);
 	recalc_sigpending();
-	spin_unlock_irqrestore(&current->sighand->siglock, flags);
+	spin_unlock_irqrestore(&sighand->siglock, flags);
+	rcu_read_unlock();
 }
 
 /*
diff --git a/net/sunrpc/svc_xprt.c b/net/sunrpc/svc_xprt.c
index 13a14897bc17..62c7919ea610 100644
--- a/net/sunrpc/svc_xprt.c
+++ b/net/sunrpc/svc_xprt.c
@@ -74,13 +74,18 @@ static LIST_HEAD(svc_xprt_class_list);
  *		  that no other thread will be using the transport or will
  *		  try to set XPT_DEAD.
  */
+
+/**
+ * svc_reg_xprt_class - Register a server-side RPC transport class
+ * @xcl: New transport class to be registered
+ *
+ * Returns zero on success; otherwise a negative errno is returned.
+ */
 int svc_reg_xprt_class(struct svc_xprt_class *xcl)
 {
 	struct svc_xprt_class *cl;
 	int res = -EEXIST;
 
-	dprintk("svc: Adding svc transport class '%s'\n", xcl->xcl_name);
-
 	INIT_LIST_HEAD(&xcl->xcl_list);
 	spin_lock(&svc_xprt_class_lock);
 	/* Make sure there isn't already a class with the same name */
@@ -96,9 +101,13 @@ out:
 }
 EXPORT_SYMBOL_GPL(svc_reg_xprt_class);
 
+/**
+ * svc_unreg_xprt_class - Unregister a server-side RPC transport class
+ * @xcl: Transport class to be unregistered
+ *
+ */
 void svc_unreg_xprt_class(struct svc_xprt_class *xcl)
 {
-	dprintk("svc: Removing svc transport class '%s'\n", xcl->xcl_name);
 	spin_lock(&svc_xprt_class_lock);
 	list_del_init(&xcl->xcl_list);
 	spin_unlock(&svc_xprt_class_lock);
@@ -685,8 +694,9 @@ static int svc_alloc_arg(struct svc_rqst *rqstp)
 	}
 
 	for (filled = 0; filled < pages; filled = ret) {
-		ret = alloc_pages_bulk_array(GFP_KERNEL, pages,
-					     rqstp->rq_pages);
+		ret = alloc_pages_bulk_array_node(GFP_KERNEL,
+						  rqstp->rq_pool->sp_id,
+						  pages, rqstp->rq_pages);
 		if (ret > filled)
 			/* Made progress, don't sleep yet */
 			continue;
@@ -843,15 +853,11 @@ static int svc_handle_xprt(struct svc_rqst *rqstp, struct svc_xprt *xprt)
 		svc_xprt_received(xprt);
 	} else if (svc_xprt_reserve_slot(rqstp, xprt)) {
 		/* XPT_DATA|XPT_DEFERRED case: */
-		dprintk("svc: server %p, pool %u, transport %p, inuse=%d\n",
-			rqstp, rqstp->rq_pool->sp_id, xprt,
-			kref_read(&xprt->xpt_ref));
 		rqstp->rq_deferred = svc_deferred_dequeue(xprt);
 		if (rqstp->rq_deferred)
 			len = svc_deferred_recv(rqstp);
 		else
 			len = xprt->xpt_ops->xpo_recvfrom(rqstp);
-		rqstp->rq_stime = ktime_get();
 		rqstp->rq_reserved = serv->sv_max_mesg;
 		atomic_add(rqstp->rq_reserved, &xprt->xpt_reserved);
 	} else
@@ -894,6 +900,7 @@ int svc_recv(struct svc_rqst *rqstp, long timeout)
 	err = -EAGAIN;
 	if (len <= 0)
 		goto out_release;
+
 	trace_svc_xdr_recvfrom(&rqstp->rq_arg);
 
 	clear_bit(XPT_OLD, &xprt->xpt_flags);
@@ -902,6 +909,7 @@ int svc_recv(struct svc_rqst *rqstp, long timeout)
 
 	if (serv->sv_stats)
 		serv->sv_stats->netcnt++;
+	rqstp->rq_stime = ktime_get();
 	return len;
 out_release:
 	rqstp->rq_res.len = 0;
diff --git a/net/sunrpc/svcsock.c b/net/sunrpc/svcsock.c
index f77cebe2c071..5f519fc0541b 100644
--- a/net/sunrpc/svcsock.c
+++ b/net/sunrpc/svcsock.c
@@ -826,12 +826,6 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
 
 	trace_sk_data_ready(sk);
 
