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) 2011, 2015, 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/osd_lproc.c
34 * Author: Mikhail Pershin <tappro@sun.com>
37 #define DEBUG_SUBSYSTEM S_OSD
39 #include <lprocfs_status.h>
41 #include "osd_internal.h"
45 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
47 struct brw_stats *s = &osd->od_brw_stats;
48 sector_t *last_block = NULL;
49 struct page **pages = iobuf->dr_pages;
50 struct page *last_page = NULL;
51 unsigned long discont_pages = 0;
52 unsigned long discont_blocks = 0;
53 sector_t *blocks = iobuf->dr_blocks;
54 int i, nr_pages = iobuf->dr_npages;
56 int rw = iobuf->dr_rw;
58 if (unlikely(nr_pages == 0))
61 blocks_per_page = PAGE_SIZE >> osd_sb(osd)->s_blocksize_bits;
63 lprocfs_oh_tally_log2(&s->hist[BRW_R_PAGES+rw], nr_pages);
65 while (nr_pages-- > 0) {
66 if (last_page && (*pages)->index != (last_page->index + 1))
70 for (i = 0; i < blocks_per_page; i++) {
71 if (last_block && *blocks != (*last_block + 1))
73 last_block = blocks++;
77 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_PAGES+rw], discont_pages);
78 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
81 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
82 struct obd_histogram *read, struct obd_histogram *write, int scale)
84 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
87 seq_printf(seq, "\n%26s read | write\n", " ");
88 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
91 read_tot = lprocfs_oh_sum(read);
92 write_tot = lprocfs_oh_sum(write);
93 for (i = 0; i < OBD_HIST_MAX; i++) {
94 r = read->oh_buckets[i];
95 w = write->oh_buckets[i];
98 if (read_cum == 0 && write_cum == 0)
102 seq_printf(seq, "%u", i);
104 seq_printf(seq, "%u", scale << i);
106 seq_printf(seq, "%uK", scale << (i-10));
108 seq_printf(seq, "%uM", scale << (i-20));
110 seq_printf(seq, ":\t\t%10lu %3u %3u | %4lu %3u %3u\n",
111 r, pct(r, read_tot), pct(read_cum, read_tot),
112 w, pct(w, write_tot), pct(write_cum, write_tot));
114 if (read_cum == read_tot && write_cum == write_tot)
119 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
121 struct timespec64 now;
123 /* this sampling races with updates */
124 ktime_get_real_ts64(&now);
126 seq_printf(seq, "snapshot_time: %lld.%09ld (secs.nsecs)\n",
127 (s64)now.tv_sec, now.tv_nsec);
129 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
130 &brw_stats->hist[BRW_R_PAGES],
131 &brw_stats->hist[BRW_W_PAGES], 1);
133 display_brw_stats(seq, "discontiguous pages", "rpcs",
134 &brw_stats->hist[BRW_R_DISCONT_PAGES],
135 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
137 display_brw_stats(seq, "discontiguous blocks", "rpcs",
138 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
139 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
141 display_brw_stats(seq, "disk fragmented I/Os", "ios",
142 &brw_stats->hist[BRW_R_DIO_FRAGS],
143 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
145 display_brw_stats(seq, "disk I/Os in flight", "ios",
146 &brw_stats->hist[BRW_R_RPC_HIST],
147 &brw_stats->hist[BRW_W_RPC_HIST], 0);
149 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
150 &brw_stats->hist[BRW_R_IO_TIME],
151 &brw_stats->hist[BRW_W_IO_TIME],
152 jiffies_to_msecs(1000) / MSEC_PER_SEC);
154 display_brw_stats(seq, "disk I/O size", "ios",
155 &brw_stats->hist[BRW_R_DISK_IOSIZE],
156 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
159 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
161 struct osd_device *osd = seq->private;
163 brw_stats_show(seq, &osd->od_brw_stats);
168 static ssize_t osd_brw_stats_seq_write(struct file *file,
169 const char __user *buf,
170 size_t len, loff_t *off)
172 struct seq_file *seq = file->private_data;
173 struct osd_device *osd = seq->private;
176 for (i = 0; i < BRW_LAST; i++)
177 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
182 LPROC_SEQ_FOPS(osd_brw_stats);
184 static int osd_stats_init(struct osd_device *osd)
189 for (i = 0; i < BRW_LAST; i++)
190 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
192 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
193 if (osd->od_stats != NULL) {
