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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd-zfs/osd_lproc.c
38 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
39 * Author: Mike Pershin <tappro@whamcloud.com>
42 #define DEBUG_SUBSYSTEM S_OSD
45 #include <obd_class.h>
46 #include <lprocfs_status.h>
47 #include <lustre/lustre_idl.h>
49 #include "osd_internal.h"
53 #define pct(a, b) (b ? a * 100 / b : 0)
55 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
56 struct obd_histogram *read,
57 struct obd_histogram *write, int scale)
59 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
62 seq_printf(seq, "\n%26s read | write\n", " ");
63 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
66 read_tot = lprocfs_oh_sum(read);
67 write_tot = lprocfs_oh_sum(write);
68 for (i = 0; i < OBD_HIST_MAX; i++) {
69 r = read->oh_buckets[i];
70 w = write->oh_buckets[i];
73 if (read_cum == 0 && write_cum == 0)
77 seq_printf(seq, "%u", i);
79 seq_printf(seq, "%u", scale << i);
81 seq_printf(seq, "%uK", scale << (i-10));
83 seq_printf(seq, "%uM", scale << (i-20));
85 seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
86 r, pct(r, read_tot), pct(read_cum, read_tot),
87 w, pct(w, write_tot), pct(write_cum, write_tot));
89 if (read_cum == read_tot && write_cum == write_tot)
94 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
98 /* this sampling races with updates */
99 do_gettimeofday(&now);
100 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
101 now.tv_sec, now.tv_usec);
103 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
104 &brw_stats->hist[BRW_R_PAGES],
105 &brw_stats->hist[BRW_W_PAGES], 1);
106 display_brw_stats(seq, "discontiguous pages", "rpcs",
107 &brw_stats->hist[BRW_R_DISCONT_PAGES],
108 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
110 display_brw_stats(seq, "discontiguous blocks", "rpcs",
111 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
112 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
114 display_brw_stats(seq, "disk fragmented I/Os", "ios",
115 &brw_stats->hist[BRW_R_DIO_FRAGS],
116 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
118 display_brw_stats(seq, "disk I/Os in flight", "ios",
119 &brw_stats->hist[BRW_R_RPC_HIST],
120 &brw_stats->hist[BRW_W_RPC_HIST], 0);
122 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
123 &brw_stats->hist[BRW_R_IO_TIME],
124 &brw_stats->hist[BRW_W_IO_TIME], 1000 / HZ);
126 display_brw_stats(seq, "disk I/O size", "ios",
127 &brw_stats->hist[BRW_R_DISK_IOSIZE],
128 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
133 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
135 struct osd_device *osd = seq->private;
137 brw_stats_show(seq, &osd->od_brw_stats);
142 static ssize_t osd_brw_stats_seq_write(struct file *file,
143 const char __user *buf,
144 size_t len, loff_t *off)
146 struct seq_file *seq = file->private_data;
147 struct osd_device *osd = seq->private;
150 for (i = 0; i < BRW_LAST; i++)
151 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
156 LPROC_SEQ_FOPS(osd_brw_stats);
158 static int osd_stats_init(struct osd_device *osd)
163 for (i = 0; i < BRW_LAST; i++)
164 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
166 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
167 if (osd->od_stats != NULL) {
168 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
173 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
174 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
176 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
177 LPROCFS_CNTR_AVGMINMAX,
178 "get_page_failures", "num");
179 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
180 LPROCFS_CNTR_AVGMINMAX,
181 "cache_access", "pages");
182 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
183 LPROCFS_CNTR_AVGMINMAX,
184 "cache_hit", "pages");
185 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
186 LPROCFS_CNTR_AVGMINMAX,
187 "cache_miss", "pages");
188 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
189 LPROCFS_CNTR_AVGMINMAX,
191 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
192 LPROCFS_CNTR_AVGMINMAX,
193 "zerocopy", "pages");
194 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
195 LPROCFS_CNTR_AVGMINMAX,
197 #ifdef OSD_THANDLE_STATS
198 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
199 LPROCFS_CNTR_AVGMINMAX,
200 "thandle_starting", "usec");
201 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
202 LPROCFS_CNTR_AVGMINMAX,
203 "thandle_open", "usec");
204 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
205 LPROCFS_CNTR_AVGMINMAX,
206 "thandle_closing", "usec");
208 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
209 0644, &osd_brw_stats_fops, osd);
218 static int zfs_osd_fstype_seq_show(struct seq_file *m, void *data)
220 return seq_printf(m, "zfs\n");
222 LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
224 static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
226 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
228 LASSERT(osd != NULL);
229 return seq_printf(m, "%s\n", osd->od_mntdev);
231 LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
234 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
235 size_t count, loff_t *off)
237 struct seq_file *m = file->private_data;
238 struct dt_device *dt = m->private;
242 rc = lu_env_init(&env, LCT_LOCAL);
245 rc = dt_sync(&env, dt);
248 return rc == 0 ? count : rc;
250 LPROC_SEQ_FOPS_WO_TYPE(zfs, osd_force_sync);
252 static int zfs_osd_iused_est_seq_show(struct seq_file *m, void *data)
254 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
255 LASSERT(osd != NULL);
257 return seq_printf(m, "%d\n", osd->od_quota_iused_est);
261 zfs_osd_iused_est_seq_write(struct file *file, const char __user *buffer,
262 size_t count, loff_t *off)
264 struct seq_file *m = file->private_data;
265 struct dt_device *dt = m->private;
266 struct osd_device *osd = osd_dt_dev(dt);
269 LASSERT(osd != NULL);
271 rc = lprocfs_write_helper(buffer, count, &val);
275 osd->od_quota_iused_est = !!val;
279 LPROC_SEQ_FOPS(zfs_osd_iused_est);
281 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
282 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
283 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
284 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
285 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
286 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
288 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
289 { .name = "blocksize",
290 .fops = &zfs_dt_blksize_fops },
291 { .name = "kbytestotal",
292 .fops = &zfs_dt_kbytestotal_fops },
293 { .name = "kbytesfree",
294 .fops = &zfs_dt_kbytesfree_fops },
295 { .name = "kbytesavail",
296 .fops = &zfs_dt_kbytesavail_fops },
297 { .name = "filestotal",
298 .fops = &zfs_dt_filestotal_fops },
299 { .name = "filesfree",
300 .fops = &zfs_dt_filesfree_fops },
302 .fops = &zfs_osd_fstype_fops },
304 .fops = &zfs_osd_mntdev_fops },
305 { .name = "force_sync",
306 .fops = &zfs_osd_force_sync_fops },
307 { .name = "quota_iused_estimate",
308 .fops = &zfs_osd_iused_est_fops },
312 int osd_procfs_init(struct osd_device *osd, const char *name)
314 struct obd_type *type;
318 if (osd->od_proc_entry)
321 /* at the moment there is no linkage between lu_type
322 * and obd_type, so we lookup obd_type this way */
323 type = class_search_type(LUSTRE_OSD_ZFS_NAME);
325 LASSERT(name != NULL);
326 LASSERT(type != NULL);
328 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
329 lprocfs_osd_obd_vars,
331 if (IS_ERR(osd->od_proc_entry)) {
332 rc = PTR_ERR(osd->od_proc_entry);
333 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
334 osd->od_proc_entry = NULL;
338 rc = osd_stats_init(osd);
343 osd_procfs_fini(osd);
347 int osd_procfs_fini(struct osd_device *osd)
352 lprocfs_free_stats(&osd->od_stats);
354 if (osd->od_proc_entry) {
355 lprocfs_remove(&osd->od_proc_entry);
356 osd->od_proc_entry = NULL;