-	if (svsk) {
-		/* Refer to svc_setup_socket() for details. */
-		rmb();
-		svsk->sk_odata(sk);
-	}
-
 	/*
 	 * This callback may called twice when a new connection
 	 * is established as a child socket inherits everything
@@ -840,13 +834,18 @@ static void svc_tcp_listen_data_ready(struct sock *sk)
 	 *    when one of child sockets become ESTABLISHED.
 	 * 2) data_ready method of the child socket may be called
 	 *    when it receives data before the socket is accepted.
-	 * In case of 2, we should ignore it silently.
+	 * In case of 2, we should ignore it silently and DO NOT
+	 * dereference svsk.
 	 */
-	if (sk->sk_state == TCP_LISTEN) {
-		if (svsk) {
-			set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
-			svc_xprt_enqueue(&svsk->sk_xprt);
-		}
+	if (sk->sk_state != TCP_LISTEN)
+		return;
+
+	if (svsk) {
+		/* Refer to svc_setup_socket() for details. */
+		rmb();
+		svsk->sk_odata(sk);
+		set_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
+		svc_xprt_enqueue(&svsk->sk_xprt);
 	}
 }
 
@@ -887,13 +886,8 @@ static struct svc_xprt *svc_tcp_accept(struct svc_xprt *xprt)
 	clear_bit(XPT_CONN, &svsk->sk_xprt.xpt_flags);
 	err = kernel_accept(sock, &newsock, O_NONBLOCK);
 	if (err < 0) {
-		if (err == -ENOMEM)
-			printk(KERN_WARNING "%s: no more sockets!\n",
-			       serv->sv_name);
-		else if (err != -EAGAIN)
-			net_warn_ratelimited("%s: accept failed (err %d)!\n",
-					     serv->sv_name, -err);
-		trace_svcsock_accept_err(xprt, serv->sv_name, err);
+		if (err != -EAGAIN)
+			trace_svcsock_accept_err(xprt, serv->sv_name, err);
 		return NULL;
 	}
 	if (IS_ERR(sock_alloc_file(newsock, O_NONBLOCK, NULL)))
@@ -1464,7 +1458,7 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
 	svsk->sk_owspace = inet->sk_write_space;
 	/*
 	 * This barrier is necessary in order to prevent race condition
-	 * with svc_data_ready(), svc_listen_data_ready() and others
+	 * with svc_data_ready(), svc_tcp_listen_data_ready(), and others
 	 * when calling callbacks above.
 	 */
 	wmb();
@@ -1476,7 +1470,7 @@ static struct svc_sock *svc_setup_socket(struct svc_serv *serv,
 	else
 		svc_tcp_init(svsk, serv);
 
-	trace_svcsock_new_socket(sock);
+	trace_svcsock_new(svsk, sock);
 	return svsk;
 }
 
@@ -1657,6 +1651,8 @@ static void svc_sock_free(struct svc_xprt *xprt)
 	struct svc_sock *svsk = container_of(xprt, struct svc_sock, sk_xprt);
 	struct socket *sock = svsk->sk_sock;
 
+	trace_svcsock_free(svsk, sock);
+
 	tls_handshake_cancel(sock->sk);
 	if (sock->file)
 		sockfd_put(sock);
diff --git a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
index 36835b2f5446..2a22e78af116 100644
--- a/net/sunrpc/xdr.c
+++ b/net/sunrpc/xdr.c
@@ -1070,22 +1070,22 @@ __be32 * xdr_reserve_space(struct xdr_stream *xdr, size_t nbytes)
 }
 EXPORT_SYMBOL_GPL(xdr_reserve_space);
 
