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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2012, 2014, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/osd-zfs/osd_lproc.c
  37  *
  38  * Author: Alex Zhuravlev <bzzz@whamcloud.com>
  39  * Author: Mike Pershin <tappro@whamcloud.com>
  40  */
  41
  42 #define DEBUG_SUBSYSTEM S_OSD
  43
  44 #include <obd.h>
  45 #include <obd_class.h>
  46 #include <lprocfs_status.h>
  47 #include <lustre/lustre_idl.h>
  48
  49 #include "osd_internal.h"
  50
  51 #ifdef LPROCFS
  52
  53 #define pct(a, b) (b ? a * 100 / b : 0)
  54
  55 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
  56                               struct obd_histogram *read,
  57                               struct obd_histogram *write, int scale)
  58 {
  59         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
  60         int i;
  61
  62         seq_printf(seq, "\n%26s read      |     write\n", " ");
  63         seq_printf(seq, "%-22s %-5s %% cum %% |  %-11s %% cum %%\n",
  64                    name, units, units);
  65
  66         read_tot = lprocfs_oh_sum(read);
  67         write_tot = lprocfs_oh_sum(write);
  68         for (i = 0; i < OBD_HIST_MAX; i++) {
  69                 r = read->oh_buckets[i];
  70                 w = write->oh_buckets[i];
  71                 read_cum += r;
  72                 write_cum += w;
  73                 if (read_cum == 0 && write_cum == 0)
  74                         continue;
  75
  76                 if (!scale)
  77                         seq_printf(seq, "%u", i);
  78                 else if (i < 10)
  79                         seq_printf(seq, "%u", scale << i);
  80                 else if (i < 20)
  81                         seq_printf(seq, "%uK", scale << (i-10));
  82                 else
  83                         seq_printf(seq, "%uM", scale << (i-20));
  84
  85                 seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
  86                            r, pct(r, read_tot), pct(read_cum, read_tot),
  87                            w, pct(w, write_tot), pct(write_cum, write_tot));
  88
  89                 if (read_cum == read_tot && write_cum == write_tot)
  90                         break;
  91         }
  92 }
  93
  94 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
  95 {
  96         struct timeval now;
  97
  98         /* this sampling races with updates */
  99         do_gettimeofday(&now);
 100         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
 101                    now.tv_sec, now.tv_usec);
 102
 103         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
 104                           &brw_stats->hist[BRW_R_PAGES],
 105                           &brw_stats->hist[BRW_W_PAGES], 1);
 106         display_brw_stats(seq, "discontiguous pages", "rpcs",
 107                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
 108                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
 109 #if 0
 110         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 111                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
 112                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
 113
 114         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 115                           &brw_stats->hist[BRW_R_DIO_FRAGS],
 116                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
 117 #endif
 118         display_brw_stats(seq, "disk I/Os in flight", "ios",
 119                           &brw_stats->hist[BRW_R_RPC_HIST],
 120                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 121
 122         display_brw_stats(seq, "I/O time (1/1000s)", "ios",
 123                           &brw_stats->hist[BRW_R_IO_TIME],
 124                           &brw_stats->hist[BRW_W_IO_TIME], 1000 / HZ);
 125
 126         display_brw_stats(seq, "disk I/O size", "ios",
 127                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
 128                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 129 }
 130
 131 #undef pct
 132
 133 static int osd_brw_stats_seq_show(struct seq_file *seq, void *v)
 134 {
 135         struct osd_device *osd = seq->private;
 136
 137         brw_stats_show(seq, &osd->od_brw_stats);
 138
 139         return 0;
 140 }
 141
 142 static ssize_t osd_brw_stats_seq_write(struct file *file, const char *buf,
 143                                        size_t len, loff_t *off)
 144 {
 145         struct seq_file *seq = file->private_data;
 146         struct osd_device *osd = seq->private;
 147         int i;
 148
 149         for (i = 0; i < BRW_LAST; i++)
 150                 lprocfs_oh_clear(&osd->od_brw_stats.hist[i]);
 151
 152         return len;
 153 }
 154
 155 LPROC_SEQ_FOPS(osd_brw_stats);
 156
 157 static int osd_stats_init(struct osd_device *osd)
 158 {
 159         int result;
 160         ENTRY;
 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         return seq_printf(m, "zfs\n");