194 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
199 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
200 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
202 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
203 LPROCFS_CNTR_AVGMINMAX,
204 "get_page_failures", "num");
205 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
206 LPROCFS_CNTR_AVGMINMAX,
207 "cache_access", "pages");
208 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
209 LPROCFS_CNTR_AVGMINMAX,
210 "cache_hit", "pages");
211 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
212 LPROCFS_CNTR_AVGMINMAX,
213 "cache_miss", "pages");
214 #if OSD_THANDLE_STATS
215 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
216 LPROCFS_CNTR_AVGMINMAX,
217 "thandle starting", "usec");
218 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
219 LPROCFS_CNTR_AVGMINMAX,
220 "thandle open", "usec");
221 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
222 LPROCFS_CNTR_AVGMINMAX,
223 "thandle closing", "usec");
225 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
226 0644, &osd_brw_stats_fops, osd);
234 static int ldiskfs_osd_fstype_seq_show(struct seq_file *m, void *data)
236 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
238 LASSERT(osd != NULL);
239 seq_puts(m, "ldiskfs\n");
242 LPROC_SEQ_FOPS_RO(ldiskfs_osd_fstype);
244 static int ldiskfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
246 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
248 LASSERT(osd != NULL);
249 if (unlikely(osd->od_mnt == NULL))
252 seq_printf(m, "%s\n", osd->od_mntdev);
255 LPROC_SEQ_FOPS_RO(ldiskfs_osd_mntdev);
257 static int ldiskfs_osd_cache_seq_show(struct seq_file *m, void *data)
259 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
261 LASSERT(osd != NULL);
262 if (unlikely(osd->od_mnt == NULL))
265 seq_printf(m, "%u\n", osd->od_read_cache);
270 ldiskfs_osd_cache_seq_write(struct file *file, const char __user *buffer,
271 size_t count, loff_t *off)
273 struct seq_file *m = file->private_data;
274 struct dt_device *dt = m->private;
275 struct osd_device *osd = osd_dt_dev(dt);
279 LASSERT(osd != NULL);
280 if (unlikely(osd->od_mnt == NULL))
283 rc = kstrtobool_from_user(buffer, count, &val);
287 osd->od_read_cache = val;
290 LPROC_SEQ_FOPS(ldiskfs_osd_cache);
292 static int ldiskfs_osd_wcache_seq_show(struct seq_file *m, void *data)
294 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
296 LASSERT(osd != NULL);
297 if (unlikely(osd->od_mnt == NULL))
300 seq_printf(m, "%u\n", osd->od_writethrough_cache);
305 ldiskfs_osd_wcache_seq_write(struct file *file, const char __user *buffer,
306 size_t count, loff_t *off)
308 struct seq_file *m = file->private_data;
309 struct dt_device *dt = m->private;
310 struct osd_device *osd = osd_dt_dev(dt);
314 LASSERT(osd != NULL);
315 if (unlikely(osd->od_mnt == NULL))
318 rc = kstrtobool_from_user(buffer, count, &val);
322 osd->od_writethrough_cache = val;
325 LPROC_SEQ_FOPS(ldiskfs_osd_wcache);
328 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
329 size_t count, loff_t *off)
331 struct seq_file *m = file->private_data;
332 struct dt_device *dt = m->private;
333 struct osd_device *osd = osd_dt_dev(dt);
337 LASSERT(osd != NULL);
338 if (unlikely(osd->od_mnt == NULL))
341 rc = lu_env_init(&env, LCT_LOCAL);
344 rc = dt_sync(&env, dt);
347 return rc == 0 ? count : rc;
349 LPROC_SEQ_FOPS_WR_ONLY(ldiskfs, osd_force_sync);
351 static int ldiskfs_osd_pdo_seq_show(struct seq_file *m, void *data)
353 seq_printf(m, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
358 ldiskfs_osd_pdo_seq_write(struct file *file, const char __user *buffer,
359 size_t count, loff_t *off)
364 rc = kstrtobool_from_user(buffer, count, &pdo);
372 LPROC_SEQ_FOPS(ldiskfs_osd_pdo);
374 static int ldiskfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
376 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
378 LASSERT(dev != NULL);
379 if (unlikely(dev->od_mnt == NULL))
382 seq_printf(m, "%lld\n", dev->od_auto_scrub_interval);
387 ldiskfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
388 size_t count, loff_t *off)
390 struct seq_file *m = file->private_data;
