Merge tag 'nfsd-5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

Pull nfsd updates from Chuck Lever:

 - Update NFSv2 and NFSv3 XDR decoding functions

 - Further improve support for re-exporting NFS mounts

 - Convert NFSD stats to per-CPU counters

 - Add batch Receive posting to the server's RPC/RDMA transport

* tag 'nfsd-5.12' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux: (65 commits)
  nfsd: skip some unnecessary stats in the v4 case
  nfs: use change attribute for NFS re-exports
  NFSv4_2: SSC helper should use its own config.
  nfsd: cstate->session->se_client -> cstate->clp
  nfsd: simplify nfsd4_check_open_reclaim
  nfsd: remove unused set_client argument
  nfsd: find_cpntf_state cleanup
  nfsd: refactor set_client
  nfsd: rename lookup_clientid->set_client
  nfsd: simplify nfsd_renew
  nfsd: simplify process_lock
  nfsd4: simplify process_lookup1
  SUNRPC: Correct a comment
  svcrdma: DMA-sync the receive buffer in svc_rdma_recvfrom()
  svcrdma: Reduce Receive doorbell rate
  svcrdma: Deprecate stat variables that are no longer used
  svcrdma: Restore read and write stats
  svcrdma: Convert rdma_stat_sq_starve to a per-CPU counter
  svcrdma: Convert rdma_stat_recv to a per-CPU counter
  svcrdma: Refactor svc_rdma_init() and svc_rdma_clean_up()
  ...

5 files changed, 175 insertions, 121 deletions

diff --git a/net/sunrpc/svc.c b/net/sunrpc/svc.c
index 4187745887f0..61fb8a18552c 100644
--- a/net/sunrpc/svc.c
+++ b/net/sunrpc/svc.c
@@ -559,7 +559,7 @@ EXPORT_SYMBOL_GPL(svc_destroy);
 
 /*
  * Allocate an RPC server's buffer space.
- * We allocate pages and place them in rq_argpages.
+ * We allocate pages and place them in rq_pages.
  */
 static int
 svc_init_buffer(struct svc_rqst *rqstp, unsigned int size, int node)
diff --git a/net/sunrpc/xprtrdma/svc_rdma.c b/net/sunrpc/xprtrdma/svc_rdma.c
index 526da5d4710b..5bc20e9d09cd 100644
--- a/net/sunrpc/xprtrdma/svc_rdma.c
+++ b/net/sunrpc/xprtrdma/svc_rdma.c
@@ -62,51 +62,47 @@ static unsigned int max_max_requests = 16384;
 unsigned int svcrdma_max_req_size = RPCRDMA_DEF_INLINE_THRESH;
 static unsigned int min_max_inline = RPCRDMA_DEF_INLINE_THRESH;
 static unsigned int max_max_inline = RPCRDMA_MAX_INLINE_THRESH;
+static unsigned int svcrdma_stat_unused;
+static unsigned int zero;
 
-atomic_t rdma_stat_recv;
-atomic_t rdma_stat_read;
-atomic_t rdma_stat_write;
-atomic_t rdma_stat_sq_starve;
-atomic_t rdma_stat_rq_starve;
-atomic_t rdma_stat_rq_poll;
-atomic_t rdma_stat_rq_prod;
-atomic_t rdma_stat_sq_poll;
-atomic_t rdma_stat_sq_prod;
+struct percpu_counter svcrdma_stat_read;
+struct percpu_counter svcrdma_stat_recv;
+struct percpu_counter svcrdma_stat_sq_starve;
+struct percpu_counter svcrdma_stat_write;
 
