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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/osd/osd_lproc.c
38 * Author: Mikhail Pershin <tappro@sun.com>
41 #define DEBUG_SUBSYSTEM S_OSD
43 #include <lprocfs_status.h>
44 #include <lustre/lustre_idl.h>
46 #include "osd_internal.h"
50 void osd_brw_stats_update(struct osd_device *osd, struct osd_iobuf *iobuf)
52 struct brw_stats *s = &osd->od_brw_stats;
53 sector_t *last_block = NULL;
54 struct page **pages = iobuf->dr_pages;
55 struct page *last_page = NULL;
56 unsigned long discont_pages = 0;
57 unsigned long discont_blocks = 0;
58 sector_t *blocks = iobuf->dr_blocks;
59 int i, nr_pages = iobuf->dr_npages;
61 int rw = iobuf->dr_rw;
63 if (unlikely(nr_pages == 0))
66 blocks_per_page = PAGE_CACHE_SIZE >> osd_sb(osd)->s_blocksize_bits;
68 lprocfs_oh_tally_log2(&s->hist[BRW_R_PAGES+rw], nr_pages);
70 while (nr_pages-- > 0) {
71 if (last_page && (*pages)->index != (last_page->index + 1))
75 for (i = 0; i < blocks_per_page; i++) {
76 if (last_block && *blocks != (*last_block + 1))
78 last_block = blocks++;
82 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_PAGES+rw], discont_pages);
83 lprocfs_oh_tally(&s->hist[BRW_R_DISCONT_BLOCKS+rw], discont_blocks);
86 #define pct(a, b) (b ? a * 100 / b : 0)
88 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
89 struct obd_histogram *read, struct obd_histogram *write, int scale)
91 unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
94 seq_printf(seq, "\n%26s read | write\n", " ");
95 seq_printf(seq, "%-22s %-5s %% cum %% | %-11s %% cum %%\n",
98 read_tot = lprocfs_oh_sum(read);
99 write_tot = lprocfs_oh_sum(write);
100 for (i = 0; i < OBD_HIST_MAX; i++) {
101 r = read->oh_buckets[i];
102 w = write->oh_buckets[i];
105 if (read_cum == 0 && write_cum == 0)
109 seq_printf(seq, "%u", i);
111 seq_printf(seq, "%u", scale << i);
113 seq_printf(seq, "%uK", scale << (i-10));
115 seq_printf(seq, "%uM", scale << (i-20));
117 seq_printf(seq, ":\t\t%10lu %3lu %3lu | %4lu %3lu %3lu\n",
118 r, pct(r, read_tot), pct(read_cum, read_tot),
119 w, pct(w, write_tot), pct(write_cum, write_tot));
121 if (read_cum == read_tot && write_cum == write_tot)
126 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
130 /* this sampling races with updates */
131 do_gettimeofday(&now);
132 seq_printf(seq, "snapshot_time: %lu.%lu (secs.usecs)\n",
133 now.tv_sec, now.tv_usec);
135 display_brw_stats(seq, "pages per bulk r/w", "rpcs",
136 &brw_stats->hist[BRW_R_PAGES],
137 &brw_stats->hist[BRW_W_PAGES], 1);
139 display_brw_stats(seq, "discontiguous pages", "rpcs",
140 &brw_stats->hist[BRW_R_DISCONT_PAGES],
141 &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
143 display_brw_stats(seq, "discontiguous blocks", "rpcs",
144 &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
145 &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
147 display_brw_stats(seq, "disk fragmented I/Os", "ios",
148 &brw_stats->hist[BRW_R_DIO_FRAGS],
149 &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
151 display_brw_stats(seq, "disk I/Os in flight", "ios",
152 &brw_stats->hist[BRW_R_RPC_HIST],
153 &brw_stats->hist[BRW_W_RPC_HIST], 0);
155 display_brw_stats(seq, "I/O time (1/1000s)", "ios",
156 &brw_stats->hist[BRW_R_IO_TIME],
157 &brw_stats->hist[BRW_W_IO_TIME],
158 jiffies_to_msecs(1000) / MSEC_PER_SEC);
160 display_brw_stats(seq, "disk I/O size", "ios",
161 &brw_stats->hist[BRW_R_DISK_IOSIZE],
162 &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
167 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
169 struct osd_device *osd = seq->private;
171 brw_stats_show(seq, &osd->od_brw_stats);
176 static ssize_t osd_brw_stats_seq_write(struct file *file, const char *buf,
177 size_t len, loff_t *off)
179 struct seq_file *seq = file->private_data;
