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, 2017, 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_scrub.h>
45 #include "osd_internal.h"
49 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
50 struct obd_histogram *read,
51 struct obd_histogram *write, int scale)
53 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
56 seq_printf(seq, "\n%26s read | write\n", " ");
57 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
60 read_tot = lprocfs_oh_sum(read);
61 write_tot = lprocfs_oh_sum(write);
62 for (i = 0; i < OBD_HIST_MAX; i++) {
63 r = read->oh_buckets[i];
64 w = write->oh_buckets[i];
67 if (read_cum == 0 && write_cum == 0)
71 seq_printf(seq, "%u", i);
73 seq_printf(seq, "%u", scale << i);
75 seq_printf(seq, "%uK", scale << (i-10));
77 seq_printf(seq, "%uM", scale << (i-20));
79 seq_printf(seq, ":\t\t%10lu %3u %3u | %4lu %3u %3u\n",
80 r, pct(r, read_tot), pct(read_cum, read_tot),
81 w, pct(w, write_tot), pct(write_cum, write_tot));
83 if (read_cum == read_tot && write_cum == write_tot)
88 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
90 struct timespec64 now;
92 /* this sampling races with updates */
93 ktime_get_real_ts64(&now);
94 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
95 (s64)now.tv_sec, now.tv_nsec);
97 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
98 &brw_stats->hist[BRW_R_PAGES],
99 &brw_stats->hist[BRW_W_PAGES], 1);
100 display_brw_stats(seq, "discontiguous pages", "rpcs",
101 &brw_stats->hist[BRW_R_DISCONT_PAGES],
102 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
104 display_brw_stats(seq, "discontiguous blocks", "rpcs",
105 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
106 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
108 display_brw_stats(seq, "disk fragmented I/Os", "ios",
109 &brw_stats->hist[BRW_R_DIO_FRAGS],
110 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
112 display_brw_stats(seq, "disk I/Os in flight", "ios",
113 &brw_stats->hist[BRW_R_RPC_HIST],
114 &brw_stats->hist[BRW_W_RPC_HIST], 0);
116 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
117 &brw_stats->hist[BRW_R_IO_TIME],
118 &brw_stats->hist[BRW_W_IO_TIME], 1);
120 display_brw_stats(seq, "disk I/O size", "ios",
121 &brw_stats->hist[BRW_R_DISK_IOSIZE],
122 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
125 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
127 struct osd_device *osd = seq->private;
129 brw_stats_show(seq, &osd->od_brw_stats);
134 static ssize_t osd_brw_stats_seq_write(struct file *file,
135 const char __user *buf,
136 size_t len, loff_t *off)
138 struct seq_file *seq = file->private_data;
139 struct osd_device *osd = seq->private;
142 for (i = 0; i < BRW_LAST; i++)
143 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
148 LPROC_SEQ_FOPS(osd_brw_stats);
150 static int osd_stats_init(struct osd_device *osd)
155 for (i = 0; i < BRW_LAST; i++)
156 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
158 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
159 if (osd->od_stats != NULL) {
160 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
165 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
166 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
168 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
169 LPROCFS_CNTR_AVGMINMAX,
170 "get_page_failures", "num");
171 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
172 LPROCFS_CNTR_AVGMINMAX,
173 "cache_access", "pages");
174 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
175 LPROCFS_CNTR_AVGMINMAX,
176 "cache_hit", "pages");
177 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
178 LPROCFS_CNTR_AVGMINMAX,
179 "cache_miss", "pages");
180 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
181 LPROCFS_CNTR_AVGMINMAX,
183 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
184 LPROCFS_CNTR_AVGMINMAX,
185 "zerocopy", "pages");
186 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
187 LPROCFS_CNTR_AVGMINMAX,
189 #ifdef OSD_THANDLE_STATS
190 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
191 LPROCFS_CNTR_AVGMINMAX,
192 "thandle_starting", "usec");
193 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
194 LPROCFS_CNTR_AVGMINMAX,
195 "thandle_open", "usec");
196 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
197 LPROCFS_CNTR_AVGMINMAX,
198 "thandle_closing", "usec");
200 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
201 0644, &osd_brw_stats_fops, osd);
210 static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
212 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
214 LASSERT(dev != NULL);
218 scrub_dump(m, &dev->od_scrub);
221 LDEBUGFS_SEQ_FOPS_RO(zfs_osd_oi_scrub);
223 static ssize_t auto_scrub_show(struct kobject *kobj, struct attribute *attr,
