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         /* this sampling races with updates */
  90         lprocfs_stats_header(seq, ktime_get(), brw_stats->bs_init, 25, ":", 1);
  91
  92         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
  93                           &brw_stats->bs_hist[BRW_R_PAGES],
  94                           &brw_stats->bs_hist[BRW_W_PAGES], 1);
  95
  96         display_brw_stats(seq, "discontiguous pages", "rpcs",
  97                           &brw_stats->bs_hist[BRW_R_DISCONT_PAGES],
  98                           &brw_stats->bs_hist[BRW_W_DISCONT_PAGES], 0);
  99 #if 0
 100         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 101                           &brw_stats->bs_hist[BRW_R_DISCONT_BLOCKS],
 102                           &brw_stats->bs_hist[BRW_W_DISCONT_BLOCKS], 0);
 103
 104         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 105                           &brw_stats->bs_hist[BRW_R_DIO_FRAGS],
 106                           &brw_stats->bs_hist[BRW_W_DIO_FRAGS], 0);
 107 #endif
 108         display_brw_stats(seq, "disk I/Os in flight", "ios",
 109                           &brw_stats->bs_hist[BRW_R_RPC_HIST],
 110                           &brw_stats->bs_hist[BRW_W_RPC_HIST], 0);
 111
 112         display_brw_stats(seq, "I/O time (1/1000s)", "ios",
 113                           &brw_stats->bs_hist[BRW_R_IO_TIME],
 114                           &brw_stats->bs_hist[BRW_W_IO_TIME], 1);
 115
 116         display_brw_stats(seq, "disk I/O size", "ios",
 117                           &brw_stats->bs_hist[BRW_R_DISK_IOSIZE],
 118                           &brw_stats->bs_hist[BRW_W_DISK_IOSIZE], 1);
 119 }
 120
 121 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
 122 {
 123         struct osd_device *osd = seq->private;
 124
 125         brw_stats_show(seq, &osd->od_brw_stats);
 126
 127         return 0;
 128 }
 129
 130 static ssize_t osd_brw_stats_seq_write(struct file *file,
 131                                        const char __user *buf,
 132                                        size_t len, loff_t *off)
 133 {
 134         struct seq_file *seq = file->private_data;
 135         struct osd_device *osd = seq->private;
 136         int i;
 137
 138         for (i = 0; i < BRW_LAST; i++)
 139                 lprocfs_oh_clear(&osd->od_brw_stats.bs_hist[i]);
 140
 141         return len;
 142 }
 143
 144 LPROC_SEQ_FOPS(osd_brw_stats);
 145
 146 static int osd_stats_init(struct osd_device *osd)
 147 {
 148         int result, i;
 149         ENTRY;
 150
 151         for (i = 0; i < BRW_LAST; i++)
 152                 spin_lock_init(&osd->od_brw_stats.bs_hist[i].oh_lock);
 153
 154         osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
 155         if (osd->od_stats != NULL) {
 156                 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
 157                                 osd->od_stats);
 158                 if (result)
 159                         GOTO(out, result);
 160
 161                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
 162                                 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
 163                                 "get_page", "usec");
 164                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
 165                                 LPROCFS_CNTR_AVGMINMAX,
 166                                 "get_page_failures", "num");
 167                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
 168                                 LPROCFS_CNTR_AVGMINMAX,
 169                                 "cache_access", "pages");
 170                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
 171                                 LPROCFS_CNTR_AVGMINMAX,
 172                                 "cache_hit", "pages");
 173                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
 174                                 LPROCFS_CNTR_AVGMINMAX,
 175                                 "cache_miss", "pages");
