lustre/osd-zfs/osd_lproc.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   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.
   9  *
  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).
  15  *
  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
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * lustre/osd-zfs/osd_lproc.c
  32  *
  33  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  34  * Author: Mike Pershin <tappro@whamcloud.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_OSD
  38
  39 #include <obd.h>
  40 #include <obd_class.h>
  41 #include <lprocfs_status.h>
  42 #include <lustre_scrub.h>
  43
  44 #include "osd_internal.h"
  45
  46 #ifdef CONFIG_PROC_FS
  47
  48 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
  49                               struct obd_histogram *read,
  50                               struct obd_histogram *write, int scale)
  51 {
  52         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
  53         int i;
  54
  55         seq_printf(seq, "\n%26s read      |     write\n", " ");
  56         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
  57                    name, units, units);
  58
  59         read_tot = lprocfs_oh_sum(read);
  60         write_tot = lprocfs_oh_sum(write);
  61         for (i = 0; i < OBD_HIST_MAX; i++) {
  62                 r = read->oh_buckets[i];
  63                 w = write->oh_buckets[i];
  64                 read_cum += r;
  65                 write_cum += w;
  66                 if (read_cum == 0 && write_cum == 0)
  67                         continue;
  68
  69                 if (!scale)
  70                         seq_printf(seq, "%u", i);
  71                 else if (i < 10)
  72                         seq_printf(seq, "%u", scale << i);
  73                 else if (i < 20)
  74                         seq_printf(seq, "%uK", scale << (i-10));
  75                 else
  76                         seq_printf(seq, "%uM", scale << (i-20));
  77
  78                 seq_printf(seq, ":\t\t%10lu %3u %3u   | %4lu %3u %3u\n",
  79                            r, pct(r, read_tot), pct(read_cum, read_tot),
  80                            w, pct(w, write_tot), pct(write_cum, write_tot));
  81
  82                 if (read_cum == read_tot && write_cum == write_tot)
  83                         break;
  84         }
  85 }
  86
  87 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
  88 {
  89         struct timespec64 now;
  90
  91         /* this sampling races with updates */
  92         ktime_get_real_ts64(&now);
  93         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
  94                    (s64)now.tv_sec, now.tv_nsec);
  95
  96         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
  97                           &brw_stats->hist[BRW_R_PAGES],
  98                           &brw_stats->hist[BRW_W_PAGES], 1);
  99         display_brw_stats(seq, "discontiguous pages", "rpcs",
 100                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
 101                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
 102 #if 0
 103         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 104                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
 105                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
 106
 107         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 108                           &brw_stats->hist[BRW_R_DIO_FRAGS],
 109                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
 110 #endif
 111         display_brw_stats(seq, "disk I/Os in flight", "ios",
 112                           &brw_stats->hist[BRW_R_RPC_HIST],
 113                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 114
 115         display_brw_stats(seq, "I/O time (1/1000s)", "ios",
 116                           &brw_stats->hist[BRW_R_IO_TIME],
 117                           &brw_stats->hist[BRW_W_IO_TIME], 1);
 118
 119         display_brw_stats(seq, "disk I/O size", "ios",
 120                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
 121                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 122 }
 123
 124 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
 125 {
 126         struct osd_device *osd = seq->private;
 127
 128         brw_stats_show(seq, &osd->od_brw_stats);
 129
 130         return 0;
 131 }
 132
 133 static ssize_t osd_brw_stats_seq_write(struct file *file,
 134                                        const char __user *buf,
 135                                        size_t len, loff_t *off)
 136 {
 137         struct seq_file *seq = file->private_data;
 138         struct osd_device *osd = seq->private;
 139         int i;
 140
 141         for (i = 0; i < BRW_LAST; i++)
 142                 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
 143
 144         return len;
 145 }
 146
 147 LPROC_SEQ_FOPS(osd_brw_stats);
 148
 149 static int osd_stats_init(struct osd_device *osd)
 150 {
 151         int result, i;
 152         ENTRY;
 153
 154         for (i = 0; i < BRW_LAST; i++)
 155                 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
 156
 157         osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
 158         if (osd->od_stats != NULL) {
 159                 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
 160                                 osd->od_stats);
 161                 if (result)
 162                         GOTO(out, result);
 163
 164                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
 165                                 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
 166                                 "get_page", "usec");
 167                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
