4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2012, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/osd-zfs/osd_lproc.c
34 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
35 * Author: Mike Pershin <tappro@whamcloud.com>
38 #define DEBUG_SUBSYSTEM S_OSD
41 #include <obd_class.h>
42 #include <lprocfs_status.h>
43 #include <lustre/lustre_idl.h>
45 #include "osd_internal.h"
49 #define pct(a, b) (b ? a * 100 / b : 0)
51 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
52 struct obd_histogram *read,
53 struct obd_histogram *write, int scale)
55 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
58 seq_printf(seq, "\n%26s read | write\n", " ");
59 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
62 read_tot = lprocfs_oh_sum(read);
63 write_tot = lprocfs_oh_sum(write);
64 for (i = 0; i < OBD_HIST_MAX; i++) {
65 r = read->oh_buckets[i];
66 w = write->oh_buckets[i];
69 if (read_cum == 0 && write_cum == 0)
73 seq_printf(seq, "%u", i);
75 seq_printf(seq, "%u", scale << i);
77 seq_printf(seq, "%uK", scale << (i-10));
79 seq_printf(seq, "%uM", scale << (i-20));
81 seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
82 r, pct(r, read_tot), pct(read_cum, read_tot),
83 w, pct(w, write_tot), pct(write_cum, write_tot));
85 if (read_cum == read_tot && write_cum == write_tot)
90 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
92 struct timespec64 now;
94 /* this sampling races with updates */
95 ktime_get_real_ts64(&now);
96 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
97 (s64)now.tv_sec, now.tv_nsec);
99 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
100 &brw_stats->hist[BRW_R_PAGES],
101 &brw_stats->hist[BRW_W_PAGES], 1);
102 display_brw_stats(seq, "discontiguous pages", "rpcs",
103 &brw_stats->hist[BRW_R_DISCONT_PAGES],
104 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
106 display_brw_stats(seq, "discontiguous blocks", "rpcs",
107 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
108 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
110 display_brw_stats(seq, "disk fragmented I/Os", "ios",
111 &brw_stats->hist[BRW_R_DIO_FRAGS],
112 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
114 display_brw_stats(seq, "disk I/Os in flight", "ios",
115 &brw_stats->hist[BRW_R_RPC_HIST],
116 &brw_stats->hist[BRW_W_RPC_HIST], 0);
118 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
119 &brw_stats->hist[BRW_R_IO_TIME],
120 &brw_stats->hist[BRW_W_IO_TIME], 1000 / HZ);
122 display_brw_stats(seq, "disk I/O size", "ios",
123 &brw_stats->hist[BRW_R_DISK_IOSIZE],
124 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
129 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
131 struct osd_device *osd = seq->private;
133 brw_stats_show(seq, &osd->od_brw_stats);
138 static ssize_t osd_brw_stats_seq_write(struct file *file,
139 const char __user *buf,
140 size_t len, loff_t *off)
142 struct seq_file *seq = file->private_data;
143 struct osd_device *osd = seq->private;
146 for (i = 0; i < BRW_LAST; i++)
147 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
152 LPROC_SEQ_FOPS(osd_brw_stats);
154 static int osd_stats_init(struct osd_device *osd)
159 for (i = 0; i < BRW_LAST; i++)
160 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
162 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
163 if (osd->od_stats != NULL) {
164 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
169 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
170 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
172 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
173 LPROCFS_CNTR_AVGMINMAX,
174 "get_page_failures", "num");
175 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
176 LPROCFS_CNTR_AVGMINMAX,
177 "cache_access", "pages");
178 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
179 LPROCFS_CNTR_AVGMINMAX,
180 "cache_hit", "pages");
181 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
182 LPROCFS_CNTR_AVGMINMAX,
183 "cache_miss", "pages");
184 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
185 LPROCFS_CNTR_AVGMINMAX,
187 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
188 LPROCFS_CNTR_AVGMINMAX,
189 "zerocopy", "pages");
190 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
191 LPROCFS_CNTR_AVGMINMAX,
193 #ifdef OSD_THANDLE_STATS
194 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
195 LPROCFS_CNTR_AVGMINMAX,
196 "thandle_starting", "usec");
197 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
198 LPROCFS_CNTR_AVGMINMAX,
199 "thandle_open", "usec");
200 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
201 LPROCFS_CNTR_AVGMINMAX,
202 "thandle_closing", "usec");
204 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
205 0644, &osd_brw_stats_fops, osd);
214 static int zfs_osd_fstype_seq_show(struct seq_file *m, void *data)
216 seq_puts(m, "zfs\n");
219 LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
221 static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
223 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
225 LASSERT(osd != NULL);
226 seq_printf(m, "%s\n", osd->od_mntdev);
229 LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
232 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
233 size_t count, loff_t *off)
235 struct seq_file *m = file->private_data;
236 struct dt_device *dt = m->private;
240 rc = lu_env_init(&env, LCT_LOCAL);
243 rc = dt_sync(&env, dt);
246 return rc == 0 ? count : rc;
248 LPROC_SEQ_FOPS_WO_TYPE(zfs, osd_force_sync);
250 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
251 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
252 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
253 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
254 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
255 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
257 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
258 { .name = "blocksize",
259 .fops = &zfs_dt_blksize_fops },
260 { .name = "kbytestotal",
261 .fops = &zfs_dt_kbytestotal_fops },
262 { .name = "kbytesfree",
263 .fops = &zfs_dt_kbytesfree_fops },
264 { .name = "kbytesavail",
265 .fops = &zfs_dt_kbytesavail_fops },
266 { .name = "filestotal",
267 .fops = &zfs_dt_filestotal_fops },
268 { .name = "filesfree",
269 .fops = &zfs_dt_filesfree_fops },
271 .fops = &zfs_osd_fstype_fops },
273 .fops = &zfs_osd_mntdev_fops },
274 { .name = "force_sync",
275 .fops = &zfs_osd_force_sync_fops },
279 int osd_procfs_init(struct osd_device *osd, const char *name)
281 struct obd_type *type;
285 if (osd->od_proc_entry)
288 /* at the moment there is no linkage between lu_type
289 * and obd_type, so we lookup obd_type this way */
290 type = class_search_type(LUSTRE_OSD_ZFS_NAME);
292 LASSERT(name != NULL);
293 LASSERT(type != NULL);
295 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
296 lprocfs_osd_obd_vars,
298 if (IS_ERR(osd->od_proc_entry)) {
299 rc = PTR_ERR(osd->od_proc_entry);
300 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
301 osd->od_proc_entry = NULL;
305 rc = osd_stats_init(osd);
310 osd_procfs_fini(osd);
314 int osd_procfs_fini(struct osd_device *osd)
319 lprocfs_free_stats(&osd->od_stats);
321 if (osd->od_proc_entry) {
322 lprocfs_remove(&osd->od_proc_entry);
323 osd->od_proc_entry = NULL;