SUNRPC: provide a mechanism for collecting stats in the RPC client

Add a simple mechanism for collecting stats in the RPC client.  Stats are
tabulated during xprt_release.  Note that per_cpu shenanigans are not
required here because the RPC client already serializes on the transport
write lock.

Test plan:
Compile kernel with CONFIG_NFS enabled.  Basic performance regression
testing with high-speed networking and high performance server.

Signed-off-by: Chuck Lever <cel@netapp.com>
Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>
diff --git a/include/linux/sunrpc/clnt.h b/include/linux/sunrpc/clnt.h
index f147e6b..0f36620 100644
--- a/include/linux/sunrpc/clnt.h
+++ b/include/linux/sunrpc/clnt.h
@@ -45,7 +45,8 @@
 	char *			cl_server;	/* server machine name */
 	char *			cl_protname;	/* protocol name */
 	struct rpc_auth *	cl_auth;	/* authenticator */
-	struct rpc_stat *	cl_stats;	/* statistics */
+	struct rpc_stat *	cl_stats;	/* per-program statistics */
+	struct rpc_iostats *	cl_metrics;	/* per-client statistics */
 
 	unsigned int		cl_softrtry : 1,/* soft timeouts */
 				cl_intr     : 1,/* interruptible */
diff --git a/include/linux/sunrpc/metrics.h b/include/linux/sunrpc/metrics.h
new file mode 100644
index 0000000..8f96e9d
--- /dev/null
+++ b/include/linux/sunrpc/metrics.h
@@ -0,0 +1,77 @@
+/*
+ *  linux/include/linux/sunrpc/metrics.h
+ *
+ *  Declarations for RPC client per-operation metrics
+ *
+ *  Copyright (C) 2005	Chuck Lever <cel@netapp.com>
+ *
+ *  RPC client per-operation statistics provide latency and retry
+ *  information about each type of RPC procedure in a given RPC program.
+ *  These statistics are not for detailed problem diagnosis, but simply
+ *  to indicate whether the problem is local or remote.
+ *
+ *  These counters are not meant to be human-readable, but are meant to be
+ *  integrated into system monitoring tools such as "sar" and "iostat".  As
+ *  such, the counters are sampled by the tools over time, and are never
+ *  zeroed after a file system is mounted.  Moving averages can be computed
+ *  by the tools by taking the difference between two instantaneous samples
+ *  and dividing that by the time between the samples.
+ *
+ *  The counters are maintained in a single array per RPC client, indexed
+ *  by procedure number.  There is no need to maintain separate counter
+ *  arrays per-CPU because these counters are always modified behind locks.
+ */
+
+#ifndef _LINUX_SUNRPC_METRICS_H
+#define _LINUX_SUNRPC_METRICS_H
+
+#include <linux/seq_file.h>
+
+#define RPC_IOSTATS_VERS	"1.0"
+
+struct rpc_iostats {
+	/*
+	 * These counters give an idea about how many request
+	 * transmissions are required, on average, to complete that
+	 * particular procedure.  Some procedures may require more
+	 * than one transmission because the server is unresponsive,
+	 * the client is retransmitting too aggressively, or the
+	 * requests are large and the network is congested.
+	 */
+	unsigned long		om_ops,		/* count of operations */
+				om_ntrans,	/* count of RPC transmissions */
+				om_timeouts;	/* count of major timeouts */
+
+	/*
+	 * These count how many bytes are sent and received for a
+	 * given RPC procedure type.  This indicates how much load a
+	 * particular procedure is putting on the network.  These
+	 * counts include the RPC and ULP headers, and the request
+	 * payload.
+	 */
+	unsigned long long      om_bytes_sent,	/* count of bytes out */
+				om_bytes_recv;	/* count of bytes in */
+
+	/*
+	 * The length of time an RPC request waits in queue before
+	 * transmission, the network + server latency of the request,
+	 * and the total time the request spent from init to release
+	 * are measured.
+	 */
+	unsigned long long	om_queue,	/* jiffies queued for xmit */
+				om_rtt,		/* jiffies for RPC RTT */
+				om_execute;	/* jiffies for RPC execution */
+} ____cacheline_aligned;
+
+struct rpc_task;
+struct rpc_clnt;
+
+/*
+ * EXPORTed functions for managing rpc_iostats structures
+ */
+struct rpc_iostats *	rpc_alloc_iostats(struct rpc_clnt *);
+void			rpc_count_iostats(struct rpc_task *);
+void			rpc_print_iostats(struct seq_file *, struct rpc_clnt *);
+void			rpc_free_iostats(struct rpc_iostats *);
+
+#endif /* _LINUX_SUNRPC_METRICS_H */
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 84eb5b4..0bb23e8 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -28,12 +28,11 @@
 #include <linux/mm.h>
 #include <linux/slab.h>
 #include <linux/utsname.h>
+#include <linux/workqueue.h>
 
 #include <linux/sunrpc/clnt.h>
-#include <linux/workqueue.h>
 #include <linux/sunrpc/rpc_pipe_fs.h>
-
-#include <linux/nfs.h>
+#include <linux/sunrpc/metrics.h>
 