180 struct osd_device *osd = seq->private;
183 for (i = 0; i < BRW_LAST; i++)
184 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
189 LPROC_SEQ_FOPS(osd_brw_stats);
191 static int osd_stats_init(struct osd_device *osd)
196 for (i = 0; i < BRW_LAST; i++)
197 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
199 osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
200 if (osd->od_stats != NULL) {
201 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
206 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
207 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
209 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
210 LPROCFS_CNTR_AVGMINMAX,
211 "get_page_failures", "num");
212 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
213 LPROCFS_CNTR_AVGMINMAX,
214 "cache_access", "pages");
215 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
216 LPROCFS_CNTR_AVGMINMAX,
217 "cache_hit", "pages");
218 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
219 LPROCFS_CNTR_AVGMINMAX,
220 "cache_miss", "pages");
221 #if OSD_THANDLE_STATS
222 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
223 LPROCFS_CNTR_AVGMINMAX,
224 "thandle starting", "usec");
225 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
226 LPROCFS_CNTR_AVGMINMAX,
227 "thandle open", "usec");
228 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
229 LPROCFS_CNTR_AVGMINMAX,
230 "thandle closing", "usec");
232 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
233 0644, &osd_brw_stats_fops, osd);
241 static int ldiskfs_osd_fstype_seq_show(struct seq_file *m, void *data)
243 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
245 LASSERT(osd != NULL);
246 return seq_printf(m, "ldiskfs\n");
248 LPROC_SEQ_FOPS_RO(ldiskfs_osd_fstype);
250 static int ldiskfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
252 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
254 LASSERT(osd != NULL);
255 if (unlikely(osd->od_mnt == NULL))
258 return 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 return seq_printf(m, "%u\n", osd->od_read_cache);
274 ldiskfs_osd_cache_seq_write(struct file *file, const char *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);
282 LASSERT(osd != NULL);
283 if (unlikely(osd->od_mnt == NULL))
286 rc = lprocfs_write_helper(buffer, count, &val);
290 osd->od_read_cache = !!val;
293 LPROC_SEQ_FOPS(ldiskfs_osd_cache);
295 static int ldiskfs_osd_wcache_seq_show(struct seq_file *m, void *data)
297 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
299 LASSERT(osd != NULL);
300 if (unlikely(osd->od_mnt == NULL))
303 return seq_printf(m, "%u\n", osd->od_writethrough_cache);
307 ldiskfs_osd_wcache_seq_write(struct file *file, const char *buffer,
308 size_t count, loff_t *off)
310 struct seq_file *m = file->private_data;
311 struct dt_device *dt = m->private;
312 struct osd_device *osd = osd_dt_dev(dt);
315 LASSERT(osd != NULL);
316 if (unlikely(osd->od_mnt == NULL))
319 rc = lprocfs_write_helper(buffer, count, &val);
323 osd->od_writethrough_cache = !!val;
326 LPROC_SEQ_FOPS(ldiskfs_osd_wcache);
329 lprocfs_osd_force_sync_seq_write(struct file *file, const char *buffer,
330 size_t count, loff_t *off)
332 struct seq_file *m = file->private_data;
333 struct dt_device *dt = m->private;
334 struct osd_device *osd = osd_dt_dev(dt);
338 LASSERT(osd != NULL);
339 if (unlikely(osd->od_mnt == NULL))
342 rc = lu_env_init(&env, LCT_LOCAL);
345 rc = dt_sync(&env, dt);
348 return rc == 0 ? count : rc;
350 LPROC_SEQ_FOPS_WO_TYPE(ldiskfs, osd_force_sync);
352 static int ldiskfs_osd_pdo_seq_show(struct seq_file *m, void *data)
354 return seq_printf(m, "%s\n", ldiskfs_pdo ? "ON" : "OFF");
358 ldiskfs_osd_pdo_seq_write(struct file *file, const char *buffer,
359 size_t count, loff_t *off)
363 rc = lprocfs_write_helper(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 return seq_printf(m, "%d\n", !dev->od_noscrub);