 176                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
 177                                 LPROCFS_CNTR_AVGMINMAX,
 178                                 "copy", "pages");
 179                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
 180                                 LPROCFS_CNTR_AVGMINMAX,
 181                                 "zerocopy", "pages");
 182                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
 183                                 LPROCFS_CNTR_AVGMINMAX,
 184                                 "tail", "pages");
 185 #ifdef OSD_THANDLE_STATS
 186                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
 187                                 LPROCFS_CNTR_AVGMINMAX,
 188                                 "thandle_starting", "usec");
 189                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
 190                                 LPROCFS_CNTR_AVGMINMAX,
 191                                 "thandle_open", "usec");
 192                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
 193                                 LPROCFS_CNTR_AVGMINMAX,
 194                                 "thandle_closing", "usec");
 195 #endif
 196                 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
 197                                             0644, &osd_brw_stats_fops, osd);
 198         } else {
 199                 result = -ENOMEM;
 200         }
 201
 202 out:
 203         RETURN(result);
 204 }
 205
 206 static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
 207 {
 208         struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 209
 210         LASSERT(dev != NULL);
 211         if (!dev->od_os)
 212                 return -EINPROGRESS;
 213
 214         scrub_dump(m, &dev->od_scrub);
 215         return 0;
 216 }
 217 LDEBUGFS_SEQ_FOPS_RO(zfs_osd_oi_scrub);
 218
 219 static ssize_t auto_scrub_show(struct kobject *kobj, struct attribute *attr,
 220                               char *buf)
 221 {
 222         struct dt_device *dt = container_of(kobj, struct dt_device,
 223                                             dd_kobj);
 224         struct osd_device *dev = osd_dt_dev(dt);
 225
 226         LASSERT(dev);
 227         if (!dev->od_os)
 228                 return -EINPROGRESS;
 229
 230         return scnprintf(buf, PAGE_SIZE, "%lld\n",
 231                          dev->od_scrub.os_auto_scrub_interval);
 232 }
 233
 234 static ssize_t auto_scrub_store(struct kobject *kobj, struct attribute *attr,
 235                                 const char *buffer, size_t count)
 236 {
 237         struct dt_device *dt = container_of(kobj, struct dt_device,
 238                                             dd_kobj);
 239         struct osd_device *dev = osd_dt_dev(dt);
 240         s64 val;
 241         int rc;
 242
 243         LASSERT(dev);
 244         if (!dev->od_os)
 245                 return -EINPROGRESS;
 246
 247         rc = kstrtoull(buffer, 0, &val);
 248         if (rc)
 249                 return rc;
 250
 251         dev->od_scrub.os_auto_scrub_interval = val;
 252         return count;
 253 }
 254 LUSTRE_RW_ATTR(auto_scrub);
 255
 256 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 257                           char *buf)
 258 {
 259         return sprintf(buf, "zfs\n");
 260 }
 261 LUSTRE_RO_ATTR(fstype);
 262
 263 static ssize_t mntdev_show(struct kobject *kobj, struct attribute *attr,
 264                            char *buf)
 265 {
 266         struct dt_device *dt = container_of(kobj, struct dt_device,
 267                                             dd_kobj);
 268         struct osd_device *osd = osd_dt_dev(dt);
 269
 270         LASSERT(osd);
 271
 272         return sprintf(buf, "%s\n", osd->od_mntdev);
 273 }
 274 LUSTRE_RO_ATTR(mntdev);
 275
 276 ssize_t force_sync_store(struct kobject *kobj, struct attribute *attr,
 277                          const char *buffer, size_t count)
 278 {
 279         struct dt_device *dt = container_of(kobj, struct dt_device,