 168                                 LPROCFS_CNTR_AVGMINMAX,
 169                                 "get_page_failures", "num");
 170                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
 171                                 LPROCFS_CNTR_AVGMINMAX,
 172                                 "cache_access", "pages");
 173                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
 174                                 LPROCFS_CNTR_AVGMINMAX,
 175                                 "cache_hit", "pages");
 176                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
 177                                 LPROCFS_CNTR_AVGMINMAX,
 178                                 "cache_miss", "pages");
 179                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
 180                                 LPROCFS_CNTR_AVGMINMAX,
 181                                 "copy", "pages");
 182                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
 183                                 LPROCFS_CNTR_AVGMINMAX,
 184                                 "zerocopy", "pages");
 185                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
 186                                 LPROCFS_CNTR_AVGMINMAX,
 187                                 "tail", "pages");
 188 #ifdef OSD_THANDLE_STATS
 189                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
 190                                 LPROCFS_CNTR_AVGMINMAX,
 191                                 "thandle_starting", "usec");
 192                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
 193                                 LPROCFS_CNTR_AVGMINMAX,
 194                                 "thandle_open", "usec");
 195                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
 196                                 LPROCFS_CNTR_AVGMINMAX,
 197                                 "thandle_closing", "usec");
 198 #endif
 199                 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
 200                                             0644, &osd_brw_stats_fops, osd);
 201         } else {
 202                 result = -ENOMEM;
 203         }
 204
 205 out:
 206         RETURN(result);
 207 }
 208
 209 static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
 210 {
 211         struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 212
 213         LASSERT(dev != NULL);
 214         if (!dev->od_os)
 215                 return -EINPROGRESS;
 216
 217         scrub_dump(m, &dev->od_scrub);
 218         return 0;
 219 }
 220 LDEBUGFS_SEQ_FOPS_RO(zfs_osd_oi_scrub);
 221
 222 static ssize_t auto_scrub_show(struct kobject *kobj, struct attribute *attr,
 223                               char *buf)
 224 {
 225         struct dt_device *dt = container_of(kobj, struct dt_device,
 226                                             dd_kobj);
 227         struct osd_device *dev = osd_dt_dev(dt);
 228
 229         LASSERT(dev);
 230         if (!dev->od_os)
 231                 return -EINPROGRESS;
 232
 233         return sprintf(buf, "%lld\n", dev->od_auto_scrub_interval);
 234 }
 235
 236 static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
 237                                 const char *buffer, size_t count)
 238 {
 239         struct dt_device *dt = container_of(kobj, struct dt_device,
 240                                             dd_kobj);
 241         struct osd_device *dev = osd_dt_dev(dt);
 242         s64 val;
 243         int rc;
 244
 245         LASSERT(dev);
 246         if (!dev->od_os)
 247                 return -EINPROGRESS;
 248
 249         rc = kstrtoull(buffer, 0, &val);
 250         if (rc)
 251                 return rc;
 252
 253         dev->od_auto_scrub_interval = val;
 254         return count;
 255 }
 256 LUSTRE_RW_ATTR(auto_scrub);
 257
 258 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 259                           char *buf)
 260 {
 261         return sprintf(buf, "zfs\n");
 262 }
 263 LUSTRE_RO_ATTR(fstype);
 264
 265 static ssize_t mntdev_show(struct kobject *kobj, struct attribute *attr,
 266                            char *buf)
 267 {
 268         struct dt_device *dt = container_of(kobj, struct dt_device,
 269                                             dd_kobj);
 270         struct osd_device *osd = osd_dt_dev(dt);
 271
 272         LASSERT(osd);
 273
 274         return sprintf(buf, "%s\n", osd->od_mntdev);
 275 }
 276 LUSTRE_RO_ATTR(mntdev);
 277
 278 ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
 279                          const char *buffer, size_t count)
 280 {
 281         struct dt_device *dt = container_of(kobj, struct dt_device,
 282                                             dd_kobj);
 283         struct lu_env env;
 284         int rc;
 285
 286         rc = lu_env_init(&env, LCT_LOCAL);
 287         if (rc)
 288                 return rc;
 289
 290         rc = dt_sync(&env, dt);
 291         lu_env_fini(&env);
 292
 293         return rc == 0 ? count : rc;
 294 }
 295 LUSTRE_WO_ATTR(force_sync);
 296
 297 static ssize_t nonrotational_show(struct kobject *kobj, struct attribute *attr,
 298                                   char *buf)
 299 {
 300         struct dt_device *dt = container_of(kobj, struct dt_device,
 301                                             dd_kobj);
 302         struct osd_device *osd = osd_dt_dev(dt);
 303
 304         LASSERT(osd);
 305         if (!osd->od_os)
 306                 return -EINPROGRESS;
 307
 308         return sprintf(buf, "%u\n", osd->od_nonrotational);
 309 }
 310
 311 static ssize_t nonrotational_store(struct kobject *kobj,
 312                                    struct attribute *attr, const char *buffer,
 313                                    size_t count)
 314 {
 315         struct dt_device *dt = container_of(kobj, struct dt_device,
