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/
31 * lustre/osd-zfs/osd_lproc.c
33 * Author: Alex Zhuravlev <bzzz@whamcloud.com>
34 * Author: Mike Pershin <tappro@whamcloud.com>
37 #define DEBUG_SUBSYSTEM S_OSD
40 #include <obd_class.h>
41 #include <lprocfs_status.h>
42 #include <lustre_scrub.h>
44 #include "osd_internal.h"
48 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
49 struct obd_histogram *read,
50 struct obd_histogram *write, int scale)
52 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
55 seq_printf(seq, "\n%26s read | write\n", " ");
56 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
59 read_tot = lprocfs_oh_sum(read);
60 write_tot = lprocfs_oh_sum(write);
61 for (i = 0; i < OBD_HIST_MAX; i++) {
62 r = read->oh_buckets[i];
63 w = write->oh_buckets[i];
66 if (read_cum == 0 && write_cum == 0)
70 seq_printf(seq, "%u", i);
72 seq_printf(seq, "%u", scale << i);
74 seq_printf(seq, "%uK", scale << (i-10));
76 seq_printf(seq, "%uM", scale << (i-20));
78 seq_printf(seq, ":\t\t%10lu %3u %3u | %4lu %3u %3u\n",
79 r, pct(r, read_tot), pct(read_cum, read_tot),
80 w, pct(w, write_tot), pct(write_cum, write_tot));
82 if (read_cum == read_tot && write_cum == write_tot)
87 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
89 struct timespec64 now;
91 /* this sampling races with updates */
92 ktime_get_real_ts64(&now);
93 seq_printf(seq, "snapshot_time: %llu.%09lu (secs.nsecs)\n",
94 (s64)now.tv_sec, now.tv_nsec);
96 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
97 &brw_stats->hist[BRW_R_PAGES],
98 &brw_stats->hist[BRW_W_PAGES], 1);
99 display_brw_stats(seq, "discontiguous pages", "rpcs",
100 &brw_stats->hist[BRW_R_DISCONT_PAGES],
101 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
103 display_brw_stats(seq, "discontiguous blocks", "rpcs",
104 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
105 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
107 display_brw_stats(seq, "disk fragmented I/Os", "ios",
108 &brw_stats->hist[BRW_R_DIO_FRAGS],
109 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
111 display_brw_stats(seq, "disk I/Os in flight", "ios",
112 &brw_stats->hist[BRW_R_RPC_HIST],
113 &brw_stats->hist[BRW_W_RPC_HIST], 0);
115 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
116 &brw_stats->hist[BRW_R_IO_TIME],
117 &brw_stats->hist[BRW_W_IO_TIME], 1);
119 display_brw_stats(seq, "disk I/O size", "ios",
120 &brw_stats->hist[BRW_R_DISK_IOSIZE],
121 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
124 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
126 struct osd_device *osd = seq->private;
128 brw_stats_show(seq, &osd->od_brw_stats);
133 static ssize_t osd_brw_stats_seq_write(struct file *file,
134 const char __user *buf,
135 size_t len, loff_t *off)
137 struct seq_file *seq = file->private_data;
138 struct osd_device *osd = seq->private;
141 for (i = 0; i < BRW_LAST; i++)
142 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
147 LPROC_SEQ_FOPS(osd_brw_stats);
149 static int osd_stats_init(struct osd_device *osd)
154 for (i = 0; i < BRW_LAST; i++)
155 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
157 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
158 if (osd->od_stats != NULL) {
159 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
164 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
165 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
167 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
168 LPROCFS_CNTR_AVGMINMAX,
169 "get_page_failures", "num");
170 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
171 LPROCFS_CNTR_AVGMINMAX,
172 "cache_access", "pages");
173 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
174 LPROCFS_CNTR_AVGMINMAX,
175 "cache_hit", "pages");
176 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
177 LPROCFS_CNTR_AVGMINMAX,
178 "cache_miss", "pages");
179 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
180 LPROCFS_CNTR_AVGMINMAX,
182 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
183 LPROCFS_CNTR_AVGMINMAX,
184 "zerocopy", "pages");
185 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
186 LPROCFS_CNTR_AVGMINMAX,
188 #ifdef OSD_THANDLE_STATS
189 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
190 LPROCFS_CNTR_AVGMINMAX,
191 "thandle_starting", "usec");
192 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
193 LPROCFS_CNTR_AVGMINMAX,
194 "thandle_open", "usec");
195 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
196 LPROCFS_CNTR_AVGMINMAX,
197 "thandle_closing", "usec");
199 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
200 0644, &osd_brw_stats_fops, osd);
209 static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
211 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
213 LASSERT(dev != NULL);
217 scrub_dump(m, &dev->od_scrub);
