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>
40 #include <lustre/lustre_idl.h>
42 #include "osd_internal.h"
46 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
48 struct brw_stats *s = &osd->od_brw_stats;
49 sector_t *last_block = NULL;
50 struct page **pages = iobuf->dr_pages;
51 struct page *last_page = NULL;
52 unsigned long discont_pages = 0;
53 unsigned long discont_blocks = 0;
54 sector_t *blocks = iobuf->dr_blocks;
55 int i, nr_pages = iobuf->dr_npages;
57 int rw = iobuf->dr_rw;
59 if (unlikely(nr_pages == 0))
62 blocks_per_page = PAGE_SIZE >> osd_sb(osd)->s_blocksize_bits;
64 lprocfs_oh_tally_log2(&s->hist[BRW_R_PAGES+rw], nr_pages);
66 while (nr_pages-- > 0) {
67 if (last_page && (*pages)->index != (last_page->index + 1))
71 for (i = 0; i < blocks_per_page; i++) {
72 if (last_block && *blocks != (*last_block + 1))
74 last_block = blocks++;
78 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_PAGES+rw], discont_pages);
79 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
82 #define pct(a, b) (b ? a * 100 / b : 0)
84 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
85 struct obd_histogram *read, struct obd_histogram *write, int scale)
87 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
90 seq_printf(seq, "\n%26s read | write\n", " ");
91 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
94 read_tot = lprocfs_oh_sum(read);
95 write_tot = lprocfs_oh_sum(write);
96 for (i = 0; i < OBD_HIST_MAX; i++) {
97 r = read->oh_buckets[i];
98 w = write->oh_buckets[i];
101 if (read_cum == 0 && write_cum == 0)
105 seq_printf(seq, "%u", i);
107 seq_printf(seq, "%u", scale << i);
109 seq_printf(seq, "%uK", scale << (i-10));
111 seq_printf(seq, "%uM", scale << (i-20));
113 seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
114 r, pct(r, read_tot), pct(read_cum, read_tot),
115 w, pct(w, write_tot), pct(write_cum, write_tot));
117 if (read_cum == read_tot && write_cum == write_tot)
122 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
124 struct timespec64 now;
126 /* this sampling races with updates */
127 ktime_get_real_ts64(&now);
129 seq_printf(seq, "snapshot_time: %lld.%09ld (secs.nsecs)\n",
130 (s64)now.tv_sec, now.tv_nsec);
132 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
133 &brw_stats->hist[BRW_R_PAGES],
134 &brw_stats->hist[BRW_W_PAGES], 1);
136 display_brw_stats(seq, "discontiguous pages", "rpcs",
137 &brw_stats->hist[BRW_R_DISCONT_PAGES],
138 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
140 display_brw_stats(seq, "discontiguous blocks", "rpcs",
141 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
142 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
144 display_brw_stats(seq, "disk fragmented I/Os", "ios",
145 &brw_stats->hist[BRW_R_DIO_FRAGS],
146 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
148 display_brw_stats(seq, "disk I/Os in flight", "ios",
149 &brw_stats->hist[BRW_R_RPC_HIST],
150 &brw_stats->hist[BRW_W_RPC_HIST], 0);
152 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
153 &brw_stats->hist[BRW_R_IO_TIME],
154 &brw_stats->hist[BRW_W_IO_TIME],
155 jiffies_to_msecs(1000) / MSEC_PER_SEC);
157 display_brw_stats(seq, "disk I/O size", "ios",
158 &brw_stats->hist[BRW_R_DISK_IOSIZE],
159 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
164 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
166 struct osd_device *osd = seq->private;
168 brw_stats_show(seq, &osd->od_brw_stats);
173 static ssize_t osd_brw_stats_seq_write(struct file *file,
174 const char __user *buf,
175 size_t len, loff_t *off)
177 struct seq_file *seq = file->private_data;
178 struct osd_device *osd = seq->private;
181 for (i = 0; i < BRW_LAST; i++)
182 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
187 LPROC_SEQ_FOPS(osd_brw_stats);
189 static int osd_stats_init(struct osd_device *osd)
194 for (i = 0; i < BRW_LAST; i++)
195 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
197 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
198 if (osd->od_stats != NULL) {
199 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
204 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
205 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
207 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
208 LPROCFS_CNTR_AVGMINMAX,
209 "get_page_failures", "num");
210 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
211 LPROCFS_CNTR_AVGMINMAX,
212 "cache_access", "pages");
213 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