-
 /**
  * xdr_reserve_space_vec - Reserves a large amount of buffer space for sending
  * @xdr: pointer to xdr_stream
- * @vec: pointer to a kvec array
  * @nbytes: number of bytes to reserve
  *
- * Reserves enough buffer space to encode 'nbytes' of data and stores the
- * pointers in 'vec'. The size argument passed to xdr_reserve_space() is
- * determined based on the number of bytes remaining in the current page to
- * avoid invalidating iov_base pointers when xdr_commit_encode() is called.
+ * The size argument passed to xdr_reserve_space() is determined based
+ * on the number of bytes remaining in the current page to avoid
+ * invalidating iov_base pointers when xdr_commit_encode() is called.
+ *
+ * Return values:
+ *   %0: success
+ *   %-EMSGSIZE: not enough space is available in @xdr
  */
-int xdr_reserve_space_vec(struct xdr_stream *xdr, struct kvec *vec, size_t nbytes)
+int xdr_reserve_space_vec(struct xdr_stream *xdr, size_t nbytes)
 {
-	int thislen;
-	int v = 0;
+	size_t thislen;
 	__be32 *p;
 
 	/*
@@ -1097,21 +1097,19 @@ int xdr_reserve_space_vec(struct xdr_stream *xdr, struct kvec *vec, size_t nbyte
 		xdr->end = xdr->p;
 	}
 
+	/* XXX: Let's find a way to make this more efficient */
 	while (nbytes) {
 		thislen = xdr->buf->page_len % PAGE_SIZE;
 		thislen = min_t(size_t, nbytes, PAGE_SIZE - thislen);
 
 		p = xdr_reserve_space(xdr, thislen);
 		if (!p)
-			return -EIO;
+			return -EMSGSIZE;
 
-		vec[v].iov_base = p;
-		vec[v].iov_len = thislen;
-		v++;
 		nbytes -= thislen;
 	}
 
-	return v;
+	return 0;
 }
 EXPORT_SYMBOL_GPL(xdr_reserve_space_vec);
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
index aa2227a7e552..7420a2c990c7 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
@@ -93,13 +93,7 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
 	 */
 	get_page(virt_to_page(rqst->rq_buffer));
 	sctxt->sc_send_wr.opcode = IB_WR_SEND;
-	ret = svc_rdma_send(rdma, sctxt);
-	if (ret < 0)
-		return ret;
-
-	ret = wait_for_completion_killable(&sctxt->sc_done);
-	svc_rdma_send_ctxt_put(rdma, sctxt);
-	return ret;
+	return svc_rdma_send(rdma, sctxt);
 }
 
 /* Server-side transport endpoint wants a whole page for its send
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index a22fe7587fa6..85c8bcaebb80 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -125,14 +125,15 @@ static void svc_rdma_recv_cid_init(struct svcxprt_rdma *rdma,
 static struct svc_rdma_recv_ctxt *
 svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma)
 {
+	int node = ibdev_to_node(rdma->sc_cm_id->device);
 	struct svc_rdma_recv_ctxt *ctxt;
 	dma_addr_t addr;
 	void *buffer;
 
-	ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL);
+	ctxt = kmalloc_node(sizeof(*ctxt), GFP_KERNEL, node);
 	if (!ctxt)
 		goto fail0;
-	buffer = kmalloc(rdma->sc_max_req_size, GFP_KERNEL);
+	buffer = kmalloc_node(rdma->sc_max_req_size, GFP_KERNEL, node);
 	if (!buffer)
 		goto fail1;
 	addr = ib_dma_map_single(rdma->sc_pd->device, buffer,
@@ -155,7 +156,6 @@ svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma)
 	ctxt->rc_recv_sge.length = rdma->sc_max_req_size;
 	ctxt->rc_recv_sge.lkey = rdma->sc_pd->local_dma_lkey;
 	ctxt->rc_recv_buf = buffer;
-	ctxt->rc_temp = false;
 	return ctxt;
 
 fail2:
@@ -232,10 +232,7 @@ void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma,
 	pcl_free(&ctxt->rc_write_pcl);
 	pcl_free(&ctxt->rc_reply_pcl);
 
-	if (!ctxt->rc_temp)
-		llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts);
-	else
-		svc_rdma_recv_ctxt_destroy(rdma, ctxt);
+	llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts);
 }
 
