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 #define pct(a, b) (b ? a * 100 / b : 0)
  50
  51 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
  52                               struct obd_histogram *read,
  53                               struct obd_histogram *write, int scale)
  54 {
  55         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
  56         int i;
  57
  58         seq_printf(seq, "\n%26s read      |     write\n", " ");
  59         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
  60                    name, units, units);
  61
  62         read_tot = lprocfs_oh_sum(read);
  63         write_tot = lprocfs_oh_sum(write);
  64         for (i = 0; i < OBD_HIST_MAX; i++) {
  65                 r = read->oh_buckets[i];
  66                 w = write->oh_buckets[i];
  67                 read_cum += r;
  68                 write_cum += w;
  69                 if (read_cum == 0 && write_cum == 0)
  70                         continue;
  71
  72                 if (!scale)
  73                         seq_printf(seq, "%u", i);
  74                 else if (i < 10)
  75                         seq_printf(seq, "%u", scale << i);
  76                 else if (i < 20)
  77                         seq_printf(seq, "%uK", scale << (i-10));
  78                 else
  79                         seq_printf(seq, "%uM", scale << (i-20));
  80
  81                 seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
  82                            r, pct(r, read_tot), pct(read_cum, read_tot),
  83                            w, pct(w, write_tot), pct(write_cum, write_tot));
  84
  85                 if (read_cum == read_tot && write_cum == write_tot)
  86                         break;
  87         }
  88 }
  89
  90 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
  91 {
  92         struct timespec64 now;
  93
  94         /* this sampling races with updates */
  95         ktime_get_real_ts64(&now);
  96         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
  97                    (s64)now.tv_sec, now.tv_nsec);
  98
  99         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
 100                           &brw_stats->hist[BRW_R_PAGES],
 101                           &brw_stats->hist[BRW_W_PAGES], 1);
 102         display_brw_stats(seq, "discontiguous pages", "rpcs",
 103                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
 104                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
 105 #if 0
 106         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 107                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
 108                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
 109
 110         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 111                           &brw_stats->hist[BRW_R_DIO_FRAGS],
 112                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
 113 #endif
 114         display_brw_stats(seq, "disk I/Os in flight", "ios",
 115                           &brw_stats->hist[BRW_R_RPC_HIST],
 116                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 117
 118         display_brw_stats(seq, "I/O time (1/1000s)", "ios",
 119                           &brw_stats->hist[BRW_R_IO_TIME],
 120                           &brw_stats->hist[BRW_W_IO_TIME], 1000 / HZ);
 121
 122         display_brw_stats(seq, "disk I/O size", "ios",
 123                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
 124                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 125 }
 126
 127 #undef pct
 128
 129 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
 130 {
 131         struct osd_device *osd = seq->private;
 132
 133         brw_stats_show(seq, &osd->od_brw_stats);
 134
 135         return 0;
 136 }
 137
 138 static ssize_t osd_brw_stats_seq_write(struct file *file,
 139                                        const char __user *buf,
 140                                        size_t len, loff_t *off)
 141 {
 142         struct seq_file *seq = file->private_data;
 143         struct osd_device *osd = seq->private;
 144         int i;
 145
 146         for (i = 0; i < BRW_LAST; i++)
 147                 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
 148
 149         return len;
 150 }
 151
 152 LPROC_SEQ_FOPS(osd_brw_stats);
 153
 154 static int osd_stats_init(struct osd_device *osd)
 155 {
 156         int result, i;
 157         ENTRY;
 158
 159         for (i = 0; i < BRW_LAST; i++)
 160                 spin_lock_init(&osd->od_brw_stats.hist[i].oh_lock);
 161
 162         osd->od_stats = lprocfs_alloc_stats(LPROC_OSD_LAST, 0);
 163         if (osd->od_stats != NULL) {
 164                 result = lprocfs_register_stats(osd->od_proc_entry, "stats",
 165                                 osd->od_stats);
 166                 if (result)
 167                         GOTO(out, result);
 168
 169                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_GET_PAGE,
 170                                 LPROCFS_CNTR_AVGMINMAX|LPROCFS_CNTR_STDDEV,
 171                                 "get_page", "usec");
 172                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_NO_PAGE,
 173                                 LPROCFS_CNTR_AVGMINMAX,
 174                                 "get_page_failures", "num");
