I created a small experimental patch (leveraging the existing code for
the slow command check we already have in mid processing) to track the
total time per SMB2/SMB3 command. Seems like it could be useful -
here is sample output (see below) from running seven xfstests to
generate the stats (these stats require CONFIG_CIFS_STATS2 to be
compiled as do the slow command counters). I wanted to keep the
patch small and low risk so this seemed like the best place to capture
it. This could be helpful in calculating average time per command
(and perhaps help isolate performance problems, and isolate what is
really causing slowdowns). As with other counters: "echo 0 >
/proc/fs/cifs/Stats" resets them. The new stats counters start at
the line "Total time spent processing each command"
Thoughts?
$ cat /proc/fs/cifs/Stats
Resources in use
CIFS Session: 1
Share (unique mount targets): 3
SMB Request/Response Buffer: 1 Pool size: 5
SMB Small Req/Resp Buffer: 2 Pool size: 30
Total Large 60864 Small 1165977 Allocations
Operations (MIDs): 1
0 session 0 share reconnects
Total vfs operations: 280236 maximum at one time: 3
Total time spent processing each command
cmd: 0 jiffies: 0 seconds 0
cmd: 1 jiffies: 2 seconds 0
cmd: 2 jiffies: 0 seconds 0
cmd: 3 jiffies: 1 seconds 0
cmd: 4 jiffies: 1 seconds 0
cmd: 5 jiffies: 6212 seconds 24
cmd: 6 jiffies: 3993 seconds 15
cmd: 7 jiffies: 135 seconds 0
cmd: 8 jiffies: 83 seconds 0
cmd: 9 jiffies: 44 seconds 0
cmd: 10 jiffies: 0 seconds 0
cmd: 11 jiffies: 1 seconds 0
cmd: 12 jiffies: 0 seconds 0
cmd: 13 jiffies: 0 seconds 0
cmd: 14 jiffies: 47 seconds 0
cmd: 15 jiffies: 0 seconds 0
cmd: 16 jiffies: 1890 seconds 7
cmd: 17 jiffies: 1368 seconds 5
cmd: 18 jiffies: 0 seconds 0
1) \\localhost\btrfs-scratch
SMBs: 42255
Bytes read: 25379425 Bytes written: 53689661
Open files: 1 total (local), 1 open on server
TreeConnects: 1 total 0 failed
TreeDisconnects: 0 total 0 failed
Creates: 14012 total 0 failed
Closes: 14009 total 0 failed
Flushes: 27 total 0 failed
Reads: 41 total 0 failed
Writes: 112 total 0 failed
Locks: 0 total 0 failed
IOCTLs: 2 total 1 failed
QueryDirectories: 28 total 0 failed
ChangeNotifies: 0 total 0 failed
QueryInfos: 11312 total 0 failed
SetInfos: 2711 total 0 failed
OplockBreaks: 0 sent 0 failed
2) \\localhost\btrfs
SMBs: 77
Bytes read: 282 Bytes written: 94
Open files: 0 total (local), 0 open on server
TreeConnects: 1 total 0 failed
TreeDisconnects: 0 total 0 failed
Creates: 23 total 0 failed
Closes: 21 total 0 failed
Flushes: 0 total 0 failed
Reads: 3 total 0 failed
Writes: 1 total 0 failed
Locks: 0 total 0 failed
--
Thanks,
Steve
From 3f1ea43711cc8298263323a25e26227cbb22873d Mon Sep 17 00:00:00 2001
From: Steve French <stfrench@xxxxxxxxxxxxx>
Date: Tue, 26 Mar 2019 00:32:31 -0500
Subject: [PATCH] SMB3: track total time spent per command
This allows us to calculate average time spent in each SMB2/SMB3 command and
also helps determine where bottlenecks could be.