-/*
- * This function implements reading and resetting an atomic_t stat
- * variable through read/write to a proc file. Any write to the file
- * resets the associated statistic to zero. Any read returns it's
- * current value.
- */
-static int read_reset_stat(struct ctl_table *table, int write,
-			   void *buffer, size_t *lenp, loff_t *ppos)
+enum {
+	SVCRDMA_COUNTER_BUFSIZ	= sizeof(unsigned long long),
+};
+
+static int svcrdma_counter_handler(struct ctl_table *table, int write,
+				   void *buffer, size_t *lenp, loff_t *ppos)
 {
-	atomic_t *stat = (atomic_t *)table->data;
-
-	if (!stat)
-		return -EINVAL;
-
-	if (write)
-		atomic_set(stat, 0);
-	else {
-		char str_buf[32];
-		int len = snprintf(str_buf, 32, "%d\n", atomic_read(stat));
-		if (len >= 32)
-			return -EFAULT;
-		len = strlen(str_buf);
-		if (*ppos > len) {
-			*lenp = 0;
-			return 0;
-		}
-		len -= *ppos;
-		if (len > *lenp)
-			len = *lenp;
-		if (len)
-			memcpy(buffer, str_buf, len);
-		*lenp = len;
-		*ppos += len;
+	struct percpu_counter *stat = (struct percpu_counter *)table->data;
+	char tmp[SVCRDMA_COUNTER_BUFSIZ + 1];
+	int len;
+
+	if (write) {
+		percpu_counter_set(stat, 0);
+		return 0;
 	}
+
+	len = snprintf(tmp, SVCRDMA_COUNTER_BUFSIZ, "%lld\n",
+		       percpu_counter_sum_positive(stat));
+	if (len >= SVCRDMA_COUNTER_BUFSIZ)
+		return -EFAULT;
+	len = strlen(tmp);
+	if (*ppos > len) {
+		*lenp = 0;
+		return 0;
+	}
+	len -= *ppos;
+	if (len > *lenp)
+		len = *lenp;
+	if (len)
+		memcpy(buffer, tmp, len);
+	*lenp = len;
+	*ppos += len;
+
 	return 0;
 }
 
@@ -142,66 +138,76 @@ static struct ctl_table svcrdma_parm_table[] = {
 
 	{
 		.procname	= "rdma_stat_read",
-		.data		= &rdma_stat_read,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_read,
+		.maxlen		= SVCRDMA_COUNTER_BUFSIZ,
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= svcrdma_counter_handler,
 	},
 	{
 		.procname	= "rdma_stat_recv",
-		.data		= &rdma_stat_recv,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_recv,
+		.maxlen		= SVCRDMA_COUNTER_BUFSIZ,
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= svcrdma_counter_handler,
 	},
 	{
 		.procname	= "rdma_stat_write",
-		.data		= &rdma_stat_write,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_write,
+		.maxlen		= SVCRDMA_COUNTER_BUFSIZ,
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= svcrdma_counter_handler,
 	},
 	{
 		.procname	= "rdma_stat_sq_starve",
-		.data		= &rdma_stat_sq_starve,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_sq_starve,
+		.maxlen		= SVCRDMA_COUNTER_BUFSIZ,
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= svcrdma_counter_handler,
 	},
 	{
 		.procname	= "rdma_stat_rq_starve",
-		.data		= &rdma_stat_rq_starve,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_unused,
+		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &zero,
 	},
 	{
 		.procname	= "rdma_stat_rq_poll",
-		.data		= &rdma_stat_rq_poll,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_unused,
+		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &zero,
 	},
 	{
 		.procname	= "rdma_stat_rq_prod",
-		.data		= &rdma_stat_rq_prod,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_unused,
+		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &zero,
 	},
 	{
 		.procname	= "rdma_stat_sq_poll",
-		.data		= &rdma_stat_sq_poll,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_unused,
+		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &zero,
 	},
 	{
 		.procname	= "rdma_stat_sq_prod",
-		.data		= &rdma_stat_sq_prod,
-		.maxlen		= sizeof(atomic_t),
+		.data		= &svcrdma_stat_unused,
+		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= read_reset_stat,
+		.proc_handler	= proc_dointvec_minmax,
+		.extra1		= &zero,
+		.extra2		= &zero,
 	},
 	{ },
 };
@@ -224,27 +230,69 @@ static struct ctl_table svcrdma_root_table[] = {
 	{ },
 };
 
+static void svc_rdma_proc_cleanup(void)
+{
+	if (!svcrdma_table_header)
+		return;
+	unregister_sysctl_table(svcrdma_table_header);
+	svcrdma_table_header = NULL;
+
+	percpu_counter_destroy(&svcrdma_stat_write);
+	percpu_counter_destroy(&svcrdma_stat_sq_starve);
+	percpu_counter_destroy(&svcrdma_stat_recv);
+	percpu_counter_destroy(&svcrdma_stat_read);
+}
+
+static int svc_rdma_proc_init(void)
+{
+	int rc;
+
+	if (svcrdma_table_header)
+		return 0;
+
+	rc = percpu_counter_init(&svcrdma_stat_read, 0, GFP_KERNEL);
+	if (rc)
+		goto out_err;
+	rc = percpu_counter_init(&svcrdma_stat_recv, 0, GFP_KERNEL);
+	if (rc)
+		goto out_err;
+	rc = percpu_counter_init(&svcrdma_stat_sq_starve, 0, GFP_KERNEL);
+	if (rc)
+		goto out_err;
+	rc = percpu_counter_init(&svcrdma_stat_write, 0, GFP_KERNEL);
+	if (rc)
+		goto out_err;
+
+	svcrdma_table_header = register_sysctl_table(svcrdma_root_table);
+	return 0;
+
+out_err:
+	percpu_counter_destroy(&svcrdma_stat_sq_starve);
+	percpu_counter_destroy(&svcrdma_stat_recv);
+	percpu_counter_destroy(&svcrdma_stat_read);
+	return rc;
+}
+
 void svc_rdma_cleanup(void)
 {
 	dprintk("SVCRDMA Module Removed, deregister RPC RDMA transport\n");
-	if (svcrdma_table_header) {
-		unregister_sysctl_table(svcrdma_table_header);
-		svcrdma_table_header = NULL;
-	}
 	svc_unreg_xprt_class(&svc_rdma_class);
+	svc_rdma_proc_cleanup();
 }
 