 /**
@@ -258,7 +255,7 @@ void svc_rdma_release_ctxt(struct svc_xprt *xprt, void *vctxt)
 }
 
 static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma,
-				   unsigned int wanted, bool temp)
+				   unsigned int wanted)
 {
 	const struct ib_recv_wr *bad_wr = NULL;
 	struct svc_rdma_recv_ctxt *ctxt;
@@ -275,7 +272,6 @@ static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma,
 			break;
 
 		trace_svcrdma_post_recv(ctxt);
-		ctxt->rc_temp = temp;
 		ctxt->rc_recv_wr.next = recv_chain;
 		recv_chain = &ctxt->rc_recv_wr;
 		rdma->sc_pending_recvs++;
@@ -309,7 +305,7 @@ err_free:
  */
 bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma)
 {
-	return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests, true);
+	return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests);
 }
 
 /**
@@ -343,7 +339,7 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	 * client reconnects.
 	 */
 	if (rdma->sc_pending_recvs < rdma->sc_max_requests)
-		if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch, false))
+		if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch))
 			goto dropped;
 
 	/* All wc fields are now known to be valid */
@@ -775,9 +771,6 @@ static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt,
  *
  * The next ctxt is removed from the "receive" lists.
  *
- * - If the ctxt completes a Read, then finish assembling the Call
- *   message and return the number of bytes in the message.
- *
  * - If the ctxt completes a Receive, then construct the Call
  *   message from the contents of the Receive buffer.
  *
@@ -786,7 +779,8 @@ static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt,
  *     in the message.
  *
  *   - If there are Read chunks in this message, post Read WRs to
- *     pull that payload and return 0.
+ *     pull that payload. When the Read WRs complete, build the
+ *     full message and return the number of bytes in it.
  */
 int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 {
@@ -796,6 +790,12 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 	struct svc_rdma_recv_ctxt *ctxt;
 	int ret;
 
+	/* Prevent svc_xprt_release() from releasing pages in rq_pages
+	 * when returning 0 or an error.
+	 */
+	rqstp->rq_respages = rqstp->rq_pages;
+	rqstp->rq_next_page = rqstp->rq_respages;
+
 	rqstp->rq_xprt_ctxt = NULL;
 
 	ctxt = NULL;
@@ -819,12 +819,6 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 				   DMA_FROM_DEVICE);
 	svc_rdma_build_arg_xdr(rqstp, ctxt);
 
-	/* Prevent svc_xprt_release from releasing pages in rq_pages
-	 * if we return 0 or an error.
-	 */
-	rqstp->rq_respages = rqstp->rq_pages;
-	rqstp->rq_next_page = rqstp->rq_respages;
-
 	ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg, ctxt);
 	if (ret < 0)
 		goto out_err;
diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c
index 11cf7c646644..e460e25a1d6d 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_rw.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c
@@ -62,8 +62,8 @@ svc_rdma_get_rw_ctxt(struct svcxprt_rdma *rdma, unsigned int sges)
 	if (node) {
 		ctxt = llist_entry(node, struct svc_rdma_rw_ctxt, rw_node);
 	} else {
-		ctxt = kmalloc(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE),
-			       GFP_KERNEL);
+		ctxt = kmalloc_node(struct_size(ctxt, rw_first_sgl, SG_CHUNK_SIZE),
+				    GFP_KERNEL, ibdev_to_node(rdma->sc_cm_id->device));
 		if (!ctxt)
 			goto out_noctx;
 
@@ -84,8 +84,7 @@ out_noctx:
 	return NULL;
 }
 
-static void __svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
-				   struct svc_rdma_rw_ctxt *ctxt,
+static void __svc_rdma_put_rw_ctxt(struct svc_rdma_rw_ctxt *ctxt,
 				   struct llist_head *list)
 {
 	sg_free_table_chained(&ctxt->rw_sg_table, SG_CHUNK_SIZE);
@@ -95,7 +94,7 @@ static void __svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
 static void svc_rdma_put_rw_ctxt(struct svcxprt_rdma *rdma,
 				 struct svc_rdma_rw_ctxt *ctxt)
 {
-	__svc_rdma_put_rw_ctxt(rdma, ctxt, &rdma->sc_rw_ctxts);
+	__svc_rdma_put_rw_ctxt(ctxt, &rdma->sc_rw_ctxts);
 }
 
