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, 2016, 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/lustre_idl.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 timeval now;
  93
  94         /* this sampling races with updates */
  95         do_gettimeofday(&now);
  96         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
  97                    now.tv_sec, now.tv_usec);
  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_fstype_seq_show(struct seq_file *m, void *data)
 215 {
 216         seq_puts(m, "zfs\n");
 217         return 0;
 218 }
 219 LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
 220
 221 static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
 222 {
 223         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 224
 225         LASSERT(osd != NULL);
 226         seq_printf(m, "%s\n", osd->od_mntdev);
 227         return 0;
 228 }
 229 LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
 230
 231 static ssize_t
 232 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
 233                                 size_t count, loff_t *off)
 234 {
 235         struct seq_file   *m = file->private_data;
 236         struct dt_device  *dt = m->private;
 237         struct lu_env      env;
 238         int rc;
 239
 240         rc = lu_env_init(&env, LCT_LOCAL);
 241         if (rc)
 242                 return rc;
 243         rc = dt_sync(&env, dt);
 244         lu_env_fini(&env);
 245
 246         return rc == 0 ? count : rc;
 247 }
 248 LPROC_SEQ_FOPS_WO_TYPE(zfs, osd_force_sync);
 249
 250 static int zfs_osd_iused_est_seq_show(struct seq_file *m, void *data)
 251 {
 252         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 253         LASSERT(osd != NULL);
 254
 255         seq_printf(m, "%d\n", osd->od_quota_iused_est);
 256         return 0;
 257 }
 258
 259 static ssize_t
 260 zfs_osd_iused_est_seq_write(struct file *file, const char __user *buffer,
 261                              size_t count, loff_t *off)
 262 {
 263         struct seq_file *m = file->private_data;
 264         struct dt_device *dt = m->private;
 265         struct osd_device *osd = osd_dt_dev(dt);
 266         int rc;
 267         __s64 val;
 268
 269         LASSERT(osd != NULL);
 270
 271         rc = lprocfs_str_to_s64(buffer, count, &val);
 272         if (rc)
 273                 return rc;
 274
 275         osd->od_quota_iused_est = !!val;
 276
 277         return count;
 278 }
 279 LPROC_SEQ_FOPS(zfs_osd_iused_est);
 280
 281 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
 282 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
 283 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
 284 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
 285 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
 286 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
 287
 288 struct lprocfs_vars lprocfs_osd_obd_vars[] = {
 289         { .name =       "blocksize",
 290           .fops =       &zfs_dt_blksize_fops            },
 291         { .name =       "kbytestotal",
 292           .fops =       &zfs_dt_kbytestotal_fops        },
 293         { .name =       "kbytesfree",
 294           .fops =       &zfs_dt_kbytesfree_fops         },
 295         { .name =       "kbytesavail",
 296           .fops =       &zfs_dt_kbytesavail_fops        },
 297         { .name =       "filestotal",
 298           .fops =       &zfs_dt_filestotal_fops         },
 299         { .name =       "filesfree",
 300           .fops =       &zfs_dt_filesfree_fops          },
 301         { .name =       "fstype",
 302           .fops =       &zfs_osd_fstype_fops            },
 303         { .name =       "mntdev",
 304           .fops =       &zfs_osd_mntdev_fops            },
 305         { .name =       "force_sync",
 306           .fops =       &zfs_osd_force_sync_fops        },
 307         { .name =       "quota_iused_estimate",
 308           .fops =       &zfs_osd_iused_est_fops         },
 309         { 0 }
 310 };
 311
 312 int osd_procfs_init(struct osd_device *osd, const char *name)
 313 {
 314         struct obd_type *type;
 315         int              rc;
 316         ENTRY;
 317
 318         if (osd->od_proc_entry)
 319                 RETURN(0);
 320
 321         /* at the moment there is no linkage between lu_type
 322          * and obd_type, so we lookup obd_type this way */
 323         type = class_search_type(LUSTRE_OSD_ZFS_NAME);
 324
 325         LASSERT(name != NULL);
 326         LASSERT(type != NULL);
 327
 328         osd->od_proc_entry = lprocfs_register(name, type->typ_procroot,
 329                                               lprocfs_osd_obd_vars,
 330                                               &osd->od_dt_dev);
 331         if (IS_ERR(osd->od_proc_entry)) {
 332                 rc = PTR_ERR(osd->od_proc_entry);
 333                 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
 334                 osd->od_proc_entry = NULL;
 335                 GOTO(out, rc);
 336         }
 337
 338         rc = osd_stats_init(osd);
 339
 340         GOTO(out, rc);
 341 out:
 342         if (rc)
 343                 osd_procfs_fini(osd);
 344         return rc;
 345 }
 346
 347 int osd_procfs_fini(struct osd_device *osd)
 348 {
 349         ENTRY;
 350
 351         if (osd->od_stats)
 352                 lprocfs_free_stats(&osd->od_stats);
 353
 354         if (osd->od_proc_entry) {
 355                 lprocfs_remove(&osd->od_proc_entry);
 356                 osd->od_proc_entry = NULL;
 357         }
 358
 359         RETURN(0);
 360 }
 361
 362 #endif