391 struct dt_device *dt = m->private;
392 struct osd_device *dev = osd_dt_dev(dt);
396 LASSERT(dev != NULL);
397 if (unlikely(dev->od_mnt == NULL))
400 rc = kstrtoll_from_user(buffer, count, 0, &val);
404 dev->od_auto_scrub_interval = val;
407 LPROC_SEQ_FOPS(ldiskfs_osd_auto_scrub);
409 static int ldiskfs_osd_full_scrub_ratio_seq_show(struct seq_file *m, void *data)
411 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
413 LASSERT(dev != NULL);
414 if (unlikely(dev->od_mnt == NULL))
417 seq_printf(m, "%llu\n", dev->od_full_scrub_ratio);
422 ldiskfs_osd_full_scrub_ratio_seq_write(struct file *file,
423 const char __user *buffer,
424 size_t count, loff_t *off)
426 struct seq_file *m = file->private_data;
427 struct dt_device *dt = m->private;
428 struct osd_device *dev = osd_dt_dev(dt);
432 LASSERT(dev != NULL);
433 if (unlikely(dev->od_mnt == NULL))
436 rc = kstrtoll_from_user(buffer, count, 0, &val);
443 dev->od_full_scrub_ratio = val;
446 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_ratio);
448 static int ldiskfs_osd_full_scrub_threshold_rate_seq_show(struct seq_file *m,
451 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
453 LASSERT(dev != NULL);
454 if (unlikely(dev->od_mnt == NULL))
457 seq_printf(m, "%llu (bad OI mappings/minute)\n",
458 dev->od_full_scrub_threshold_rate);
463 ldiskfs_osd_full_scrub_threshold_rate_seq_write(struct file *file,
464 const char __user *buffer,
465 size_t count, loff_t *off)
467 struct seq_file *m = file->private_data;
468 struct dt_device *dt = m->private;
469 struct osd_device *dev = osd_dt_dev(dt);
473 LASSERT(dev != NULL);
474 if (unlikely(dev->od_mnt == NULL))
477 rc = kstrtoull_from_user(buffer, count, 0, &val);
481 dev->od_full_scrub_threshold_rate = val;
484 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_threshold_rate);
487 ldiskfs_osd_track_declares_assert_seq_show(struct seq_file *m, void *data)
489 seq_printf(m, "%d\n", ldiskfs_track_declares_assert);
494 ldiskfs_osd_track_declares_assert_seq_write(struct file *file,
495 const char __user *buffer,
496 size_t count, loff_t *off)
498 bool track_declares_assert;
501 rc = kstrtobool_from_user(buffer, count, &track_declares_assert);
505 ldiskfs_track_declares_assert = track_declares_assert;
509 LPROC_SEQ_FOPS(ldiskfs_osd_track_declares_assert);
511 static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
513 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
515 LASSERT(dev != NULL);
516 if (unlikely(dev->od_mnt == NULL))
519 osd_scrub_dump(m, dev);
522 LPROC_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
524 static int ldiskfs_osd_readcache_seq_show(struct seq_file *m, void *data)
526 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
528 LASSERT(osd != NULL);
529 if (unlikely(osd->od_mnt == NULL))
532 seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
537 ldiskfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
538 size_t count, loff_t *off)
540 struct seq_file *m = file->private_data;
541 struct dt_device *dt = m->private;
542 struct osd_device *osd = osd_dt_dev(dt);
546 LASSERT(osd != NULL);
547 if (unlikely(osd->od_mnt == NULL))
550 rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
556 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
557 OSD_MAX_CACHE_SIZE : val;
560 LPROC_SEQ_FOPS(ldiskfs_osd_readcache);
562 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
563 static int ldiskfs_osd_index_in_idif_seq_show(struct seq_file *m, void *data)
565 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
567 LASSERT(dev != NULL);
568 if (unlikely(dev->od_mnt == NULL))
571 seq_printf(m, "%d\n", (int)(dev->od_index_in_idif));
576 ldiskfs_osd_index_in_idif_seq_write(struct file *file,
577 const char __user *buffer,
578 size_t count, loff_t *off)
581 struct seq_file *m = file->private_data;
582 struct dt_device *dt = m->private;
583 struct osd_device *dev = osd_dt_dev(dt);
584 struct lu_target *tgt;
588 LASSERT(dev != NULL);
589 if (unlikely(dev->od_mnt == NULL))
592 rc = kstrtobool_from_user(buffer, count, &val);
596 if (dev->od_index_in_idif) {
600 LCONSOLE_WARN("%s: OST-index in IDIF has been enabled, "