 /**
@@ -191,6 +190,8 @@ static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc,
 	struct svc_rdma_rw_ctxt *ctxt;
 	LLIST_HEAD(free);
 
+	trace_svcrdma_cc_release(&cc->cc_cid, cc->cc_sqecount);
+
 	first = last = NULL;
 	while ((ctxt = svc_rdma_next_ctxt(&cc->cc_rwctxts)) != NULL) {
 		list_del(&ctxt->rw_list);
@@ -198,7 +199,7 @@ static void svc_rdma_cc_release(struct svc_rdma_chunk_ctxt *cc,
 		rdma_rw_ctx_destroy(&ctxt->rw_ctx, rdma->sc_qp,
 				    rdma->sc_port_num, ctxt->rw_sg_table.sgl,
 				    ctxt->rw_nents, dir);
-		__svc_rdma_put_rw_ctxt(rdma, ctxt, &free);
+		__svc_rdma_put_rw_ctxt(ctxt, &free);
 
 		ctxt->rw_node.next = first;
 		first = &ctxt->rw_node;
@@ -234,7 +235,8 @@ svc_rdma_write_info_alloc(struct svcxprt_rdma *rdma,
 {
 	struct svc_rdma_write_info *info;
 
-	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	info = kmalloc_node(sizeof(*info), GFP_KERNEL,
+			    ibdev_to_node(rdma->sc_cm_id->device));
 	if (!info)
 		return info;
 
@@ -304,7 +306,8 @@ svc_rdma_read_info_alloc(struct svcxprt_rdma *rdma)
 {
 	struct svc_rdma_read_info *info;
 
-	info = kmalloc(sizeof(*info), GFP_KERNEL);
+	info = kmalloc_node(sizeof(*info), GFP_KERNEL,
+			    ibdev_to_node(rdma->sc_cm_id->device));
 	if (!info)
 		return info;
 
@@ -351,8 +354,7 @@ static void svc_rdma_wc_read_done(struct ib_cq *cq, struct ib_wc *wc)
 	return;
 }
 
-/* This function sleeps when the transport's Send Queue is congested.
- *
+/*
  * Assumptions:
  * - If ib_post_send() succeeds, only one completion is expected,
  *   even if one or more WRs are flushed. This is true when posting
@@ -367,6 +369,8 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc)
 	struct ib_cqe *cqe;
 	int ret;
 
+	might_sleep();
+
 	if (cc->cc_sqecount > rdma->sc_sq_depth)
 		return -EINVAL;
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 22a871e6fe4d..c6644cca52c5 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -123,18 +123,17 @@ static void svc_rdma_send_cid_init(struct svcxprt_rdma *rdma,
 static struct svc_rdma_send_ctxt *
 svc_rdma_send_ctxt_alloc(struct svcxprt_rdma *rdma)
 {
+	int node = ibdev_to_node(rdma->sc_cm_id->device);
 	struct svc_rdma_send_ctxt *ctxt;
 	dma_addr_t addr;
 	void *buffer;
-	size_t size;
 	int i;
 
-	size = sizeof(*ctxt);
-	size += rdma->sc_max_send_sges * sizeof(struct ib_sge);
-	ctxt = kmalloc(size, GFP_KERNEL);
+	ctxt = kmalloc_node(struct_size(ctxt, sc_sges, rdma->sc_max_send_sges),
+			    GFP_KERNEL, node);
 	if (!ctxt)
 		goto fail0;
-	buffer = kmalloc(rdma->sc_max_req_size, GFP_KERNEL);
+	buffer = kmalloc_node(rdma->sc_max_req_size, GFP_KERNEL, node);
 	if (!buffer)
 		goto fail1;
 	addr = ib_dma_map_single(rdma->sc_pd->device, buffer,
@@ -148,7 +147,6 @@ svc_rdma_send_ctxt_alloc(struct svcxprt_rdma *rdma)
 	ctxt->sc_send_wr.wr_cqe = &ctxt->sc_cqe;
 	ctxt->sc_send_wr.sg_list = ctxt->sc_sges;
 	ctxt->sc_send_wr.send_flags = IB_SEND_SIGNALED;
-	init_completion(&ctxt->sc_done);
 	ctxt->sc_cqe.done = svc_rdma_wc_send;
 	ctxt->sc_xprt_buf = buffer;
 	xdr_buf_init(&ctxt->sc_hdrbuf, ctxt->sc_xprt_buf,
@@ -214,6 +212,7 @@ out:
 