 
 #define RPC_SLACK_SPACE		(1024)	/* total overkill */
@@ -147,6 +146,7 @@
 	clnt->cl_vers     = version->number;
 	clnt->cl_prot     = xprt->prot;
 	clnt->cl_stats    = program->stats;
+	clnt->cl_metrics  = rpc_alloc_iostats(clnt);
 	rpc_init_wait_queue(&clnt->cl_pmap_default.pm_bindwait, "bindwait");
 
 	if (!clnt->cl_port)
@@ -245,6 +245,7 @@
 	if (new->cl_auth)
 		atomic_inc(&new->cl_auth->au_count);
 	new->cl_pmap		= &new->cl_pmap_default;
+	new->cl_metrics         = rpc_alloc_iostats(clnt);
 	rpc_init_wait_queue(&new->cl_pmap_default.pm_bindwait, "bindwait");
 	return new;
 out_no_clnt:
@@ -315,6 +316,8 @@
 	if (clnt->cl_server != clnt->cl_inline_name)
 		kfree(clnt->cl_server);
 out_free:
+	rpc_free_iostats(clnt->cl_metrics);
+	clnt->cl_metrics = NULL;
 	if (clnt->cl_dentry)
 		dput(clnt->cl_dentry);
 	kfree(clnt);
diff --git a/net/sunrpc/stats.c b/net/sunrpc/stats.c
index 4979f22..24ac716 100644
--- a/net/sunrpc/stats.c
+++ b/net/sunrpc/stats.c
@@ -21,6 +21,7 @@
 #include <linux/seq_file.h>
 #include <linux/sunrpc/clnt.h>
 #include <linux/sunrpc/svcsock.h>
+#include <linux/sunrpc/metrics.h>
 
 #define RPCDBG_FACILITY	RPCDBG_MISC
 
@@ -106,6 +107,110 @@
 	}
 }
 
+/**
+ * rpc_alloc_iostats - allocate an rpc_iostats structure
+ * @clnt: RPC program, version, and xprt
+ *
+ */
+struct rpc_iostats *rpc_alloc_iostats(struct rpc_clnt *clnt)
+{
+	unsigned int ops = clnt->cl_maxproc;
+	size_t size = ops * sizeof(struct rpc_iostats);
+	struct rpc_iostats *new;
+
+	new = kmalloc(size, GFP_KERNEL);
+	if (new)
+		memset(new, 0 , size);
+	return new;
+}
+EXPORT_SYMBOL(rpc_alloc_iostats);
+
+/**
+ * rpc_free_iostats - release an rpc_iostats structure
+ * @stats: doomed rpc_iostats structure
+ *
+ */
+void rpc_free_iostats(struct rpc_iostats *stats)
+{
+	kfree(stats);
+}
+EXPORT_SYMBOL(rpc_free_iostats);
+
+/**
+ * rpc_count_iostats - tally up per-task stats
+ * @task: completed rpc_task
+ *
+ * Relies on the caller for serialization.
+ */
+void rpc_count_iostats(struct rpc_task *task)
+{
+	struct rpc_rqst *req = task->tk_rqstp;
+	struct rpc_iostats *stats = task->tk_client->cl_metrics;
+	struct rpc_iostats *op_metrics;
+	long rtt, execute, queue;
+
+	if (!stats || !req)
+		return;
+	op_metrics = &stats[task->tk_msg.rpc_proc->p_proc];
+
+	op_metrics->om_ops++;
+	op_metrics->om_ntrans += req->rq_ntrans;
+	op_metrics->om_timeouts += task->tk_timeouts;
+
+	op_metrics->om_bytes_sent += task->tk_bytes_sent;
+	op_metrics->om_bytes_recv += req->rq_received;
+
+	queue = (long)req->rq_xtime - task->tk_start;
+	if (queue < 0)
+		queue = -queue;
+	op_metrics->om_queue += queue;
+
+	rtt = task->tk_rtt;
+	if (rtt < 0)
+		rtt = -rtt;
+	op_metrics->om_rtt += rtt;
+
+	execute = (long)jiffies - task->tk_start;
+	if (execute < 0)
+		execute = -execute;
+	op_metrics->om_execute += execute;
+}
+
+#define MILLISECS_PER_JIFFY	(1000UL / HZ)
+
+void rpc_print_iostats(struct seq_file *seq, struct rpc_clnt *clnt)
+{
+	struct rpc_iostats *stats = clnt->cl_metrics;
+	struct rpc_xprt *xprt = clnt->cl_xprt;
+	unsigned int op, maxproc = clnt->cl_maxproc;
+
+	if (!stats)
+		return;
+
+	seq_printf(seq, "\tRPC iostats version: %s  ", RPC_IOSTATS_VERS);
+	seq_printf(seq, "p/v: %u/%u (%s)\n",
+			clnt->cl_prog, clnt->cl_vers, clnt->cl_protname);
+
+	if (xprt)
+		xprt->ops->print_stats(xprt, seq);
+
+	seq_printf(seq, "\tper-op statistics\n");
+	for (op = 0; op < maxproc; op++) {
+		struct rpc_iostats *metrics = &stats[op];
+		seq_printf(seq, "%12u: ", op);
+		seq_printf(seq, "%lu %lu %lu %Lu %Lu %Lu %Lu %Lu\n",
+				metrics->om_ops,
+				metrics->om_ntrans,
+				metrics->om_timeouts,
+				metrics->om_bytes_sent,
+				metrics->om_bytes_recv,
+				metrics->om_queue * MILLISECS_PER_JIFFY,
+				metrics->om_rtt * MILLISECS_PER_JIFFY,
+				metrics->om_execute * MILLISECS_PER_JIFFY);
+	}
+}
+EXPORT_SYMBOL(rpc_print_iostats);
+
 /*
  * Register/unregister RPC proc files
  */
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index c624197..eb5a262 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -44,6 +44,7 @@
 #include <linux/random.h>
 
 #include <linux/sunrpc/clnt.h>
+#include <linux/sunrpc/metrics.h>
 
 /*
  * Local variables
@@ -859,6 +860,7 @@
 
 	if (!(req = task->tk_rqstp))
 		return;
+	rpc_count_iostats(task);
 	spin_lock_bh(&xprt->transport_lock);
 	xprt->ops->release_xprt(xprt, task);
 	if (xprt->ops->release_request)