 316                                             dd_kobj);
 317         struct osd_device *osd = osd_dt_dev(dt);
 318         bool val;
 319         int rc;
 320
 321         LASSERT(osd);
 322         if (!osd->od_os)
 323                 return -EINPROGRESS;
 324
 325         rc = kstrtobool(buffer, &val);
 326         if (rc)
 327                 return rc;
 328
 329         osd->od_nonrotational = val;
 330         return count;
 331 }
 332 LUSTRE_RW_ATTR(nonrotational);
 333
 334 static ssize_t index_backup_show(struct kobject *kobj, struct attribute *attr,
 335                                  char *buf)
 336 {
 337         struct dt_device *dt = container_of(kobj, struct dt_device,
 338                                             dd_kobj);
 339         struct osd_device *dev = osd_dt_dev(dt);
 340
 341         LASSERT(dev);
 342         if (!dev->od_os)
 343                 return -EINPROGRESS;
 344
 345         return sprintf(buf, "%d\n", dev->od_index_backup_policy);
 346 }
 347
 348 ssize_t index_backup_store(struct kobject *kobj, struct attribute *attr,
 349                            const char *buffer, size_t count)
 350 {
 351         struct dt_device *dt = container_of(kobj, struct dt_device,
 352                                             dd_kobj);
 353         struct osd_device *dev = osd_dt_dev(dt);
 354         int val;
 355         int rc;
 356
 357         LASSERT(dev);
 358         if (!dev->od_os)
 359                 return -EINPROGRESS;
 360
 361         rc = kstrtoint(buffer, 0, &val);
 362         if (rc)
 363                 return rc;
 364
 365         dev->od_index_backup_policy = val;
 366         return count;
 367 }
 368 LUSTRE_RW_ATTR(index_backup);
 369
 370 static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
 371 {
 372         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 373
 374         LASSERT(osd != NULL);
 375         if (unlikely(osd->od_os == NULL))
 376                 return -EINPROGRESS;
 377
 378         seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
 379         return 0;
 380 }
 381
 382 static ssize_t
 383 zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
 384                             size_t count, loff_t *off)
 385 {
 386         struct seq_file *m = file->private_data;
 387         struct dt_device *dt = m->private;
 388         struct osd_device *osd = osd_dt_dev(dt);
 389         char kernbuf[22] = "";
 390         u64 val;
 391         int rc;
 392
 393         LASSERT(osd != NULL);
 394         if (unlikely(osd->od_os == NULL))
 395                 return -EINPROGRESS;
 396
 397         if (count >= sizeof(kernbuf))
 398                 return -EINVAL;
 399
 400         if (copy_from_user(kernbuf, buffer, count))
 401                 return -EFAULT;
 402         kernbuf[count] = 0;
 403
 404         rc = sysfs_memparse(kernbuf, count, &val, "B");
 405         if (rc < 0)
 406                 return rc;
 407
 408         osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
 409                                          OSD_MAX_CACHE_SIZE : val;
 410         return count;
 411 }
 412 LDEBUGFS_SEQ_FOPS(zfs_osd_readcache);
 413
 414 static struct attribute *zfs_attrs[] = {
 415         &lustre_attr_fstype.attr,
 416         &lustre_attr_mntdev.attr,
 417         &lustre_attr_force_sync.attr,
 418         &lustre_attr_nonrotational.attr,
 419         &lustre_attr_index_backup.attr,
 420         &lustre_attr_auto_scrub.attr,
 421         NULL,
 422 };
 423
 424 struct ldebugfs_vars ldebugfs_osd_obd_vars[] = {
 425         { .name =       "oi_scrub",
 426           .fops =       &zfs_osd_oi_scrub_fops          },
 427         { .name =       "readcache_max_filesize",
 428           .fops =       &zfs_osd_readcache_fops         },
 429         { 0 }
 430 };
 431
 432 int osd_procfs_init(struct osd_device *osd, const char *name)
 433 {
 434         struct obd_type *type;
 435         int rc;
 436
 437         ENTRY;
 438
 439         /* at the moment there is no linkage between lu_type
 440          * and obd_type, so we lookup obd_type this way
 441          */
 442         type = class_search_type(LUSTRE_OSD_ZFS_NAME);
 443
 444         LASSERT(type);
 445         LASSERT(name);
 446
 447         /* put reference taken by class_search_type */
 448         kobject_put(&type->typ_kobj);
 449
 450         osd->od_dt_dev.dd_ktype.default_attrs = zfs_attrs;
 451         rc = dt_tunables_init(&osd->od_dt_dev, type, name,
 452                               ldebugfs_osd_obd_vars);
 453         if (rc) {
 454                 CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
 455                        name, rc);
 456                 GOTO(out, rc);
 457         }
 458
 459         if (osd->od_proc_entry)
 460                 RETURN(0);
 461
 462         osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
 463                                               NULL, &osd->od_dt_dev);
 464         if (IS_ERR(osd->od_proc_entry)) {
 465                 rc = PTR_ERR(osd->od_proc_entry);
 466                 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
 467                 osd->od_proc_entry = NULL;
 468                 GOTO(out, rc);
 469         }
 470
 471         rc = osd_stats_init(osd);
 472
 473         GOTO(out, rc);
 474 out:
 475         if (rc)
 476                 osd_procfs_fini(osd);
 477         return rc;
 478 }
 479
 480 int osd_procfs_fini(struct osd_device *osd)
 481 {
 482         ENTRY;
 483
 484         if (osd->od_stats)
 485                 lprocfs_free_stats(&osd->od_stats);
 486
 487         if (osd->od_proc_entry) {
 488                 lprocfs_remove(&osd->od_proc_entry);
 489                 osd->od_proc_entry = NULL;
 490         }
 491
 492         return dt_tunables_fini(&osd->od_dt_dev);
 493 }
 494
 495 #endif