 175                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_ACCESS,
 176                                 LPROCFS_CNTR_AVGMINMAX,
 177                                 "cache_access", "pages");
 178                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_HIT,
 179                                 LPROCFS_CNTR_AVGMINMAX,
 180                                 "cache_hit", "pages");
 181                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_CACHE_MISS,
 182                                 LPROCFS_CNTR_AVGMINMAX,
 183                                 "cache_miss", "pages");
 184                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_COPY_IO,
 185                                 LPROCFS_CNTR_AVGMINMAX,
 186                                 "copy", "pages");
 187                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_ZEROCOPY_IO,
 188                                 LPROCFS_CNTR_AVGMINMAX,
 189                                 "zerocopy", "pages");
 190                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_TAIL_IO,
 191                                 LPROCFS_CNTR_AVGMINMAX,
 192                                 "tail", "pages");
 193 #ifdef OSD_THANDLE_STATS
 194                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_STARTING,
 195                                 LPROCFS_CNTR_AVGMINMAX,
 196                                 "thandle_starting", "usec");
 197                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_OPEN,
 198                                 LPROCFS_CNTR_AVGMINMAX,
 199                                 "thandle_open", "usec");
 200                 lprocfs_counter_init(osd->od_stats, LPROC_OSD_THANDLE_CLOSING,
 201                                 LPROCFS_CNTR_AVGMINMAX,
 202                                 "thandle_closing", "usec");
 203 #endif
 204                 result = lprocfs_seq_create(osd->od_proc_entry, "brw_stats",
 205                                             0644, &osd_brw_stats_fops, osd);
 206         } else {
 207                 result = -ENOMEM;
 208         }
 209
 210 out:
 211         RETURN(result);
 212 }
 213
 214 static int zfs_osd_auto_scrub_seq_show(struct seq_file *m, void *data)
 215 {
 216         struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 217
 218         LASSERT(dev != NULL);
 219         if (!dev->od_os)
 220                 return -EINPROGRESS;
 221
 222         seq_printf(m, "%lld\n", dev->od_auto_scrub_interval);
 223         return 0;
 224 }
 225
 226 static ssize_t
 227 zfs_osd_auto_scrub_seq_write(struct file *file, const char __user *buffer,
 228                              size_t count, loff_t *off)
 229 {
 230         struct seq_file *m = file->private_data;
 231         struct dt_device *dt = m->private;
 232         struct osd_device *dev = osd_dt_dev(dt);
 233         int rc;
 234         __s64 val;
 235
 236         LASSERT(dev != NULL);
 237         if (!dev->od_os)
 238                 return -EINPROGRESS;
 239
 240         rc = kstrtoull_from_user(buffer, count, 0, &val);
 241         if (rc)
 242                 return rc;
 243
 244         dev->od_auto_scrub_interval = val;
 245         return count;
 246 }
 247 LPROC_SEQ_FOPS(zfs_osd_auto_scrub);
 248
 249 static int zfs_osd_oi_scrub_seq_show(struct seq_file *m, void *data)
 250 {
 251         struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 252
 253         LASSERT(dev != NULL);
 254         if (!dev->od_os)
 255                 return -EINPROGRESS;
 256
 257         scrub_dump(m, &dev->od_scrub);
 258         return 0;
 259 }
 260 LPROC_SEQ_FOPS_RO(zfs_osd_oi_scrub);
 261
 262 static int zfs_osd_fstype_seq_show(struct seq_file *m, void *data)
 263 {
 264         seq_puts(m, "zfs\n");
 265         return 0;
 266 }
 267 LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
 268
 269 static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
 270 {
 271         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 272
 273         LASSERT(osd != NULL);
 274         seq_printf(m, "%s\n", osd->od_mntdev);
 275         return 0;
 276 }
 277 LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
 278
 279 static ssize_t
 280 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
 281                                 size_t count, loff_t *off)
 282 {
 283         struct seq_file   *m = file->private_data;
 284         struct dt_device  *dt = m->private;
 285         struct lu_env      env;
 286         int rc;
 287
 288         rc = lu_env_init(&env, LCT_LOCAL);
 289         if (rc)
 290                 return rc;
 291         rc = dt_sync(&env, dt);
 292         lu_env_fini(&env);
 293
 294         return rc == 0 ? count : rc;
 295 }
 296 LPROC_SEQ_FOPS_WR_ONLY(zfs, osd_force_sync);
 297
 298 static int zfs_osd_index_backup_seq_show(struct seq_file *m, void *data)
 299 {
 300         struct osd_device *dev = osd_dt_dev((struct dt_device *)m->private);
 301
 302         LASSERT(dev != NULL);
 303         if (!dev->od_os)
 304                 return -EINPROGRESS;
 305
 306         seq_printf(m, "%d\n", dev->od_index_backup_policy);
 307         return 0;
 308 }
 309
 310 static ssize_t zfs_osd_index_backup_seq_write(struct file *file,
 311                                               const char __user *buffer,