220 LDEBUGFS_SEQ_FOPS_RO(zfs_osd_oi_scrub);
222 static ssize_t auto_scrub_show(struct kobject *kobj, struct attribute *attr,
225 struct dt_device *dt = container_of(kobj, struct dt_device,
227 struct osd_device *dev = osd_dt_dev(dt);
233 return sprintf(buf, "%lld\n", dev->od_auto_scrub_interval);
236 static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
237 const char *buffer, size_t count)
239 struct dt_device *dt = container_of(kobj, struct dt_device,
241 struct osd_device *dev = osd_dt_dev(dt);
249 rc = kstrtoull(buffer, 0, &val);
253 dev->od_auto_scrub_interval = val;
256 LUSTRE_RW_ATTR(auto_scrub);
258 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
261 return sprintf(buf, "zfs\n");
263 LUSTRE_RO_ATTR(fstype);
265 static ssize_t mntdev_show(struct kobject *kobj, struct attribute *attr,
268 struct dt_device *dt = container_of(kobj, struct dt_device,
270 struct osd_device *osd = osd_dt_dev(dt);
274 return sprintf(buf, "%s\n", osd->od_mntdev);
276 LUSTRE_RO_ATTR(mntdev);
278 ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
279 const char *buffer, size_t count)
281 struct dt_device *dt = container_of(kobj, struct dt_device,
286 rc = lu_env_init(&env, LCT_LOCAL);
290 rc = dt_sync(&env, dt);
293 return rc == 0 ? count : rc;
295 LUSTRE_WO_ATTR(force_sync);
297 static ssize_t nonrotational_show(struct kobject *kobj, struct attribute *attr,
300 struct dt_device *dt = container_of(kobj, struct dt_device,
302 struct osd_device *osd = osd_dt_dev(dt);
308 return sprintf(buf, "%u\n", osd->od_nonrotational);
311 static ssize_t nonrotational_store(struct kobject *kobj,
312 struct attribute *attr, const char *buffer,
315 struct dt_device *dt = container_of(kobj, struct dt_device,
317 struct osd_device *osd = osd_dt_dev(dt);
325 rc = kstrtobool(buffer, &val);
329 osd->od_nonrotational = val;
332 LUSTRE_RW_ATTR(nonrotational);
334 static ssize_t index_backup_show(struct kobject *kobj, struct attribute *attr,
337 struct dt_device *dt = container_of(kobj, struct dt_device,
339 struct osd_device *dev = osd_dt_dev(dt);
345 return sprintf(buf, "%d\n", dev->od_index_backup_policy);
348 ssize_t index_backup_store(struct kobject *kobj, struct attribute *attr,
349 const char *buffer, size_t count)
351 struct dt_device *dt = container_of(kobj, struct dt_device,
353 struct osd_device *dev = osd_dt_dev(dt);
361 rc = kstrtoint(buffer, 0, &val);
365 dev->od_index_backup_policy = val;
368 LUSTRE_RW_ATTR(index_backup);
370 static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
372 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
374 LASSERT(osd != NULL);
375 if (unlikely(osd->od_os == NULL))
378 seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
383 zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
384 size_t count, loff_t *off)
386 struct seq_file *m = file->private_data;
387 struct dt_device *dt = m->private;
388 struct osd_device *osd = osd_dt_dev(dt);
389 char kernbuf[22] = "";
393 LASSERT(osd != NULL);
394 if (unlikely(osd->od_os == NULL))
397 if (count >= sizeof(kernbuf))
400 if (copy_from_user(kernbuf, buffer, count))
404 rc = sysfs_memparse(kernbuf, count, &val, "B");
408 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
409 OSD_MAX_CACHE_SIZE : val;
412 LDEBUGFS_SEQ_FOPS(zfs_osd_readcache);
414 static struct attribute *zfs_attrs[] = {
415 &lustre_attr_fstype.attr,
416 &lustre_attr_mntdev.attr,
417 &lustre_attr_force_sync.attr,
418 &lustre_attr_nonrotational.attr,
419 &lustre_attr_index_backup.attr,
420 &lustre_attr_auto_scrub.attr,
424 struct ldebugfs_vars ldebugfs_osd_obd_vars[] = {
425 { .name = "oi_scrub",
426 .fops = &zfs_osd_oi_scrub_fops },
427 { .name = "readcache_max_filesize",
428 .fops = &zfs_osd_readcache_fops },
432 int osd_procfs_init(struct osd_device *osd, const char *name)
434 struct obd_type *type;
439 /* at the moment there is no linkage between lu_type
440 * and obd_type, so we lookup obd_type this way
442 type = class_search_type(LUSTRE_OSD_ZFS_NAME);
447 /* put reference taken by class_search_type */
448 kobject_put(&type->typ_kobj);
450 osd->od_dt_dev.dd_ktype.default_attrs = zfs_attrs;
451 rc = dt_tunables_init(&osd->od_dt_dev, type, name,
452 ldebugfs_osd_obd_vars);
454 CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
459 if (osd->od_proc_entry)
462 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
463 NULL, &osd->od_dt_dev);
464 if (IS_ERR(osd->od_proc_entry)) {
465 rc = PTR_ERR(osd->od_proc_entry);
466 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
467 osd->od_proc_entry = NULL;
471 rc = osd_stats_init(osd);
476 osd_procfs_fini(osd);
480 int osd_procfs_fini(struct osd_device *osd)
485 lprocfs_free_stats(&osd->od_stats);
487 if (osd->od_proc_entry) {
488 lprocfs_remove(&osd->od_proc_entry);
489 osd->od_proc_entry = NULL;
492 return dt_tunables_fini(&osd->od_dt_dev);