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 #define pct(a, b) (b ? a * 100 / b : 0)
83 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
84 struct obd_histogram *read, struct obd_histogram *write, int scale)
86 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
89 seq_printf(seq, "\n%26s read | write\n", " ");
90 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
93 read_tot = lprocfs_oh_sum(read);
94 write_tot = lprocfs_oh_sum(write);
95 for (i = 0; i < OBD_HIST_MAX; i++) {
96 r = read->oh_buckets[i];
97 w = write->oh_buckets[i];
100 if (read_cum == 0 && write_cum == 0)
104 seq_printf(seq, "%u", i);
106 seq_printf(seq, "%u", scale << i);
108 seq_printf(seq, "%uK", scale << (i-10));
110 seq_printf(seq, "%uM", scale << (i-20));
112 seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
113 r, pct(r, read_tot), pct(read_cum, read_tot),
114 w, pct(w, write_tot), pct(write_cum, write_tot));
116 if (read_cum == read_tot && write_cum == write_tot)
121 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
123 struct timespec64 now;
125 /* this sampling races with updates */
126 ktime_get_real_ts64(&now);
128 seq_printf(seq, "snapshot_time: %lld.%09ld (secs.nsecs)\n",
129 (s64)now.tv_sec, now.tv_nsec);
131 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
132 &brw_stats->hist[BRW_R_PAGES],
133 &brw_stats->hist[BRW_W_PAGES], 1);
135 display_brw_stats(seq, "discontiguous pages", "rpcs",
136 &brw_stats->hist[BRW_R_DISCONT_PAGES],
137 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
139 display_brw_stats(seq, "discontiguous blocks", "rpcs",
140 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
141 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
143 display_brw_stats(seq, "disk fragmented I/Os", "ios",
144 &brw_stats->hist[BRW_R_DIO_FRAGS],
145 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
147 display_brw_stats(seq, "disk I/Os in flight", "ios",
148 &brw_stats->hist[BRW_R_RPC_HIST],
149 &brw_stats->hist[BRW_W_RPC_HIST], 0);
151 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
152 &brw_stats->hist[BRW_R_IO_TIME],
153 &brw_stats->hist[BRW_W_IO_TIME],
154 jiffies_to_msecs(1000) / MSEC_PER_SEC);
156 display_brw_stats(seq, "disk I/O size", "ios",
157 &brw_stats->hist[BRW_R_DISK_IOSIZE],
158 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
163 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
165 struct osd_device *osd = seq->private;
167 brw_stats_show(seq, &osd->od_brw_stats);
172 static ssize_t osd_brw_stats_seq_write(struct file *file,
173 const char __user *buf,
174 size_t len, loff_t *off)
176 struct seq_file *seq = file->private_data;
177 struct osd_device *osd = seq->private;
180 for (i = 0; i < BRW_LAST; i++)
181 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
186 LPROC_SEQ_FOPS(osd_brw_stats);
188 static int osd_stats_init(struct osd_device *osd)
193 for (i = 0; i < BRW_LAST; i++)
194 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
196 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
197 if (osd->od_stats != NULL) {
198 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
203 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
204 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
206 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
207 LPROCFS_CNTR_AVGMINMAX,
208 "get_page_failures", "num");
209 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
210 LPROCFS_CNTR_AVGMINMAX,
211 "cache_access", "pages");
212 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
213 LPROCFS_CNTR_AVGMINMAX,
214 "cache_hit", "pages");
215 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
216 LPROCFS_CNTR_AVGMINMAX,
217 "cache_miss", "pages");
218 #if OSD_THANDLE_STATS
219 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
220 LPROCFS_CNTR_AVGMINMAX,
221 "thandle starting", "usec");
222 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