385 ldiskfs_osd_auto_scrub_seq_write(struct file *file, const char *buffer,
386 size_t count, loff_t *off)
388 struct seq_file *m = file->private_data;
389 struct dt_device *dt = m->private;
390 struct osd_device *dev = osd_dt_dev(dt);
393 LASSERT(dev != NULL);
394 if (unlikely(dev->od_mnt == NULL))
397 rc = lprocfs_write_helper(buffer, count, &val);
401 dev->od_noscrub = !val;
404 LPROC_SEQ_FOPS(ldiskfs_osd_auto_scrub);
406 static int ldiskfs_osd_full_scrub_ratio_seq_show(struct seq_file *m, void *data)
408 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
410 LASSERT(dev != NULL);
411 if (unlikely(dev->od_mnt == NULL))
414 return seq_printf(m, LPU64"\n", dev->od_full_scrub_ratio);
418 ldiskfs_osd_full_scrub_ratio_seq_write(struct file *file, const char *buffer,
419 size_t count, loff_t *off)
421 struct seq_file *m = file->private_data;
422 struct dt_device *dt = m->private;
423 struct osd_device *dev = osd_dt_dev(dt);
426 LASSERT(dev != NULL);
427 if (unlikely(dev->od_mnt == NULL))
430 rc = lprocfs_write_helper(buffer, count, &val);
437 dev->od_full_scrub_ratio = val;
440 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_ratio);
442 static int ldiskfs_osd_full_scrub_threshold_rate_seq_show(struct seq_file *m,
445 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
447 LASSERT(dev != NULL);
448 if (unlikely(dev->od_mnt == NULL))
451 return seq_printf(m, LPU64" (bad OI mappings/minute)\n",
452 dev->od_full_scrub_threshold_rate);
456 ldiskfs_osd_full_scrub_threshold_rate_seq_write(struct file *file,
458 size_t count, loff_t *off)
460 struct seq_file *m = file->private_data;
461 struct dt_device *dt = m->private;
462 struct osd_device *dev = osd_dt_dev(dt);
465 LASSERT(dev != NULL);
466 if (unlikely(dev->od_mnt == NULL))
469 rc = lprocfs_write_helper(buffer, count, &val);
476 dev->od_full_scrub_threshold_rate = val;
479 LPROC_SEQ_FOPS(ldiskfs_osd_full_scrub_threshold_rate);
482 ldiskfs_osd_track_declares_assert_seq_show(struct seq_file *m, void *data)
484 return seq_printf(m, "%d\n", ldiskfs_track_declares_assert);
488 ldiskfs_osd_track_declares_assert_seq_write(struct file *file,
490 size_t count, loff_t *off)
492 int track_declares_assert;
495 rc = lprocfs_write_helper(buffer, count, &track_declares_assert);
499 ldiskfs_track_declares_assert = !!track_declares_assert;
503 LPROC_SEQ_FOPS(ldiskfs_osd_track_declares_assert);
505 static int ldiskfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
507 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
509 LASSERT(dev != NULL);
510 if (unlikely(dev->od_mnt == NULL))
513 return osd_scrub_dump(m, dev);
515 LPROC_SEQ_FOPS_RO(ldiskfs_osd_oi_scrub);
517 static int ldiskfs_osd_readcache_seq_show(struct seq_file *m, void *data)
519 struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
521 LASSERT(osd != NULL);
522 if (unlikely(osd->od_mnt == NULL))
525 return seq_printf(m, LPU64"\n", osd->od_readcache_max_filesize);
529 ldiskfs_osd_readcache_seq_write(struct file *file, const char *buffer,
530 size_t count, loff_t *off)
532 struct seq_file *m = file->private_data;
533 struct dt_device *dt = m->private;
534 struct osd_device *osd = osd_dt_dev(dt);
538 LASSERT(osd != NULL);
539 if (unlikely(osd->od_mnt == NULL))
542 rc = lprocfs_write_u64_helper(buffer, count, &val);
546 osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
547 OSD_MAX_CACHE_SIZE : val;
550 LPROC_SEQ_FOPS(ldiskfs_osd_readcache);
552 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(3, 0, 52, 0)
553 static int ldiskfs_osd_index_in_idif_seq_show(struct seq_file *m, void *data)
555 struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
557 LASSERT(dev != NULL);
558 if (unlikely(dev->od_mnt == NULL))
561 return seq_printf(m, "%d\n", (int)(dev->od_index_in_idif));
