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], 1);
153 display_brw_stats(seq, "disk I/O size", "ios",
154 &brw_stats->hist[BRW_R_DISK_IOSIZE],
155 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
158 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
160 struct osd_device *osd = seq->private;
162 brw_stats_show(seq, &osd->od_brw_stats);
167 static ssize_t osd_brw_stats_seq_write(struct file *file,
168 const char __user *buf,
169 size_t len, loff_t *off)
171 struct seq_file *seq = file->private_data;
172 struct osd_device *osd = seq->private;
175 for (i = 0; i < BRW_LAST; i++)
176 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
181 LPROC_SEQ_FOPS(osd_brw_stats);
183 static int osd_stats_init(struct osd_device *osd)
188 for (i = 0; i < BRW_LAST; i++)
189 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
191 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
192 if (osd->od_stats != NULL) {
193 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
198 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
199 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
201 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
202 LPROCFS_CNTR_AVGMINMAX,
203 "get_page_failures", "num");
204 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
205 LPROCFS_CNTR_AVGMINMAX,
206 "cache_access", "pages");
207 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
208 LPROCFS_CNTR_AVGMINMAX,
209 "cache_hit", "pages");
210 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
211 LPROCFS_CNTR_AVGMINMAX,
212 "cache_miss", "pages");
213 #if OSD_THANDLE_STATS
214 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
215 LPROCFS_CNTR_AVGMINMAX,
216 "thandle starting", "usec");
217 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
218 LPROCFS_CNTR_AVGMINMAX,
219 "thandle open", "usec");
220 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
221 LPROCFS_CNTR_AVGMINMAX,
222 "thandle closing", "usec");
224 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
225 0644, &osd_brw_stats_fops, osd);
233 static int ldiskfs_osd_fstype_seq_show(struct seq_file *m, void *data)
235 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
237 LASSERT(osd != NULL);
238 seq_puts(m, "ldiskfs\n");
241 LPROC_SEQ_FOPS_RO(ldiskfs_osd_fstype);
243 static int ldiskfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
245 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
247 LASSERT(osd != NULL);
248 if (unlikely(osd->od_mnt == NULL))
251 seq_printf(m, "%s\n", osd->od_mntdev);
254 LPROC_SEQ_FOPS_RO(ldiskfs_osd_mntdev);
256 static int ldiskfs_osd_cache_seq_show(struct seq_file *m, void *data)
258 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
260 LASSERT(osd != NULL);
261 if (unlikely(osd->od_mnt == NULL))
264 seq_printf(m, "%u\n", osd->od_read_cache);
269 ldiskfs_osd_cache_seq_write(struct file *file, const char __user *buffer,
270 size_t count, loff_t *off)
272 struct seq_file *m = file->private_data;
273 struct dt_device *dt = m->private;
274 struct osd_device *osd = osd_dt_dev(dt);
278 LASSERT(osd != NULL);
279 if (unlikely(osd->od_mnt == NULL))
282 rc = kstrtobool_from_user(buffer, count, &val);
286 osd->od_read_cache = val;
289 LPROC_SEQ_FOPS(ldiskfs_osd_cache);
291 static int ldiskfs_osd_wcache_seq_show(struct seq_file *m, void *data)
293 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
295 LASSERT(osd != NULL);
296 if (unlikely(osd->od_mnt == NULL))
299 seq_printf(m, "%u\n", osd->od_writethrough_cache);
304 ldiskfs_osd_wcache_seq_write(struct file *file, const char __user *buffer,
305 size_t count, loff_t *off)
307 struct seq_file *m = file->private_data;
308 struct dt_device *dt = m->private;
309 struct osd_device *osd = osd_dt_dev(dt);
313 LASSERT(osd != NULL);
314 if (unlikely(osd->od_mnt == NULL))
317 rc = kstrtobool_from_user(buffer, count, &val);
321 osd->od_writethrough_cache = val;
324 LPROC_SEQ_FOPS(ldiskfs_osd_wcache);
327 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
328 size_t count, loff_t *off)
330 struct seq_file *m = file->private_data;
331 struct dt_device *dt = m->private;
332 struct osd_device *osd = osd_dt_dev(dt);
336 LASSERT(osd != NULL);
337 if (unlikely(osd->od_mnt == NULL))
340 rc = lu_env_init(&env, LCT_LOCAL);
343 rc = dt_sync(&env, dt);
346 return rc == 0 ? count : rc;
348 LPROC_SEQ_FOPS_WR_ONLY(ldiskfs, osd_force_sync);
350 static int ldiskfs_osd_pdo_seq_show(struct seq_file *m, void *data)