 	ctxt->sc_send_wr.num_sge = 0;
 	ctxt->sc_cur_sge_no = 0;
+	ctxt->sc_page_count = 0;
 	return ctxt;
 
 out_empty:
@@ -228,6 +227,8 @@ out_empty:
  * svc_rdma_send_ctxt_put - Return send_ctxt to free list
  * @rdma: controlling svcxprt_rdma
  * @ctxt: object to return to the free list
+ *
+ * Pages left in sc_pages are DMA unmapped and released.
  */
 void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
 			    struct svc_rdma_send_ctxt *ctxt)
@@ -235,6 +236,9 @@ void svc_rdma_send_ctxt_put(struct svcxprt_rdma *rdma,
 	struct ib_device *device = rdma->sc_cm_id->device;
 	unsigned int i;
 
+	if (ctxt->sc_page_count)
+		release_pages(ctxt->sc_pages, ctxt->sc_page_count);
+
 	/* The first SGE contains the transport header, which
 	 * remains mapped until @ctxt is destroyed.
 	 */
@@ -281,12 +285,12 @@ static void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
 		container_of(cqe, struct svc_rdma_send_ctxt, sc_cqe);
 
 	svc_rdma_wake_send_waiters(rdma, 1);
-	complete(&ctxt->sc_done);
 
 	if (unlikely(wc->status != IB_WC_SUCCESS))
 		goto flushed;
 
 	trace_svcrdma_wc_send(wc, &ctxt->sc_cid);
+	svc_rdma_send_ctxt_put(rdma, ctxt);
 	return;
 
 flushed:
@@ -294,6 +298,7 @@ flushed:
 		trace_svcrdma_wc_send_err(wc, &ctxt->sc_cid);
 	else
 		trace_svcrdma_wc_send_flush(wc, &ctxt->sc_cid);
+	svc_rdma_send_ctxt_put(rdma, ctxt);
 	svc_xprt_deferred_close(&rdma->sc_xprt);
 }
 
@@ -310,7 +315,7 @@ int svc_rdma_send(struct svcxprt_rdma *rdma, struct svc_rdma_send_ctxt *ctxt)
 	struct ib_send_wr *wr = &ctxt->sc_send_wr;
 	int ret;
 
-	reinit_completion(&ctxt->sc_done);
+	might_sleep();
 
 	/* Sync the transport header buffer */
 	ib_dma_sync_single_for_device(rdma->sc_pd->device,
@@ -799,6 +804,25 @@ int svc_rdma_map_reply_msg(struct svcxprt_rdma *rdma,
 				       svc_rdma_xb_dma_map, &args);
 }
 
+/* The svc_rqst and all resources it owns are released as soon as
+ * svc_rdma_sendto returns. Transfer pages under I/O to the ctxt
+ * so they are released by the Send completion handler.
+ */
+static void svc_rdma_save_io_pages(struct svc_rqst *rqstp,
+				   struct svc_rdma_send_ctxt *ctxt)
+{
+	int i, pages = rqstp->rq_next_page - rqstp->rq_respages;
+
+	ctxt->sc_page_count += pages;
+	for (i = 0; i < pages; i++) {
+		ctxt->sc_pages[i] = rqstp->rq_respages[i];
+		rqstp->rq_respages[i] = NULL;
+	}
+
+	/* Prevent svc_xprt_release from releasing pages in rq_pages */
+	rqstp->rq_next_page = rqstp->rq_respages;
+}
+
 /* Prepare the portion of the RPC Reply that will be transmitted
  * via RDMA Send. The RPC-over-RDMA transport header is prepared
  * in sc_sges[0], and the RPC xdr_buf is prepared in following sges.
@@ -828,6 +852,8 @@ static int svc_rdma_send_reply_msg(struct svcxprt_rdma *rdma,
 	if (ret < 0)
 		return ret;
 
