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