214 LPROCFS_CNTR_AVGMINMAX,
215 "cache_hit", "pages");
216 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
217 LPROCFS_CNTR_AVGMINMAX,
218 "cache_miss", "pages");
219 #if OSD_THANDLE_STATS
220 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
221 LPROCFS_CNTR_AVGMINMAX,
222 "thandle starting", "usec");
223 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
224 LPROCFS_CNTR_AVGMINMAX,
225 "thandle open", "usec");
226 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
227 LPROCFS_CNTR_AVGMINMAX,
228 "thandle closing", "usec");
230 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
231 0644, &osd_brw_stats_fops, osd);
239 static int ldiskfs_osd_fstype_seq_show(struct seq_file *m, void *data)
241 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
243 LASSERT(osd != NULL);
244 seq_puts(m, "ldiskfs\n");
247 LPROC_SEQ_FOPS_RO(ldiskfs_osd_fstype);
249 static int ldiskfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
251 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
253 LASSERT(osd != NULL);
254 if (unlikely(osd->od_mnt == NULL))
257 seq_printf(m, "%s\n", osd->od_mntdev);
260 LPROC_SEQ_FOPS_RO(ldiskfs_osd_mntdev);
262 static int ldiskfs_osd_cache_seq_show(struct seq_file *m, void *data)
264 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
266 LASSERT(osd != NULL);
267 if (unlikely(osd->od_mnt == NULL))
270 seq_printf(m, "%u\n", osd->od_read_cache);
275 ldiskfs_osd_cache_seq_write(struct file *file, const char __user *buffer,
276 size_t count, loff_t *off)
278 struct seq_file *m = file->private_data;
279 struct dt_device *dt = m->private;
280 struct osd_device *osd = osd_dt_dev(dt);
284 LASSERT(osd != NULL);
285 if (unlikely(osd->od_mnt == NULL))
288 rc = lprocfs_str_to_s64(buffer, count, &val);
292 osd->od_read_cache = !!val;
295 LPROC_SEQ_FOPS(ldiskfs_osd_cache);
297 static int ldiskfs_osd_wcache_seq_show(struct seq_file *m, void *data)
299 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
301 LASSERT(osd != NULL);
302 if (unlikely(osd->od_mnt == NULL))
305 seq_printf(m, "%u\n", osd->od_writethrough_cache);
310 ldiskfs_osd_wcache_seq_write(struct file *file, const char __user *buffer,
311 size_t count, loff_t *off)
313 struct seq_file *m = file->private_data;
314 struct dt_device *dt = m->private;
315 struct osd_device *osd = osd_dt_dev(dt);
319 LASSERT(osd != NULL);
320 if (unlikely(osd->od_mnt == NULL))
323 rc = lprocfs_str_to_s64(buffer, count, &val);
327 osd->od_writethrough_cache = !!val;
330 LPROC_SEQ_FOPS(ldiskfs_osd_wcache);
333 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
334 size_t count, loff_t *off)
336 struct seq_file *m = file->private_data;
337 struct dt_device *dt = m->private;
338 struct osd_device *osd = osd_dt_dev(dt);
342 LASSERT(osd != NULL);
343 if (unlikely(osd->od_mnt == NULL))
346 rc = lu_env_init(&env, LCT_LOCAL);
349 rc = dt_sync(&env, dt);
352 return rc == 0 ? count : rc;
354 LPROC_SEQ_FOPS_WO_TYPE(ldiskfs, osd_force_sync);
356 static int ldiskfs_osd_pdo_seq_show(struct seq_file *m, void *data)
358 seq_printf(m, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
363 ldiskfs_osd_pdo_seq_write(struct file *file, const char __user *buffer,
364 size_t count, loff_t *off)
369 rc = lprocfs_str_to_s64(buffer, count, &pdo);
377 LPROC_SEQ_FOPS(ldiskfs_osd_pdo);
379 static int ldiskfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
381 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
383 LASSERT(dev != NULL);
384 if (unlikely(dev->od_mnt == NULL))
387 seq_printf(m, "%d\n", !dev->od_noscrub);
392 ldiskfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
393 size_t count, loff_t *off)
395 struct seq_file *m = file->private_data;
396 struct dt_device *dt = m->private;
397 struct osd_device *dev = osd_dt_dev(dt);
401 LASSERT(dev != NULL);
402 if (unlikely(dev->od_mnt == NULL))
405 rc = lprocfs_str_to_s64(buffer, count, &val);
409 dev->od_noscrub = !val;
412 LPROC_SEQ_FOPS(ldiskfs_osd_auto_scrub);
414 static int ldiskfs_osd_full_scrub_ratio_seq_show(struct seq_file *m, void *data)
416 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
418 LASSERT(dev != NULL);
419 if (unlikely(dev->od_mnt == NULL))
422 seq_printf(m, "%llu\n", dev->od_full_scrub_ratio);
427 ldiskfs_osd_full_scrub_ratio_seq_write(struct file *file,
428 const char __user *buffer,
429 size_t count, loff_t *off)
431 struct seq_file *m = file->private_data;
432 struct dt_device *dt = m->private;