601 "it cannot be reverted back.\n", osd_name(dev));
608 rc = lu_env_init(&env, LCT_DT_THREAD);
612 tgt = dev->od_dt_dev.dd_lu_dev.ld_site->ls_tgt;
613 tgt->lut_lsd.lsd_feature_rocompat |= OBD_ROCOMPAT_IDX_IN_IDIF;
614 rc = tgt_server_data_update(&env, tgt, 1);
619 LCONSOLE_INFO("%s: enable OST-index in IDIF successfully, "
620 "it cannot be reverted back.\n", osd_name(dev));
622 dev->od_index_in_idif = 1;
625 LPROC_SEQ_FOPS(ldiskfs_osd_index_in_idif);
627 int osd_register_proc_index_in_idif(struct osd_device *osd)
629 struct proc_dir_entry *proc;
631 proc = proc_create_data("index_in_idif", 0, osd->od_proc_entry,
632 &ldiskfs_osd_index_in_idif_fops,
641 static int ldiskfs_osd_index_backup_seq_show(struct seq_file *m, void *data)
643 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
645 LASSERT(dev != NULL);
646 if (unlikely(dev->od_mnt == NULL))
649 seq_printf(m, "%d\n", dev->od_index_backup_policy);
653 static ssize_t ldiskfs_osd_index_backup_seq_write(struct file *file,
654 const char __user *buffer,
655 size_t count, loff_t *off)
657 struct seq_file *m = file->private_data;
658 struct dt_device *dt = m->private;
659 struct osd_device *dev = osd_dt_dev(dt);
663 LASSERT(dev != NULL);
664 if (unlikely(dev->od_mnt == NULL))
667 rc = kstrtoint_from_user(buffer, count, 0, &val);
671 dev->od_index_backup_policy = val;
674 LPROC_SEQ_FOPS(ldiskfs_osd_index_backup);
676 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_blksize);
677 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytestotal);
678 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesfree);
679 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesavail);
680 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filestotal);
681 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filesfree);
683 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
684 { .name = "blocksize",
685 .fops = &ldiskfs_dt_blksize_fops },
686 { .name = "kbytestotal",
687 .fops = &ldiskfs_dt_kbytestotal_fops },
688 { .name = "kbytesfree",
689 .fops = &ldiskfs_dt_kbytesfree_fops },
690 { .name = "kbytesavail",
691 .fops = &ldiskfs_dt_kbytesavail_fops },
692 { .name = "filestotal",
693 .fops = &ldiskfs_dt_filestotal_fops },
694 { .name = "filesfree",
695 .fops = &ldiskfs_dt_filesfree_fops },
697 .fops = &ldiskfs_osd_fstype_fops },
699 .fops = &ldiskfs_osd_mntdev_fops },
700 { .name = "force_sync",
701 .fops = &ldiskfs_osd_force_sync_fops },
703 .fops = &ldiskfs_osd_pdo_fops },
704 { .name = "auto_scrub",
705 .fops = &ldiskfs_osd_auto_scrub_fops },
706 { .name = "full_scrub_ratio",
707 .fops = &ldiskfs_osd_full_scrub_ratio_fops },
708 { .name = "full_scrub_threshold_rate",
709 .fops = &ldiskfs_osd_full_scrub_threshold_rate_fops },
710 { .name = "oi_scrub",
711 .fops = &ldiskfs_osd_oi_scrub_fops },
712 { .name = "read_cache_enable",
713 .fops = &ldiskfs_osd_cache_fops },
714 { .name = "writethrough_cache_enable",
715 .fops = &ldiskfs_osd_wcache_fops },
716 { .name = "readcache_max_filesize",
717 .fops = &ldiskfs_osd_readcache_fops },
718 { .name = "index_backup",
719 .fops = &ldiskfs_osd_index_backup_fops },
723 struct lprocfs_vars lprocfs_osd_module_vars[] = {
724 { .name = "track_declares_assert",
725 .fops = &ldiskfs_osd_track_declares_assert_fops },
730 int osd_procfs_init(struct osd_device *osd, const char *name)
732 struct obd_type *type;
736 if (osd->od_proc_entry)
739 /* at the moment there is no linkage between lu_type
740 * and obd_type, so we lookup obd_type this way */
741 type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
743 LASSERT(name != NULL);
744 LASSERT(type != NULL);
746 /* Find the type procroot and add the proc entry for this device */
747 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
748 lprocfs_osd_obd_vars,
750 if (IS_ERR(osd->od_proc_entry)) {
751 rc = PTR_ERR(osd->od_proc_entry);
752 CERROR("Error %d setting up lprocfs for %s\n",
754 osd->od_proc_entry = NULL;
758 rc = osd_stats_init(osd);
763 osd_procfs_fini(osd);
767 int osd_procfs_fini(struct osd_device *osd)
770 lprocfs_free_stats(&osd->od_stats);
772 if (osd->od_proc_entry)
773 lprocfs_remove(&osd->od_proc_entry);