226 struct dt_device *dt = container_of(kobj, struct dt_device,
228 struct osd_device *dev = osd_dt_dev(dt);
234 return sprintf(buf, "%lld\n", dev->od_auto_scrub_interval);
237 static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
238 const char *buffer, size_t count)
240 struct dt_device *dt = container_of(kobj, struct dt_device,
242 struct osd_device *dev = osd_dt_dev(dt);
250 rc = kstrtoull(buffer, 0, &val);
254 dev->od_auto_scrub_interval = val;
257 LUSTRE_RW_ATTR(auto_scrub);
259 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
262 return sprintf(buf, "zfs\n");
264 LUSTRE_RO_ATTR(fstype);
266 static ssize_t mntdev_show(struct kobject *kobj, struct attribute *attr,
269 struct dt_device *dt = container_of(kobj, struct dt_device,
271 struct osd_device *osd = osd_dt_dev(dt);
275 return sprintf(buf, "%s\n", osd->od_mntdev);
277 LUSTRE_RO_ATTR(mntdev);
279 ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
280 const char *buffer, size_t count)
282 struct dt_device *dt = container_of(kobj, struct dt_device,
287 rc = lu_env_init(&env, LCT_LOCAL);
291 rc = dt_sync(&env, dt);
294 return rc == 0 ? count : rc;
296 LUSTRE_WO_ATTR(force_sync);
298 static ssize_t nonrotational_show(struct kobject *kobj, struct attribute *attr,
301 struct dt_device *dt = container_of(kobj, struct dt_device,
303 struct osd_device *osd = osd_dt_dev(dt);
309 return sprintf(buf, "%u\n", osd->od_nonrotational);
312 static ssize_t nonrotational_store(struct kobject *kobj,
313 struct attribute *attr, const char *buffer,
316 struct dt_device *dt = container_of(kobj, struct dt_device,
318 struct osd_device *osd = osd_dt_dev(dt);
326 rc = kstrtobool(buffer, &val);
330 osd->od_nonrotational = val;
333 LUSTRE_RW_ATTR(nonrotational);
335 static ssize_t index_backup_show(struct kobject *kobj, struct attribute *attr,
338 struct dt_device *dt = container_of(kobj, struct dt_device,
340 struct osd_device *dev = osd_dt_dev(dt);
346 return sprintf(buf, "%d\n", dev->od_index_backup_policy);
349 ssize_t index_backup_store(struct kobject *kobj, struct attribute *attr,
350 const char *buffer, size_t count)
352 struct dt_device *dt = container_of(kobj, struct dt_device,
354 struct osd_device *dev = osd_dt_dev(dt);
362 rc = kstrtoint(buffer, 0, &val);
366 dev->od_index_backup_policy = val;
369 LUSTRE_RW_ATTR(index_backup);
371 static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
373 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
375 LASSERT(osd != NULL);
376 if (unlikely(osd->od_os == NULL))
379 seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
384 zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
385 size_t count, loff_t *off)
387 struct seq_file *m = file->private_data;
388 struct dt_device *dt = m->private;
389 struct osd_device *osd = osd_dt_dev(dt);
390 char kernbuf[22] = "";
394 LASSERT(osd != NULL);
395 if (unlikely(osd->od_os == NULL))
398 if (count >= sizeof(kernbuf))
401 if (copy_from_user(kernbuf, buffer, count))
405 rc = sysfs_memparse(kernbuf, count, &val, "B");
409 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
410 OSD_MAX_CACHE_SIZE : val;
413 LDEBUGFS_SEQ_FOPS(zfs_osd_readcache);
415 static struct attribute *zfs_attrs[] = {
416 &lustre_attr_fstype.attr,
417 &lustre_attr_mntdev.attr,
418 &lustre_attr_force_sync.attr,
419 &lustre_attr_nonrotational.attr,
420 &lustre_attr_index_backup.attr,
421 &lustre_attr_auto_scrub.attr,
425 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
426 { .name = "oi_scrub",
427 .fops = &zfs_osd_oi_scrub_fops },
428 { .name = "readcache_max_filesize",
429 .fops = &zfs_osd_readcache_fops },
433 int osd_procfs_init(struct osd_device *osd, const char *name)
435 struct obd_type *type;
440 /* at the moment there is no linkage between lu_type
441 * and obd_type, so we lookup obd_type this way
443 type = class_search_type(LUSTRE_OSD_ZFS_NAME);
448 /* put reference taken by class_search_type */
449 kobject_put(&type->typ_kobj);
451 osd->od_dt_dev.dd_ktype.default_attrs = zfs_attrs;
452 rc = dt_tunables_init(&osd->od_dt_dev, type, name,
453 lprocfs_osd_obd_vars);
455 CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
460 if (osd->od_proc_entry)
463 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
464 NULL, &osd->od_dt_dev);
465 if (IS_ERR(osd->od_proc_entry)) {
466 rc = PTR_ERR(osd->od_proc_entry);
467 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
468 osd->od_proc_entry = NULL;
472 rc = osd_stats_init(osd);
477 osd_procfs_fini(osd);
481 int osd_procfs_fini(struct osd_device *osd)
486 lprocfs_free_stats(&osd->od_stats);
488 if (osd->od_proc_entry) {
489 lprocfs_remove(&osd->od_proc_entry);
490 osd->od_proc_entry = NULL;
493 return dt_tunables_fini(&osd->od_dt_dev);