Signed-off-by: Steve French <stfrench@xxxxxxxxxxxxx>
---
fs/cifs/cifs_debug.c | 9 ++++++++-
fs/cifs/cifsglob.h | 3 ++-
fs/cifs/transport.c | 3 +++
3 files changed, 13 insertions(+), 2 deletions(-)
diff --git a/fs/cifs/cifs_debug.c b/fs/cifs/cifs_debug.c
index 13c1288b04a7..8fcfc7770ebb 100644
--- a/fs/cifs/cifs_debug.c
+++ b/fs/cifs/cifs_debug.c
@@ -462,8 +462,10 @@ static ssize_t cifs_stats_proc_write(struct file *file,
server = list_entry(tmp1, struct TCP_Server_Info,
tcp_ses_list);
#ifdef CONFIG_CIFS_STATS2
- for (i = 0; i < NUMBER_OF_SMB2_COMMANDS; i++)
+ for (i = 0; i < NUMBER_OF_SMB2_COMMANDS; i++) {
atomic_set(&server->smb2slowcmd[i], 0);
+ server->time_per_cmd[i] = 0;
+ }
#endif /* CONFIG_CIFS_STATS2 */
list_for_each(tmp2, &server->smb_ses_list) {
ses = list_entry(tmp2, struct cifs_ses,
@@ -531,6 +533,11 @@ static int cifs_stats_proc_show(struct seq_file *m, void *v)
server = list_entry(tmp1, struct TCP_Server_Info,
tcp_ses_list);
#ifdef CONFIG_CIFS_STATS2
+ seq_printf(m, "Total time spent processing each command\n");
+ for (j = 0; j < NUMBER_OF_SMB2_COMMANDS; j++)
+ seq_printf(m, "cmd: %d jiffies: %llu seconds %llu\n", j,
+ server->time_per_cmd[j],
+ server->time_per_cmd[j]/HZ);
for (j = 0; j < NUMBER_OF_SMB2_COMMANDS; j++)
if (atomic_read(&server->smb2slowcmd[j]))
seq_printf(m, "%d slow responses from %s for command %d\n",
diff --git a/fs/cifs/cifsglob.h b/fs/cifs/cifsglob.h
index 38feae812b47..7232788fffc0 100644
--- a/fs/cifs/cifsglob.h
+++ b/fs/cifs/cifsglob.h
@@ -18,7 +18,7 @@
*/
#ifndef _CIFS_GLOB_H
#define _CIFS_GLOB_H
-
+#define CONFIG_CIFS_STATS2 1 /* BB REMOVEME */
#include <linux/in.h>
#include <linux/in6.h>
#include <linux/slab.h>
@@ -708,6 +708,7 @@ struct TCP_Server_Info {
atomic_t in_send; /* requests trying to send */
atomic_t num_waiters; /* blocked waiting to get in sendrecv */
atomic_t smb2slowcmd[NUMBER_OF_SMB2_COMMANDS]; /* count resps > 1 sec */
+ __u64 time_per_cmd[NUMBER_OF_SMB2_COMMANDS]; /* total time per cmd */
#endif /* STATS2 */
unsigned int max_read;
unsigned int max_write;
diff --git a/fs/cifs/transport.c b/fs/cifs/transport.c
index 1de8e996e566..0c7f264403bd 100644
--- a/fs/cifs/transport.c
+++ b/fs/cifs/transport.c
@@ -114,6 +114,9 @@ DeleteMidQEntry(struct mid_q_entry *midEntry)
cifs_small_buf_release(midEntry->resp_buf);
#ifdef CONFIG_CIFS_STATS2
now = jiffies;
+ if (le16_to_cpu(midEntry->command) < NUMBER_OF_SMB2_COMMANDS)
+ midEntry->server->time_per_cmd[le16_to_cpu(midEntry->command)]
+ += (now - midEntry->when_alloc);
/*
* commands taking longer than one second (default) can be indications
* that something is wrong, unless it is quite a slow link or a very
--
2.17.1