 int svc_rdma_init(void)
 {
+	int rc;
+
 	dprintk("SVCRDMA Module Init, register RPC RDMA transport\n");
 	dprintk("\tsvcrdma_ord      : %d\n", svcrdma_ord);
 	dprintk("\tmax_requests     : %u\n", svcrdma_max_requests);
 	dprintk("\tmax_bc_requests  : %u\n", svcrdma_max_bc_requests);
 	dprintk("\tmax_inline       : %d\n", svcrdma_max_req_size);
 
-	if (!svcrdma_table_header)
-		svcrdma_table_header =
-			register_sysctl_table(svcrdma_root_table);
+	rc = svc_rdma_proc_init();
+	if (rc)
+		return rc;
 
 	/* Register RDMA with the SVC transport switch */
 	svc_reg_xprt_class(&svc_rdma_class);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index cbdb71247755..6d28f23ceb35 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -266,33 +266,46 @@ void svc_rdma_release_rqst(struct svc_rqst *rqstp)
 		svc_rdma_recv_ctxt_put(rdma, ctxt);
 }
 
-static int __svc_rdma_post_recv(struct svcxprt_rdma *rdma,
-				struct svc_rdma_recv_ctxt *ctxt)
+static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma,
+				   unsigned int wanted, bool temp)
 {
+	const struct ib_recv_wr *bad_wr = NULL;
+	struct svc_rdma_recv_ctxt *ctxt;
+	struct ib_recv_wr *recv_chain;
 	int ret;
 
-	trace_svcrdma_post_recv(ctxt);
-	ret = ib_post_recv(rdma->sc_qp, &ctxt->rc_recv_wr, NULL);
+	recv_chain = NULL;
+	while (wanted--) {
+		ctxt = svc_rdma_recv_ctxt_get(rdma);
+		if (!ctxt)
+			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++;
+	}
+	if (!recv_chain)
+		return false;
+
+	ret = ib_post_recv(rdma->sc_qp, recv_chain, &bad_wr);
 	if (ret)
 		goto err_post;
-	return 0;
+	return true;
 
 err_post:
-	trace_svcrdma_rq_post_err(rdma, ret);
-	svc_rdma_recv_ctxt_put(rdma, ctxt);
-	return ret;
-}
-
-static int svc_rdma_post_recv(struct svcxprt_rdma *rdma)
-{
-	struct svc_rdma_recv_ctxt *ctxt;
+	while (bad_wr) {
+		ctxt = container_of(bad_wr, struct svc_rdma_recv_ctxt,
+				    rc_recv_wr);
+		bad_wr = bad_wr->next;
+		svc_rdma_recv_ctxt_put(rdma, ctxt);
+	}
 
-	if (test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags))
-		return 0;
-	ctxt = svc_rdma_recv_ctxt_get(rdma);
-	if (!ctxt)
-		return -ENOMEM;
-	return __svc_rdma_post_recv(rdma, ctxt);
+	trace_svcrdma_rq_post_err(rdma, ret);
+	/* Since we're destroying the xprt, no need to reset
+	 * sc_pending_recvs. */
+	return false;
 }
 
 /**
@@ -303,20 +316,7 @@ static int svc_rdma_post_recv(struct svcxprt_rdma *rdma)
  */
 bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma)
 {
-	struct svc_rdma_recv_ctxt *ctxt;
-	unsigned int i;
-	int ret;
-
-	for (i = 0; i < rdma->sc_max_requests; i++) {
-		ctxt = svc_rdma_recv_ctxt_get(rdma);
-		if (!ctxt)
-			return false;
-		ctxt->rc_temp = true;
-		ret = __svc_rdma_post_recv(rdma, ctxt);
-		if (ret)
-			return false;
-	}
-	return true;
+	return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests, true);
 }
 