352 seq_printf(m, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
357 ldiskfs_osd_pdo_seq_write(struct file *file, const char __user *buffer,
358 size_t count, loff_t *off)
363 rc = kstrtobool_from_user(buffer, count, &pdo);
371 LPROC_SEQ_FOPS(ldiskfs_osd_pdo);
373 static int ldiskfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
375 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
377 LASSERT(dev != NULL);
378 if (unlikely(dev->od_mnt == NULL))
381 seq_printf(m, "%lld\n", dev->od_auto_scrub_interval);
386 ldiskfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
387 size_t count, loff_t *off)
389 struct seq_file *m = file->private_data;
390 struct dt_device *dt = m->private;
391 struct osd_device *dev = osd_dt_dev(dt);
395 LASSERT(dev != NULL);
396 if (unlikely(dev->od_mnt == NULL))
399 rc = kstrtoll_from_user(buffer, count, 0, &val);
403 dev->od_auto_scrub_interval = val;
406 LPROC_SEQ_FOPS(ldiskfs_osd_auto_scrub);
408 static int ldiskfs_osd_full_scrub_ratio_seq_show(struct seq_file *m, void *data)
410 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
412 LASSERT(dev != NULL);
413 if (unlikely(dev->od_mnt == NULL))
416 seq_printf(m, "%llu\n", dev->od_full_scrub_ratio);
421 ldiskfs_osd_full_scrub_ratio_seq_write(struct file *file,
422 const char __user *buffer,
423 size_t count, loff_t *off)
425 struct seq_file *m = file->private_data;
426 struct dt_device *dt = m->private;
427 struct osd_device *dev = osd_dt_dev(dt);
431 LASSERT(dev != NULL);
432 if (unlikely(dev->od_mnt == NULL))
435 rc = kstrtoll_from_user(buffer, count, 0, &val);
442 dev->od_full_scrub_ratio = val;
445 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_ratio);
447 static int ldiskfs_osd_full_scrub_threshold_rate_seq_show(struct seq_file *m,
450 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
452 LASSERT(dev != NULL);
453 if (unlikely(dev->od_mnt == NULL))
456 seq_printf(m, "%llu (bad OI mappings/minute)\n",
457 dev->od_full_scrub_threshold_rate);
462 ldiskfs_osd_full_scrub_threshold_rate_seq_write(struct file *file,
463 const char __user *buffer,
464 size_t count, loff_t *off)
466 struct seq_file *m = file->private_data;
467 struct dt_device *dt = m->private;
468 struct osd_device *dev = osd_dt_dev(dt);
472 LASSERT(dev != NULL);
473 if (unlikely(dev->od_mnt == NULL))
476 rc = kstrtoull_from_user(buffer, count, 0, &val);
480 dev->od_full_scrub_threshold_rate = val;
483 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_threshold_rate);
486 ldiskfs_osd_track_declares_assert_seq_show(struct seq_file *m, void *data)
488 seq_printf(m, "%d\n", ldiskfs_track_declares_assert);
493 ldiskfs_osd_track_declares_assert_seq_write(struct file *file,
494 const char __user *buffer,
495 size_t count, loff_t *off)
497 bool track_declares_assert;
500 rc = kstrtobool_from_user(buffer, count, &track_declares_assert);
504 ldiskfs_track_declares_assert = track_declares_assert;
508 LPROC_SEQ_FOPS(ldiskfs_osd_track_declares_assert);
510 static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
512 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
514 LASSERT(dev != NULL);
515 if (unlikely(dev->od_mnt == NULL))
518 osd_scrub_dump(m, dev);
521 LPROC_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
523 static int ldiskfs_osd_readcache_seq_show(struct seq_file *m, void *data)
525 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
527 LASSERT(osd != NULL);
528 if (unlikely(osd->od_mnt == NULL))
531 seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
536 ldiskfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
537 size_t count, loff_t *off)
539 struct seq_file *m = file->private_data;
540 struct dt_device *dt = m->private;
541 struct osd_device *osd = osd_dt_dev(dt);
545 LASSERT(osd != NULL);
546 if (unlikely(osd->od_mnt == NULL))
549 rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
555 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
556 OSD_MAX_CACHE_SIZE : val;
559 LPROC_SEQ_FOPS(ldiskfs_osd_readcache);
561 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
562 static int ldiskfs_osd_index_in_idif_seq_show(struct seq_file *m, void *data)
564 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
566 LASSERT(dev != NULL);
567 if (unlikely(dev->od_mnt == NULL))
570 seq_printf(m, "%d\n", (int)(dev->od_index_in_idif));
575 ldiskfs_osd_index_in_idif_seq_write(struct file *file,
576 const char __user *buffer,
577 size_t count, loff_t *off)
580 struct seq_file *m = file->private_data;