223 LPROCFS_CNTR_AVGMINMAX,
224 "thandle open", "usec");
225 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
226 LPROCFS_CNTR_AVGMINMAX,
227 "thandle closing", "usec");
229 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
230 0644, &osd_brw_stats_fops, osd);
238 static int ldiskfs_osd_fstype_seq_show(struct seq_file *m, void *data)
240 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
242 LASSERT(osd != NULL);
243 seq_puts(m, "ldiskfs\n");
246 LPROC_SEQ_FOPS_RO(ldiskfs_osd_fstype);
248 static int ldiskfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
250 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
252 LASSERT(osd != NULL);
253 if (unlikely(osd->od_mnt == NULL))
256 seq_printf(m, "%s\n", osd->od_mntdev);
259 LPROC_SEQ_FOPS_RO(ldiskfs_osd_mntdev);
261 static int ldiskfs_osd_cache_seq_show(struct seq_file *m, void *data)
263 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
265 LASSERT(osd != NULL);
266 if (unlikely(osd->od_mnt == NULL))
269 seq_printf(m, "%u\n", osd->od_read_cache);
274 ldiskfs_osd_cache_seq_write(struct file *file, const char __user *buffer,
275 size_t count, loff_t *off)
277 struct seq_file *m = file->private_data;
278 struct dt_device *dt = m->private;
279 struct osd_device *osd = osd_dt_dev(dt);
283 LASSERT(osd != NULL);
284 if (unlikely(osd->od_mnt == NULL))
287 rc = lprocfs_str_to_s64(buffer, count, &val);
291 osd->od_read_cache = !!val;
294 LPROC_SEQ_FOPS(ldiskfs_osd_cache);
296 static int ldiskfs_osd_wcache_seq_show(struct seq_file *m, void *data)
298 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
300 LASSERT(osd != NULL);
301 if (unlikely(osd->od_mnt == NULL))
304 seq_printf(m, "%u\n", osd->od_writethrough_cache);
309 ldiskfs_osd_wcache_seq_write(struct file *file, const char __user *buffer,
310 size_t count, loff_t *off)
312 struct seq_file *m = file->private_data;
313 struct dt_device *dt = m->private;
314 struct osd_device *osd = osd_dt_dev(dt);
318 LASSERT(osd != NULL);
319 if (unlikely(osd->od_mnt == NULL))
322 rc = lprocfs_str_to_s64(buffer, count, &val);
326 osd->od_writethrough_cache = !!val;
329 LPROC_SEQ_FOPS(ldiskfs_osd_wcache);
332 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
333 size_t count, loff_t *off)
335 struct seq_file *m = file->private_data;
336 struct dt_device *dt = m->private;
337 struct osd_device *osd = osd_dt_dev(dt);
341 LASSERT(osd != NULL);
342 if (unlikely(osd->od_mnt == NULL))
345 rc = lu_env_init(&env, LCT_LOCAL);
348 rc = dt_sync(&env, dt);
351 return rc == 0 ? count : rc;
353 LPROC_SEQ_FOPS_WR_ONLY(ldiskfs, osd_force_sync);
355 static int ldiskfs_osd_pdo_seq_show(struct seq_file *m, void *data)
357 seq_printf(m, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
362 ldiskfs_osd_pdo_seq_write(struct file *file, const char __user *buffer,
363 size_t count, loff_t *off)
368 rc = lprocfs_str_to_s64(buffer, count, &pdo);
376 LPROC_SEQ_FOPS(ldiskfs_osd_pdo);
378 static int ldiskfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
380 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
382 LASSERT(dev != NULL);
383 if (unlikely(dev->od_mnt == NULL))
386 seq_printf(m, "%lld\n", dev->od_auto_scrub_interval);
391 ldiskfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
392 size_t count, loff_t *off)
394 struct seq_file *m = file->private_data;
395 struct dt_device *dt = m->private;
396 struct osd_device *dev = osd_dt_dev(dt);
400 LASSERT(dev != NULL);
401 if (unlikely(dev->od_mnt == NULL))
404 rc = lprocfs_str_to_s64(buffer, count, &val);
408 dev->od_auto_scrub_interval = val;
411 LPROC_SEQ_FOPS(ldiskfs_osd_auto_scrub);
413 static int ldiskfs_osd_full_scrub_ratio_seq_show(struct seq_file *m, void *data)
415 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
417 LASSERT(dev != NULL);
418 if (unlikely(dev->od_mnt == NULL))
421 seq_printf(m, "%llu\n", dev->od_full_scrub_ratio);