+	svc_rdma_save_io_pages(rqstp, sctxt);
+
 	if (rctxt->rc_inv_rkey) {
 		sctxt->sc_send_wr.opcode = IB_WR_SEND_WITH_INV;
 		sctxt->sc_send_wr.ex.invalidate_rkey = rctxt->rc_inv_rkey;
@@ -835,13 +861,7 @@ static int svc_rdma_send_reply_msg(struct svcxprt_rdma *rdma,
 		sctxt->sc_send_wr.opcode = IB_WR_SEND;
 	}
 
-	ret = svc_rdma_send(rdma, sctxt);
-	if (ret < 0)
-		return ret;
-
-	ret = wait_for_completion_killable(&sctxt->sc_done);
-	svc_rdma_send_ctxt_put(rdma, sctxt);
-	return ret;
+	return svc_rdma_send(rdma, sctxt);
 }
 
 /**
@@ -907,8 +927,7 @@ void svc_rdma_send_error_msg(struct svcxprt_rdma *rdma,
 	sctxt->sc_sges[0].length = sctxt->sc_hdrbuf.len;
 	if (svc_rdma_send(rdma, sctxt))
 		goto put_ctxt;
-
-	wait_for_completion_killable(&sctxt->sc_done);
+	return;
 
 put_ctxt:
 	svc_rdma_send_ctxt_put(rdma, sctxt);
@@ -976,17 +995,16 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	ret = svc_rdma_send_reply_msg(rdma, sctxt, rctxt, rqstp);
 	if (ret < 0)
 		goto put_ctxt;
-
-	/* Prevent svc_xprt_release() from releasing the page backing
-	 * rq_res.head[0].iov_base. It's no longer being accessed by
-	 * the I/O device. */
-	rqstp->rq_respages++;
 	return 0;
 
 reply_chunk:
 	if (ret != -E2BIG && ret != -EINVAL)
 		goto put_ctxt;
 
+	/* Send completion releases payload pages that were part
+	 * of previously posted RDMA Writes.
+	 */
+	svc_rdma_save_io_pages(rqstp, sctxt);
 	svc_rdma_send_error_msg(rdma, sctxt, rctxt, ret);
 	return 0;
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index ca04f7a6a085..2abd895046ee 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -64,7 +64,7 @@
 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
 
 static struct svcxprt_rdma *svc_rdma_create_xprt(struct svc_serv *serv,
-						 struct net *net);
+						 struct net *net, int node);
 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 					struct net *net,
 					struct sockaddr *sa, int salen,
@@ -123,14 +123,14 @@ static void qp_event_handler(struct ib_event *event, void *context)
 }
 
 static struct svcxprt_rdma *svc_rdma_create_xprt(struct svc_serv *serv,
-						 struct net *net)
+						 struct net *net, int node)
 {
-	struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
+	struct svcxprt_rdma *cma_xprt;
 
-	if (!cma_xprt) {
-		dprintk("svcrdma: failed to create new transport\n");
+	cma_xprt = kzalloc_node(sizeof(*cma_xprt), GFP_KERNEL, node);
+	if (!cma_xprt)
 		return NULL;
-	}
+
 	svc_xprt_init(net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
 	INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
 	INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
@@ -193,9 +193,9 @@ static void handle_connect_req(struct rdma_cm_id *new_cma_id,
 	struct svcxprt_rdma *newxprt;
 	struct sockaddr *sa;
 
-	/* Create a new transport */
 	newxprt = svc_rdma_create_xprt(listen_xprt->sc_xprt.xpt_server,
-				       listen_xprt->sc_xprt.xpt_net);
+				       listen_xprt->sc_xprt.xpt_net,
+				       ibdev_to_node(new_cma_id->device));
 	if (!newxprt)
 		return;
 	newxprt->sc_cm_id = new_cma_id;
@@ -304,7 +304,7 @@ static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 
 	if (sa->sa_family != AF_INET && sa->sa_family != AF_INET6)
 		return ERR_PTR(-EAFNOSUPPORT);
-	cma_xprt = svc_rdma_create_xprt(serv, net);
+	cma_xprt = svc_rdma_create_xprt(serv, net, NUMA_NO_NODE);
 	if (!cma_xprt)
 		return ERR_PTR(-ENOMEM);
 	set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);