 280                                             dd_kobj);
 281         struct lu_env env;
 282         int rc;
 283
 284         rc = lu_env_init(&env, LCT_LOCAL);
 285         if (rc)
 286                 return rc;
 287
 288         rc = dt_sync(&env, dt);
 289         lu_env_fini(&env);
 290
 291         return rc == 0 ? count : rc;
 292 }
 293 LUSTRE_WO_ATTR(force_sync);
 294
 295 static ssize_t sync_on_lseek_show(struct kobject *kobj, struct attribute *attr,
 296                                   char *buf)
 297 {
 298         struct dt_device *dt = container_of(kobj, struct dt_device, dd_kobj);
 299         struct osd_device *osd = osd_dt_dev(dt);
 300
 301         if (!osd->od_os)
 302                 return -EINPROGRESS;
 303
 304         return sprintf(buf, "%u\n", osd->od_sync_on_lseek);
 305 }
 306
 307 ssize_t sync_on_lseek_store(struct kobject *kobj, struct attribute *attr,
 308                             const char *buffer, size_t count)
 309 {
 310         struct dt_device *dt = container_of(kobj, struct dt_device, dd_kobj);
 311         struct osd_device *osd = osd_dt_dev(dt);
 312         bool val;
 313         int rc;
 314
 315         if (!osd->od_os)
 316                 return -EINPROGRESS;
 317
 318         rc = kstrtobool(buffer, &val);
 319         if (rc)
 320                 return rc;
 321
 322         osd->od_sync_on_lseek = !!val;
 323
 324         return count;
 325 }
 326 LUSTRE_RW_ATTR(sync_on_lseek);
 327
 328 static ssize_t nonrotational_show(struct kobject *kobj, struct attribute *attr,
 329                                   char *buf)
 330 {
 331         struct dt_device *dt = container_of(kobj, struct dt_device,
 332                                             dd_kobj);
 333         struct osd_device *osd = osd_dt_dev(dt);
 334
 335         LASSERT(osd);
 336         if (!osd->od_os)
 337                 return -EINPROGRESS;
 338
 339         return sprintf(buf, "%u\n", osd->od_nonrotational);
 340 }
 341
 342 static ssize_t nonrotational_store(struct kobject *kobj,
 343                                    struct attribute *attr, const char *buffer,
 344                                    size_t count)
 345 {
 346         struct dt_device *dt = container_of(kobj, struct dt_device,
 347                                             dd_kobj);
 348         struct osd_device *osd = osd_dt_dev(dt);
 349         bool val;
 350         int rc;
 351
 352         LASSERT(osd);
 353         if (!osd->od_os)
 354                 return -EINPROGRESS;
 355
 356         rc = kstrtobool(buffer, &val);
 357         if (rc)
 358                 return rc;
 359
 360         osd->od_nonrotational = val;
 361         return count;
 362 }
 363 LUSTRE_RW_ATTR(nonrotational);
 364
 365 static ssize_t index_backup_show(struct kobject *kobj, struct attribute *attr,
 366                                  char *buf)
 367 {
 368         struct dt_device *dt = container_of(kobj, struct dt_device,
 369                                             dd_kobj);
 370         struct osd_device *dev = osd_dt_dev(dt);
 371
 372         LASSERT(dev);
 373         if (!dev->od_os)
 374                 return -EINPROGRESS;
 375
 376         return sprintf(buf, "%d\n", dev->od_index_backup_policy);
 377 }
 378
 379 ssize_t index_backup_store(struct kobject *kobj, struct attribute *attr,
 380                            const char *buffer, size_t count)
 381 {
 382         struct dt_device *dt = container_of(kobj, struct dt_device,
 383                                             dd_kobj);
 384         struct osd_device *dev = osd_dt_dev(dt);
 385         int val;
 386         int rc;
 387
 388         LASSERT(dev);
 389         if (!dev->od_os)
 390                 return -EINPROGRESS;
 391
 392         rc = kstrtoint(buffer, 0, &val);
 393         if (rc)
 394                 return rc;
 395
 396         dev->od_index_backup_policy = val;
 397         return count;
 398 }
 399 LUSTRE_RW_ATTR(index_backup);
 400
 401 static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
 402 {