581 struct dt_device *dt = m->private;
582 struct osd_device *dev = osd_dt_dev(dt);
583 struct lu_target *tgt;
587 LASSERT(dev != NULL);
588 if (unlikely(dev->od_mnt == NULL))
591 rc = kstrtobool_from_user(buffer, count, &val);
595 if (dev->od_index_in_idif) {
599 LCONSOLE_WARN("%s: OST-index in IDIF has been enabled, "
600 "it cannot be reverted back.\n", osd_name(dev));
607 rc = lu_env_init(&env, LCT_DT_THREAD);
611 tgt = dev->od_dt_dev.dd_lu_dev.ld_site->ls_tgt;
612 tgt->lut_lsd.lsd_feature_rocompat |= OBD_ROCOMPAT_IDX_IN_IDIF;
613 rc = tgt_server_data_update(&env, tgt, 1);
618 LCONSOLE_INFO("%s: enable OST-index in IDIF successfully, "
619 "it cannot be reverted back.\n", osd_name(dev));
621 dev->od_index_in_idif = 1;
624 LPROC_SEQ_FOPS(ldiskfs_osd_index_in_idif);
626 int osd_register_proc_index_in_idif(struct osd_device *osd)
628 struct proc_dir_entry *proc;
630 proc = proc_create_data("index_in_idif", 0, osd->od_proc_entry,
631 &ldiskfs_osd_index_in_idif_fops,
640 static int ldiskfs_osd_index_backup_seq_show(struct seq_file *m, void *data)
642 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
644 LASSERT(dev != NULL);
645 if (unlikely(dev->od_mnt == NULL))
648 seq_printf(m, "%d\n", dev->od_index_backup_policy);
652 static ssize_t ldiskfs_osd_index_backup_seq_write(struct file *file,
653 const char __user *buffer,
654 size_t count, loff_t *off)
656 struct seq_file *m = file->private_data;
657 struct dt_device *dt = m->private;
658 struct osd_device *dev = osd_dt_dev(dt);
662 LASSERT(dev != NULL);
663 if (unlikely(dev->od_mnt == NULL))
666 rc = kstrtoint_from_user(buffer, count, 0, &val);
670 dev->od_index_backup_policy = val;
673 LPROC_SEQ_FOPS(ldiskfs_osd_index_backup);
675 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_blksize);
676 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytestotal);
677 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesfree);
678 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesavail);
679 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filestotal);
680 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filesfree);
682 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
683 { .name = "blocksize",
684 .fops = &ldiskfs_dt_blksize_fops },
685 { .name = "kbytestotal",
686 .fops = &ldiskfs_dt_kbytestotal_fops },
687 { .name = "kbytesfree",
688 .fops = &ldiskfs_dt_kbytesfree_fops },
689 { .name = "kbytesavail",
690 .fops = &ldiskfs_dt_kbytesavail_fops },
691 { .name = "filestotal",
692 .fops = &ldiskfs_dt_filestotal_fops },
693 { .name = "filesfree",
694 .fops = &ldiskfs_dt_filesfree_fops },
696 .fops = &ldiskfs_osd_fstype_fops },
698 .fops = &ldiskfs_osd_mntdev_fops },
699 { .name = "force_sync",
700 .fops = &ldiskfs_osd_force_sync_fops },
702 .fops = &ldiskfs_osd_pdo_fops },
703 { .name = "auto_scrub",
704 .fops = &ldiskfs_osd_auto_scrub_fops },
705 { .name = "full_scrub_ratio",
706 .fops = &ldiskfs_osd_full_scrub_ratio_fops },
707 { .name = "full_scrub_threshold_rate",
708 .fops = &ldiskfs_osd_full_scrub_threshold_rate_fops },
709 { .name = "oi_scrub",
710 .fops = &ldiskfs_osd_oi_scrub_fops },
711 { .name = "read_cache_enable",
712 .fops = &ldiskfs_osd_cache_fops },
713 { .name = "writethrough_cache_enable",
714 .fops = &ldiskfs_osd_wcache_fops },
715 { .name = "readcache_max_filesize",
716 .fops = &ldiskfs_osd_readcache_fops },
717 { .name = "index_backup",
718 .fops = &ldiskfs_osd_index_backup_fops },
722 struct lprocfs_vars lprocfs_osd_module_vars[] = {
723 { .name = "track_declares_assert",
724 .fops = &ldiskfs_osd_track_declares_assert_fops },
729 int osd_procfs_init(struct osd_device *osd, const char *name)
731 struct obd_type *type;
735 if (osd->od_proc_entry)
738 /* at the moment there is no linkage between lu_type
739 * and obd_type, so we lookup obd_type this way */
740 type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
742 LASSERT(name != NULL);
743 LASSERT(type != NULL);
745 /* Find the type procroot and add the proc entry for this device */
746 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
747 lprocfs_osd_obd_vars,
749 if (IS_ERR(osd->od_proc_entry)) {
750 rc = PTR_ERR(osd->od_proc_entry);
751 CERROR("Error %d setting up lprocfs for %s\n",
753 osd->od_proc_entry = NULL;
757 rc = osd_stats_init(osd);
762 osd_procfs_fini(osd);
766 int osd_procfs_fini(struct osd_device *osd)
769 lprocfs_free_stats(&osd->od_stats);
771 if (osd->od_proc_entry)
772 lprocfs_remove(&osd->od_proc_entry);