 /**
@@ -324,8 +324,6 @@ bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma)
  * @cq: Completion Queue context
  * @wc: Work Completion object
  *
- * NB: The svc_xprt/svcxprt_rdma is pinned whenever it's possible that
- * the Receive completion handler could be running.
  */
 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 {
@@ -333,6 +331,8 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	struct ib_cqe *cqe = wc->wr_cqe;
 	struct svc_rdma_recv_ctxt *ctxt;
 
+	rdma->sc_pending_recvs--;
+
 	/* WARNING: Only wc->wr_cqe and wc->status are reliable */
 	ctxt = container_of(cqe, struct svc_rdma_recv_ctxt, rc_cqe);
 
@@ -340,14 +340,8 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	if (wc->status != IB_WC_SUCCESS)
 		goto flushed;
 
-	if (svc_rdma_post_recv(rdma))
-		goto post_err;
-
 	/* All wc fields are now known to be valid */
 	ctxt->rc_byte_len = wc->byte_len;
-	ib_dma_sync_single_for_cpu(rdma->sc_pd->device,
-				   ctxt->rc_recv_sge.addr,
-				   wc->byte_len, DMA_FROM_DEVICE);
 
 	spin_lock(&rdma->sc_rq_dto_lock);
 	list_add_tail(&ctxt->rc_list, &rdma->sc_rq_dto_q);
@@ -356,11 +350,18 @@ static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
 	spin_unlock(&rdma->sc_rq_dto_lock);
 	if (!test_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags))
 		svc_xprt_enqueue(&rdma->sc_xprt);
+
+	if (!test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags) &&
+	    rdma->sc_pending_recvs < rdma->sc_max_requests)
+		if (!svc_rdma_refresh_recvs(rdma, RPCRDMA_MAX_RECV_BATCH,
+					    false))
+			goto post_err;
+
 	return;
 
 flushed:
-post_err:
 	svc_rdma_recv_ctxt_put(rdma, ctxt);
+post_err:
 	set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
 	svc_xprt_enqueue(&rdma->sc_xprt);
 }
@@ -845,9 +846,11 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 	}
 	list_del(&ctxt->rc_list);
 	spin_unlock(&rdma_xprt->sc_rq_dto_lock);
+	percpu_counter_inc(&svcrdma_stat_recv);
 
-	atomic_inc(&rdma_stat_recv);
-
+	ib_dma_sync_single_for_cpu(rdma_xprt->sc_pd->device,
+				   ctxt->rc_recv_sge.addr, ctxt->rc_byte_len,
+				   DMA_FROM_DEVICE);
 	svc_rdma_build_arg_xdr(rqstp, ctxt);
 
 	/* Prevent svc_xprt_release from releasing pages in rq_pages
diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c
index 0b63e1321d74..693d139a8633 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_rw.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c
@@ -364,6 +364,7 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc)
 			return 0;
 		}
 
+		percpu_counter_inc(&svcrdma_stat_sq_starve);
 		trace_svcrdma_sq_full(rdma);
 		atomic_add(cc->cc_sqecount, &rdma->sc_sq_avail);
 		wait_event(rdma->sc_send_wait,
@@ -468,6 +469,7 @@ svc_rdma_build_writes(struct svc_rdma_write_info *info,
 					   DMA_TO_DEVICE);
 		if (ret < 0)
 			return -EIO;
+		percpu_counter_inc(&svcrdma_stat_write);
 
 		list_add(&ctxt->rw_list, &cc->cc_rwctxts);
 		cc->cc_sqecount += ret;
@@ -718,6 +720,7 @@ static int svc_rdma_build_read_segment(struct svc_rdma_read_info *info,
 				   segment->rs_handle, DMA_FROM_DEVICE);
 	if (ret < 0)
 		return -EIO;
+	percpu_counter_inc(&svcrdma_stat_read);
 
 	list_add(&ctxt->rw_list, &cc->cc_rwctxts);
 	cc->cc_sqecount += ret;
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 68af79d4f04f..52c759a8543e 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -317,7 +317,7 @@ int svc_rdma_send(struct svcxprt_rdma *rdma, struct svc_rdma_send_ctxt *ctxt)
 	/* If the SQ is full, wait until an SQ entry is available */
 	while (1) {
 		if ((atomic_dec_return(&rdma->sc_sq_avail) < 0)) {
-			atomic_inc(&rdma_stat_sq_starve);
+			percpu_counter_inc(&svcrdma_stat_sq_starve);
 			trace_svcrdma_sq_full(rdma);
 			atomic_inc(&rdma->sc_sq_avail);
 			wait_event(rdma->sc_send_wait,