426 ldiskfs_osd_full_scrub_ratio_seq_write(struct file *file,
427 const char __user *buffer,
428 size_t count, loff_t *off)
430 struct seq_file *m = file->private_data;
431 struct dt_device *dt = m->private;
432 struct osd_device *dev = osd_dt_dev(dt);
436 LASSERT(dev != NULL);
437 if (unlikely(dev->od_mnt == NULL))
440 rc = lprocfs_str_to_s64(buffer, count, &val);
447 dev->od_full_scrub_ratio = val;
450 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_ratio);
452 static int ldiskfs_osd_full_scrub_threshold_rate_seq_show(struct seq_file *m,
455 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
457 LASSERT(dev != NULL);
458 if (unlikely(dev->od_mnt == NULL))
461 seq_printf(m, "%llu (bad OI mappings/minute)\n",
462 dev->od_full_scrub_threshold_rate);
467 ldiskfs_osd_full_scrub_threshold_rate_seq_write(struct file *file,
468 const char __user *buffer,
469 size_t count, loff_t *off)
471 struct seq_file *m = file->private_data;
472 struct dt_device *dt = m->private;
473 struct osd_device *dev = osd_dt_dev(dt);
477 LASSERT(dev != NULL);
478 if (unlikely(dev->od_mnt == NULL))
481 rc = lprocfs_str_to_s64(buffer, count, &val);
488 dev->od_full_scrub_threshold_rate = val;
491 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_threshold_rate);
494 ldiskfs_osd_track_declares_assert_seq_show(struct seq_file *m, void *data)
496 seq_printf(m, "%d\n", ldiskfs_track_declares_assert);
501 ldiskfs_osd_track_declares_assert_seq_write(struct file *file,
502 const char __user *buffer,
503 size_t count, loff_t *off)
505 __s64 track_declares_assert;
508 rc = lprocfs_str_to_s64(buffer, count, &track_declares_assert);
512 ldiskfs_track_declares_assert = !!track_declares_assert;
516 LPROC_SEQ_FOPS(ldiskfs_osd_track_declares_assert);
518 static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
520 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
522 LASSERT(dev != NULL);
523 if (unlikely(dev->od_mnt == NULL))
526 osd_scrub_dump(m, dev);
529 LPROC_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
531 static int ldiskfs_osd_readcache_seq_show(struct seq_file *m, void *data)
533 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
535 LASSERT(osd != NULL);
536 if (unlikely(osd->od_mnt == NULL))
539 seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
544 ldiskfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
545 size_t count, loff_t *off)
547 struct seq_file *m = file->private_data;
548 struct dt_device *dt = m->private;
549 struct osd_device *osd = osd_dt_dev(dt);
553 LASSERT(osd != NULL);
554 if (unlikely(osd->od_mnt == NULL))
557 rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
563 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
564 OSD_MAX_CACHE_SIZE : val;
567 LPROC_SEQ_FOPS(ldiskfs_osd_readcache);
569 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
570 static int ldiskfs_osd_index_in_idif_seq_show(struct seq_file *m, void *data)
572 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
574 LASSERT(dev != NULL);
575 if (unlikely(dev->od_mnt == NULL))
578 seq_printf(m, "%d\n", (int)(dev->od_index_in_idif));
583 ldiskfs_osd_index_in_idif_seq_write(struct file *file,
584 const char __user *buffer,
585 size_t count, loff_t *off)
588 struct seq_file *m = file->private_data;
589 struct dt_device *dt = m->private;
590 struct osd_device *dev = osd_dt_dev(dt);
591 struct lu_target *tgt;
595 LASSERT(dev != NULL);
596 if (unlikely(dev->od_mnt == NULL))
599 rc = lprocfs_str_to_s64(buffer, count, &val);
603 if (dev->od_index_in_idif) {
607 LCONSOLE_WARN("%s: OST-index in IDIF has been enabled, "
608 "it cannot be reverted back.\n", osd_name(dev));
615 rc = lu_env_init(&env, LCT_DT_THREAD);
619 tgt = dev->od_dt_dev.dd_lu_dev.ld_site->ls_tgt;
620 tgt->lut_lsd.lsd_feature_rocompat |= OBD_ROCOMPAT_IDX_IN_IDIF;
621 rc = tgt_server_data_update(&env, tgt, 1);