 217 }
 218 LPROC_SEQ_FOPS_RO(zfs_osd_fstype);
 219
 220 static int zfs_osd_mntdev_seq_show(struct seq_file *m, void *data)
 221 {
 222         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 223
 224         LASSERT(osd != NULL);
 225         return seq_printf(m, "%s\n", osd->od_mntdev);
 226 }
 227 LPROC_SEQ_FOPS_RO(zfs_osd_mntdev);
 228
 229 static ssize_t
 230 lprocfs_osd_force_sync_seq_write(struct file *file, const char __user *buffer,
 231                                 size_t count, loff_t *off)
 232 {
 233         struct seq_file   *m = file->private_data;
 234         struct dt_device  *dt = m->private;
 235         struct lu_env      env;
 236         int rc;
 237
 238         rc = lu_env_init(&env, LCT_LOCAL);
 239         if (rc)
 240                 return rc;
 241         rc = dt_sync(&env, dt);
 242         lu_env_fini(&env);
 243
 244         return rc == 0 ? count : rc;
 245 }
 246 LPROC_SEQ_FOPS_WO_TYPE(zfs, osd_force_sync);
 247
 248 static int zfs_osd_iused_est_seq_show(struct seq_file *m, void *data)
 249 {
 250         struct osd_device *osd = osd_dt_dev((struct dt_device *)m->private);
 251         LASSERT(osd != NULL);
 252
 253         return seq_printf(m, "%d\n", osd->od_quota_iused_est);
 254 }
 255
 256 static ssize_t
 257 zfs_osd_iused_est_seq_write(struct file *file, const char __user *buffer,
 258                              size_t count, loff_t *off)
 259 {
 260         struct seq_file   *m = file->private_data;
 261         struct dt_device  *dt = m->private;
 262         struct osd_device *osd = osd_dt_dev(dt);
 263         int                rc, val;
 264
 265         LASSERT(osd != NULL);
 266
 267         rc = lprocfs_write_helper(buffer, count, &val);
 268         if (rc)
 269                 return rc;
 270
 271         osd->od_quota_iused_est = !!val;
 272
 273         return count;
 274 }
 275 LPROC_SEQ_FOPS(zfs_osd_iused_est);
 276
 277 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_blksize);
 278 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytestotal);
 279 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesfree);
 280 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_kbytesavail);
 281 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filestotal);
 282 LPROC_SEQ_FOPS_RO_TYPE(zfs, dt_filesfree);
 283
 284 struct lprocfs_seq_vars lprocfs_osd_obd_vars[] = {
 285         { .name =       "blocksize",
 286           .fops =       &zfs_dt_blksize_fops            },
 287         { .name =       "kbytestotal",
 288           .fops =       &zfs_dt_kbytestotal_fops        },
 289         { .name =       "kbytesfree",
 290           .fops =       &zfs_dt_kbytesfree_fops         },
 291         { .name =       "kbytesavail",
 292           .fops =       &zfs_dt_kbytesavail_fops        },
 293         { .name =       "filestotal",
 294           .fops =       &zfs_dt_filestotal_fops         },
 295         { .name =       "filesfree",
 296           .fops =       &zfs_dt_filesfree_fops          },
 297         { .name =       "fstype",
 298           .fops =       &zfs_osd_fstype_fops            },
 299         { .name =       "mntdev",
 300           .fops =       &zfs_osd_mntdev_fops            },
 301         { .name =       "force_sync",
 302           .fops =       &zfs_osd_force_sync_fops        },
 303         { .name =       "quota_iused_estimate",
 304           .fops =       &zfs_osd_iused_est_fops         },
 305         { 0 }
 306 };
 307
 308 int osd_procfs_init(struct osd_device *osd, const char *name)
 309 {
 310         struct obd_type *type;
 311         int              rc;
 312         ENTRY;
 313
 314         if (osd->od_proc_entry)
 315                 RETURN(0);
 316
 317         /* at the moment there is no linkage between lu_type
 318          * and obd_type, so we lookup obd_type this way */
 319         type = class_search_type(LUSTRE_OSD_ZFS_NAME);
 320
 321         LASSERT(name != NULL);
 322         LASSERT(type != NULL);
 323
 324         osd->od_proc_entry = lprocfs_seq_register(name, type->typ_procroot,
 325                         lprocfs_osd_obd_vars, &osd->od_dt_dev);
 326         if (IS_ERR(osd->od_proc_entry)) {
 327                 rc = PTR_ERR(osd->od_proc_entry);
 328                 CERROR("Error %d setting up lprocfs for %s\n", rc, name);
 329                 osd->od_proc_entry = NULL;
 330                 GOTO(out, rc);
 331         }
 332
 333         rc = osd_stats_init(osd);
 334
 335         GOTO(out, rc);
 336 out:
 337         if (rc)
 338                 osd_procfs_fini(osd);
 339         return rc;
 340 }
 341
 342 int osd_procfs_fini(struct osd_device *osd)
 343 {
 344         ENTRY;
 345
 346         if (osd->od_stats)
 347                 lprocfs_free_stats(&osd->od_stats);
 348
 349         if (osd->od_proc_entry) {
 350                 lprocfs_remove(&osd->od_proc_entry);
 351                 osd->od_proc_entry = NULL;
 352         }
 353
 354         RETURN(0);
 355 }
 356
 357 #endif