565 ldiskfs_osd_index_in_idif_seq_write(struct file *file, const char *buffer,
566 size_t count, loff_t *off)
569 struct seq_file *m = file->private_data;
570 struct dt_device *dt = m->private;
571 struct osd_device *dev = osd_dt_dev(dt);
572 struct lu_target *tgt;
576 LASSERT(dev != NULL);
577 if (unlikely(dev->od_mnt == NULL))
580 rc = lprocfs_write_helper(buffer, count, &val);
584 if (dev->od_index_in_idif) {
588 LCONSOLE_WARN("%s: OST-index in IDIF has been enabled, "
589 "it cannot be reverted back.\n", osd_name(dev));
596 rc = lu_env_init(&env, LCT_DT_THREAD);
600 tgt = dev->od_dt_dev.dd_lu_dev.ld_site->ls_tgt;
601 tgt->lut_lsd.lsd_feature_rocompat |= OBD_ROCOMPAT_IDX_IN_IDIF;
602 rc = tgt_server_data_update(&env, tgt, 1);
607 LCONSOLE_INFO("%s: enable OST-index in IDIF successfully, "
608 "it cannot be reverted back.\n", osd_name(dev));
610 dev->od_index_in_idif = 1;
613 LPROC_SEQ_FOPS(ldiskfs_osd_index_in_idif);
615 int osd_register_proc_index_in_idif(struct osd_device *osd)
617 struct proc_dir_entry *proc;
619 proc = proc_create_data("index_in_idif", 0, osd->od_proc_entry,
620 &ldiskfs_osd_index_in_idif_fops,
629 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_blksize);
630 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytestotal);
631 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesfree);
632 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_kbytesavail);
633 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filestotal);
634 LPROC_SEQ_FOPS_RO_TYPE(ldiskfs, dt_filesfree);
636 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
637 { .name = "blocksize",
638 .fops = &ldiskfs_dt_blksize_fops },
639 { .name = "kbytestotal",
640 .fops = &ldiskfs_dt_kbytestotal_fops },
641 { .name = "kbytesfree",
642 .fops = &ldiskfs_dt_kbytesfree_fops },
643 { .name = "kbytesavail",
644 .fops = &ldiskfs_dt_kbytesavail_fops },
645 { .name = "filestotal",
646 .fops = &ldiskfs_dt_filestotal_fops },
647 { .name = "filesfree",
648 .fops = &ldiskfs_dt_filesfree_fops },
650 .fops = &ldiskfs_osd_fstype_fops },
652 .fops = &ldiskfs_osd_mntdev_fops },
653 { .name = "force_sync",
654 .fops = &ldiskfs_osd_force_sync_fops },
656 .fops = &ldiskfs_osd_pdo_fops },
657 { .name = "auto_scrub",
658 .fops = &ldiskfs_osd_auto_scrub_fops },
659 { .name = "full_scrub_ratio",
660 .fops = &ldiskfs_osd_full_scrub_ratio_fops },
661 { .name = "full_scrub_threshold_rate",
662 .fops = &ldiskfs_osd_full_scrub_threshold_rate_fops },
663 { .name = "oi_scrub",
664 .fops = &ldiskfs_osd_oi_scrub_fops },
665 { .name = "read_cache_enable",
666 .fops = &ldiskfs_osd_cache_fops },
667 { .name = "writethrough_cache_enable",
668 .fops = &ldiskfs_osd_wcache_fops },
669 { .name = "readcache_max_filesize",
670 .fops = &ldiskfs_osd_readcache_fops },
674 struct lprocfs_vars lprocfs_osd_module_vars[] = {
675 { .name = "track_declares_assert",
676 .fops = &ldiskfs_osd_track_declares_assert_fops },
681 int osd_procfs_init(struct osd_device *osd, const char *name)
683 struct obd_type *type;
687 if (osd->od_proc_entry)
690 /* at the moment there is no linkage between lu_type
691 * and obd_type, so we lookup obd_type this way */
692 type = class_search_type(LUSTRE_OSD_LDISKFS_NAME);
694 LASSERT(name != NULL);
695 LASSERT(type != NULL);
697 /* Find the type procroot and add the proc entry for this device */
698 osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
699 lprocfs_osd_obd_vars,
701 if (IS_ERR(osd->od_proc_entry)) {
702 rc = PTR_ERR(osd->od_proc_entry);
703 CERROR("Error %d setting up lprocfs for %s\n",
705 osd->od_proc_entry = NULL;
709 rc = osd_stats_init(osd);
714 osd_procfs_fini(osd);
718 int osd_procfs_fini(struct osd_device *osd)
721 lprocfs_free_stats(&osd->od_stats);
723 if (osd->od_proc_entry)
724 lprocfs_remove(&osd->od_proc_entry);