626 LCONSOLE_INFO("%s: enable OST-index in IDIF successfully, "
627 "it cannot be reverted back.\n", osd_name(dev));
629 dev->od_index_in_idif = 1;
632 LPROC_SEQ_FOPS(ldiskfs_osd_index_in_idif);
634 int osd_register_proc_index_in_idif(struct osd_device *osd)
636 struct proc_dir_entry *proc;
638 proc = proc_create_data("index_in_idif", 0, osd->od_proc_entry,
639 &ldiskfs_osd_index_in_idif_fops,
648 static int ldiskfs_osd_index_backup_seq_show(struct seq_file *m, void *data)
650 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
652 LASSERT(dev != NULL);
653 if (unlikely(dev->od_mnt == NULL))
656 seq_printf(m, "%d\n", dev->od_index_backup_policy);
660 static ssize_t ldiskfs_osd_index_backup_seq_write(struct file *file,
661 const char __user *buffer,
662 size_t count, loff_t *off)
664 struct seq_file *m = file->private_data;
665 struct dt_device *dt = m->private;
666 struct osd_device *dev = osd_dt_dev(dt);
670 LASSERT(dev != NULL);
671 if (unlikely(dev->od_mnt == NULL))
674 rc = lprocfs_str_to_s64(buffer, count, &val);
678 dev->od_index_backup_policy = val;
681 LPROC_SEQ_FOPS(ldiskfs_osd_index_backup);
683 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_blksize);
684 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytestotal);
685 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesfree);
686 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesavail);
687 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filestotal);
688 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filesfree);
690 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
691 { .name = "blocksize",
692 .fops = &ldiskfs_dt_blksize_fops },
693 { .name = "kbytestotal",
694 .fops = &ldiskfs_dt_kbytestotal_fops },
695 { .name = "kbytesfree",
696 .fops = &ldiskfs_dt_kbytesfree_fops },
697 { .name = "kbytesavail",
698 .fops = &ldiskfs_dt_kbytesavail_fops },
699 { .name = "filestotal",
700 .fops = &ldiskfs_dt_filestotal_fops },
701 { .name = "filesfree",
702 .fops = &ldiskfs_dt_filesfree_fops },
704 .fops = &ldiskfs_osd_fstype_fops },
706 .fops = &ldiskfs_osd_mntdev_fops },
707 { .name = "force_sync",
708 .fops = &ldiskfs_osd_force_sync_fops },
710 .fops = &ldiskfs_osd_pdo_fops },
711 { .name = "auto_scrub",
712 .fops = &ldiskfs_osd_auto_scrub_fops },
713 { .name = "full_scrub_ratio",
714 .fops = &ldiskfs_osd_full_scrub_ratio_fops },
715 { .name = "full_scrub_threshold_rate",
716 .fops = &ldiskfs_osd_full_scrub_threshold_rate_fops },
717 { .name = "oi_scrub",
718 .fops = &ldiskfs_osd_oi_scrub_fops },
719 { .name = "read_cache_enable",
720 .fops = &ldiskfs_osd_cache_fops },
721 { .name = "writethrough_cache_enable",
722 .fops = &ldiskfs_osd_wcache_fops },
723 { .name = "readcache_max_filesize",
724 .fops = &ldiskfs_osd_readcache_fops },
725 { .name = "index_backup",
726 .fops = &ldiskfs_osd_index_backup_fops },
730 struct lprocfs_vars lprocfs_osd_module_vars[] = {
731 { .name = "track_declares_assert",
732 .fops = &ldiskfs_osd_track_declares_assert_fops },
737 int osd_procfs_init(struct osd_device *osd, const char *name)
739 struct obd_type *type;
743 if (osd->od_proc_entry)
746 /* at the moment there is no linkage between lu_type
747 * and obd_type, so we lookup obd_type this way */
748 type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
750 LASSERT(name != NULL);
751 LASSERT(type != NULL);
753 /* Find the type procroot and add the proc entry for this device */
754 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
755 lprocfs_osd_obd_vars,
757 if (IS_ERR(osd->od_proc_entry)) {
758 rc = PTR_ERR(osd->od_proc_entry);
759 CERROR("Error %d setting up lprocfs for %s\n",
761 osd->od_proc_entry = NULL;
765 rc = osd_stats_init(osd);
770 osd_procfs_fini(osd);
774 int osd_procfs_fini(struct osd_device *osd)
777 lprocfs_free_stats(&osd->od_stats);
779 if (osd->od_proc_entry)
780 lprocfs_remove(&osd->od_proc_entry);