433 struct osd_device *dev = osd_dt_dev(dt);
437 LASSERT(dev != NULL);
438 if (unlikely(dev->od_mnt == NULL))
441 rc = lprocfs_str_to_s64(buffer, count, &val);
448 dev->od_full_scrub_ratio = val;
451 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_ratio);
453 static int ldiskfs_osd_full_scrub_threshold_rate_seq_show(struct seq_file *m,
456 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
458 LASSERT(dev != NULL);
459 if (unlikely(dev->od_mnt == NULL))
462 seq_printf(m, "%llu (bad OI mappings/minute)\n",
463 dev->od_full_scrub_threshold_rate);
468 ldiskfs_osd_full_scrub_threshold_rate_seq_write(struct file *file,
469 const char __user *buffer,
470 size_t count, loff_t *off)
472 struct seq_file *m = file->private_data;
473 struct dt_device *dt = m->private;
474 struct osd_device *dev = osd_dt_dev(dt);
478 LASSERT(dev != NULL);
479 if (unlikely(dev->od_mnt == NULL))
482 rc = lprocfs_str_to_s64(buffer, count, &val);
489 dev->od_full_scrub_threshold_rate = val;
492 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_threshold_rate);
495 ldiskfs_osd_track_declares_assert_seq_show(struct seq_file *m, void *data)
497 seq_printf(m, "%d\n", ldiskfs_track_declares_assert);
502 ldiskfs_osd_track_declares_assert_seq_write(struct file *file,
503 const char __user *buffer,
504 size_t count, loff_t *off)
506 __s64 track_declares_assert;
509 rc = lprocfs_str_to_s64(buffer, count, &track_declares_assert);
513 ldiskfs_track_declares_assert = !!track_declares_assert;
517 LPROC_SEQ_FOPS(ldiskfs_osd_track_declares_assert);
519 static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
521 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
523 LASSERT(dev != NULL);
524 if (unlikely(dev->od_mnt == NULL))
527 return 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 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_blksize);
649 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytestotal);
650 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesfree);
651 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesavail);
652 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filestotal);
653 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filesfree);
655 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
656 { .name = "blocksize",
657 .fops = &ldiskfs_dt_blksize_fops },
658 { .name = "kbytestotal",
659 .fops = &ldiskfs_dt_kbytestotal_fops },
660 { .name = "kbytesfree",
661 .fops = &ldiskfs_dt_kbytesfree_fops },
662 { .name = "kbytesavail",
663 .fops = &ldiskfs_dt_kbytesavail_fops },
664 { .name = "filestotal",
665 .fops = &ldiskfs_dt_filestotal_fops },
666 { .name = "filesfree",
667 .fops = &ldiskfs_dt_filesfree_fops },
669 .fops = &ldiskfs_osd_fstype_fops },
671 .fops = &ldiskfs_osd_mntdev_fops },
672 { .name = "force_sync",
673 .fops = &ldiskfs_osd_force_sync_fops },
675 .fops = &ldiskfs_osd_pdo_fops },
676 { .name = "auto_scrub",
677 .fops = &ldiskfs_osd_auto_scrub_fops },
678 { .name = "full_scrub_ratio",
679 .fops = &ldiskfs_osd_full_scrub_ratio_fops },
680 { .name = "full_scrub_threshold_rate",
681 .fops = &ldiskfs_osd_full_scrub_threshold_rate_fops },
682 { .name = "oi_scrub",
683 .fops = &ldiskfs_osd_oi_scrub_fops },
684 { .name = "read_cache_enable",
685 .fops = &ldiskfs_osd_cache_fops },
686 { .name = "writethrough_cache_enable",
687 .fops = &ldiskfs_osd_wcache_fops },
688 { .name = "readcache_max_filesize",
689 .fops = &ldiskfs_osd_readcache_fops },
693 struct lprocfs_vars lprocfs_osd_module_vars[] = {
694 { .name = "track_declares_assert",
695 .fops = &ldiskfs_osd_track_declares_assert_fops },
700 int osd_procfs_init(struct osd_device *osd, const char *name)
702 struct obd_type *type;
706 if (osd->od_proc_entry)
709 /* at the moment there is no linkage between lu_type
710 * and obd_type, so we lookup obd_type this way */
711 type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
713 LASSERT(name != NULL);
714 LASSERT(type != NULL);
716 /* Find the type procroot and add the proc entry for this device */
717 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
718 lprocfs_osd_obd_vars,
720 if (IS_ERR(osd->od_proc_entry)) {
721 rc = PTR_ERR(osd->od_proc_entry);
722 CERROR("Error %d setting up lprocfs for %s\n",
724 osd->od_proc_entry = NULL;
728 rc = osd_stats_init(osd);
733 osd_procfs_fini(osd);
737 int osd_procfs_fini(struct osd_device *osd)
740 lprocfs_free_stats(&osd->od_stats);
742 if (osd->od_proc_entry)
743 lprocfs_remove(&osd->od_proc_entry);