 403         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 404
 405         LASSERT(osd != NULL);
 406         if (unlikely(osd->od_os == NULL))
 407                 return -EINPROGRESS;
 408
 409         seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
 410         return 0;
 411 }
 412
 413 static ssize_t
 414 zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
 415                             size_t count, loff_t *off)
 416 {
 417         struct seq_file *m = file->private_data;
 418         struct dt_device *dt = m->private;
 419         struct osd_device *osd = osd_dt_dev(dt);
 420         char kernbuf[22] = "";
 421         u64 val;
 422         int rc;
 423
 424         LASSERT(osd != NULL);
 425         if (unlikely(osd->od_os == NULL))
 426                 return -EINPROGRESS;
 427
 428         if (count >= sizeof(kernbuf))
 429                 return -EINVAL;
 430
 431         if (copy_from_user(kernbuf, buffer, count))
 432                 return -EFAULT;
 433         kernbuf[count] = 0;
 434
 435         rc = sysfs_memparse(kernbuf, count, &val, "B");
 436         if (rc < 0)
 437                 return rc;
 438
 439         osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
 440                                          OSD_MAX_CACHE_SIZE : val;
 441         return count;
 442 }
 443 LDEBUGFS_SEQ_FOPS(zfs_osd_readcache);
 444
 445 static struct attribute *zfs_attrs[] = {
 446         &lustre_attr_fstype.attr,
 447         &lustre_attr_mntdev.attr,
 448         &lustre_attr_force_sync.attr,
 449         &lustre_attr_nonrotational.attr,
 450         &lustre_attr_index_backup.attr,
 451         &lustre_attr_auto_scrub.attr,
 452         &lustre_attr_sync_on_lseek.attr,
 453         NULL,
 454 };
 455
 456 struct ldebugfs_vars ldebugfs_osd_obd_vars[] = {
 457         { .name =       "oi_scrub",
 458           .fops =       &zfs_osd_oi_scrub_fops          },
 459         { .name =       "readcache_max_filesize",
 460           .fops =       &zfs_osd_readcache_fops         },
 461         { 0 }
 462 };
 463
 464 int osd_procfs_init(struct osd_device *osd, const char *name)
 465 {
 466         struct obd_type *type;
 467         int rc;
 468
 469         ENTRY;
 470
 471         /* at the moment there is no linkage between lu_type
 472          * and obd_type, so we lookup obd_type this way
 473          */
 474         type = class_search_type(LUSTRE_OSD_ZFS_NAME);
 475
 476         LASSERT(type);
 477         LASSERT(name);
 478
 479         /* put reference taken by class_search_type */
 480         kobject_put(&type->typ_kobj);
 481
 482         osd->od_dt_dev.dd_ktype.default_attrs = zfs_attrs;
 483         rc = dt_tunables_init(&osd->od_dt_dev, type, name,
 484                               ldebugfs_osd_obd_vars);
 485         if (rc) {
 486                 CERROR("%s: cannot setup sysfs / debugfs entry: %d\n",
 487                        name, rc);
 488                 GOTO(out, rc);
 489         }
 490
 491         if (osd->od_proc_entry)
 492                 RETURN(0);
 493
 494         osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
 495                                               NULL, &osd->od_dt_dev);
 496         if (IS_ERR(osd->od_proc_entry)) {
 497                 rc = PTR_ERR(osd->od_proc_entry);
 498                 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
 499                 osd->od_proc_entry = NULL;
 500                 GOTO(out, rc);
 501         }
 502
 503         rc = osd_stats_init(osd);
 504
 505         GOTO(out, rc);
 506 out:
 507         if (rc)
 508                 osd_procfs_fini(osd);
 509         return rc;
 510 }
 511
 512 int osd_procfs_fini(struct osd_device *osd)
 513 {
 514         ENTRY;
 515
 516         if (osd->od_stats)
 517                 lprocfs_free_stats(&osd->od_stats);
 518
 519         if (osd->od_proc_entry) {
 520                 lprocfs_remove(&osd->od_proc_entry);
 521                 osd->od_proc_entry = NULL;
 522         }
 523
 524         return dt_tunables_fini(&osd->od_dt_dev);
 525 }
 526
 527 #endif