 312                                               size_t count, loff_t *off)
 313 {
 314         struct seq_file *m = file->private_data;
 315         struct dt_device *dt = m->private;
 316         struct osd_device *dev = osd_dt_dev(dt);
 317         int val;
 318         int rc;
 319
 320         LASSERT(dev != NULL);
 321         if (!dev->od_os)
 322                 return -EINPROGRESS;
 323
 324         rc = kstrtoint_from_user(buffer, count, 0, &val);
 325         if (rc)
 326                 return rc;
 327
 328         dev->od_index_backup_policy = val;
 329         return count;
 330 }
 331 LPROC_SEQ_FOPS(zfs_osd_index_backup);
 332
 333 static int zfs_osd_readcache_seq_show(struct seq_file *m, void *data)
 334 {
 335         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 336
 337         LASSERT(osd != NULL);
 338         if (unlikely(osd->od_os == NULL))
 339                 return -EINPROGRESS;
 340
 341         seq_printf(m, "%llu\n", osd->od_readcache_max_filesize);
 342         return 0;
 343 }
 344
 345 static ssize_t
 346 zfs_osd_readcache_seq_write(struct file *file, const char __user *buffer,
 347                             size_t count, loff_t *off)
 348 {
 349         struct seq_file *m = file->private_data;
 350         struct dt_device *dt = m->private;
 351         struct osd_device *osd = osd_dt_dev(dt);
 352         s64 val;
 353         int rc;
 354
 355         LASSERT(osd != NULL);
 356         if (unlikely(osd->od_os == NULL))
 357                 return -EINPROGRESS;
 358
 359         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
 360         if (rc)
 361                 return rc;
 362         if (val < 0)
 363                 return -ERANGE;
 364
 365         osd->od_readcache_max_filesize = val > OSD_MAX_CACHE_SIZE ?
 366                                          OSD_MAX_CACHE_SIZE : val;
 367         return count;
 368 }
 369 LPROC_SEQ_FOPS(zfs_osd_readcache);
 370
 371 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
 372 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
 373 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
 374 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
 375 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
 376 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
 377
 378 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
 379         { .name =       "blocksize",
 380           .fops =       &zfs_dt_blksize_fops            },
 381         { .name =       "kbytestotal",
 382           .fops =       &zfs_dt_kbytestotal_fops        },
 383         { .name =       "kbytesfree",
 384           .fops =       &zfs_dt_kbytesfree_fops         },
 385         { .name =       "kbytesavail",
 386           .fops =       &zfs_dt_kbytesavail_fops        },
 387         { .name =       "filestotal",
 388           .fops =       &zfs_dt_filestotal_fops         },
 389         { .name =       "filesfree",
 390           .fops =       &zfs_dt_filesfree_fops          },
 391         { .name =       "auto_scrub",
 392           .fops =       &zfs_osd_auto_scrub_fops        },
 393         { .name =       "oi_scrub",
 394           .fops =       &zfs_osd_oi_scrub_fops          },
 395         { .name =       "fstype",
 396           .fops =       &zfs_osd_fstype_fops            },
 397         { .name =       "mntdev",
 398           .fops =       &zfs_osd_mntdev_fops            },
 399         { .name =       "force_sync",
 400           .fops =       &zfs_osd_force_sync_fops        },
 401         { .name =       "index_backup",
 402           .fops =       &zfs_osd_index_backup_fops      },
 403         { .name =       "readcache_max_filesize",
 404           .fops =       &zfs_osd_readcache_fops },
 405         { 0 }
 406 };
 407
 408 int osd_procfs_init(struct osd_device *osd, const char *name)
 409 {
 410         struct obd_type *type;
 411         int              rc;
 412         ENTRY;
 413
 414         if (osd->od_proc_entry)
 415                 RETURN(0);
 416
 417         /* at the moment there is no linkage between lu_type
 418          * and obd_type, so we lookup obd_type this way */
 419         type = class_search_type(LUSTRE_OSD_ZFS_NAME);
 420
 421         LASSERT(name != NULL);
 422         LASSERT(type != NULL);
 423
 424         osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
 425                                               lprocfs_osd_obd_vars,
 426                                               &osd->od_dt_dev);
 427         if (IS_ERR(osd->od_proc_entry)) {
 428                 rc = PTR_ERR(osd->od_proc_entry);
 429                 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
 430                 osd->od_proc_entry = NULL;
 431                 GOTO(out, rc);
 432         }
 433
 434         rc = osd_stats_init(osd);
 435
 436         GOTO(out, rc);
 437 out:
 438         if (rc)
 439                 osd_procfs_fini(osd);
 440         return rc;
 441 }
 442
 443 int osd_procfs_fini(struct osd_device *osd)
 444 {
 445         ENTRY;
 446
 447         if (osd->od_stats)
 448                 lprocfs_free_stats(&osd->od_stats);
 449
 450         if (osd->od_proc_entry) {
 451                 lprocfs_remove(&osd->od_proc_entry);
 452                 osd->od_proc_entry = NULL;
 453         }
 454
 455         RETURN(0);
 456 }
 457
 458 #endif