lustre/obdfilter/lproc_obdfilter.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  *  Copyright (C) 2002, 2003 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of the Lustre file system, http://www.lustre.org
   7  *   Lustre is a trademark of Cluster File Systems, Inc.
   8  *
   9  *   You may have signed or agreed to another license before downloading
  10  *   this software.  If so, you are bound by the terms and conditions
  11  *   of that agreement, and the following does not apply to you.  See the
  12  *   LICENSE file included with this distribution for more information.
  13  *
  14  *   If you did not agree to a different license, then this copy of Lustre
  15  *   is open source software; you can redistribute it and/or modify it
  16  *   under the terms of version 2 of the GNU General Public License as
  17  *   published by the Free Software Foundation.
  18  *
  19  *   In either case, Lustre is distributed in the hope that it will be
  20  *   useful, but WITHOUT ANY WARRANTY; without even the implied warranty
  21  *   of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  22  *   license text for more details.
  23  *
  24  */
  25 #define DEBUG_SUBSYSTEM S_CLASS
  26
  27 #include <linux/version.h>
  28 #include <lprocfs_status.h>
  29 #include <obd.h>
  30 #include <linux/seq_file.h>
  31 #include <linux/version.h>
  32
  33 #include "filter_internal.h"
  34
  35 #ifdef LPROCFS
  36 static int lprocfs_filter_rd_groups(char *page, char **start, off_t off,
  37                                     int count, int *eof, void *data)
  38 {
  39         struct obd_device *obd = (struct obd_device *)data;
  40         *eof = 1;
  41         return snprintf(page, count, "%u\n", obd->u.filter.fo_group_count);
  42 }
  43
  44 static int lprocfs_filter_rd_tot_dirty(char *page, char **start, off_t off,
  45                                        int count, int *eof, void *data)
  46 {
  47         struct obd_device *obd = (struct obd_device *)data;
  48
  49         LASSERT(obd != NULL);
  50         *eof = 1;
  51         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_dirty);
  52 }
  53
  54 static int lprocfs_filter_rd_tot_granted(char *page, char **start, off_t off,
  55                                          int count, int *eof, void *data)
  56 {
  57         struct obd_device *obd = (struct obd_device *)data;
  58
  59         LASSERT(obd != NULL);
  60         *eof = 1;
  61         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_granted);
  62 }
  63
  64 static int lprocfs_filter_rd_tot_pending(char *page, char **start, off_t off,
  65                                          int count, int *eof, void *data)
  66 {
  67         struct obd_device *obd = (struct obd_device *)data;
  68
  69         LASSERT(obd != NULL);
  70         *eof = 1;
  71         return snprintf(page, count, LPU64"\n", obd->u.filter.fo_tot_pending);
  72 }
  73
  74 static int lprocfs_filter_rd_mntdev(char *page, char **start, off_t off,
  75                                     int count, int *eof, void *data)
  76 {
  77         struct obd_device *obd = (struct obd_device *)data;
  78
  79         LASSERT(obd != NULL);
  80         LASSERT(obd->u.filter.fo_vfsmnt->mnt_devname);
  81         *eof = 1;
  82         return snprintf(page, count, "%s\n",
  83                         obd->u.filter.fo_vfsmnt->mnt_devname);
  84 }
  85
  86 static int lprocfs_filter_rd_last_id(char *page, char **start, off_t off,
  87                                      int count, int *eof, void *data)
  88 {
  89         struct obd_device *obd = data;
  90         struct filter_obd *filter = &obd->u.filter;
  91         int retval = 0, rc, i;
  92
  93         if (obd == NULL)
  94                 return 0;
  95
  96         for (i = FILTER_GROUP_MDS0; i < filter->fo_group_count; i++) {
  97                 rc = snprintf(page, count, LPU64"\n",filter_last_id(filter, i));
  98                 if (rc < 0) {
  99                         retval = rc;
 100                         break;
 101                 }
 102                 page += rc;
 103                 count -= rc;
 104                 retval += rc;
 105         }
 106         return retval;
 107 }
 108
 109 int lprocfs_filter_rd_readcache(char *page, char **start, off_t off, int count,
 110                                 int *eof, void *data)
 111 {
 112         struct obd_device *obd = data;
 113         int rc;
 114
 115         rc = snprintf(page, count, LPU64"\n",
 116                       obd->u.filter.fo_readcache_max_filesize);
 117         return rc;
 118 }
 119
 120 int lprocfs_filter_wr_readcache(struct file *file, const char *buffer,
 121                                 unsigned long count, void *data)
 122 {
 123         struct obd_device *obd = data;
 124         __u64 val;
 125         int rc;
 126
 127         rc = lprocfs_write_u64_helper(buffer, count, &val);
 128         if (rc)
 129                 return rc;
 130
 131         obd->u.filter.fo_readcache_max_filesize = val;
 132         return count;
 133 }
 134
 135
 136 int lprocfs_filter_rd_fmd_max_num(char *page, char **start, off_t off,
 137                                   int count, int *eof, void *data)
 138 {
 139         struct obd_device *obd = data;
 140         int rc;
 141
 142         rc = snprintf(page, count, "%u\n", obd->u.filter.fo_fmd_max_num);
 143         return rc;
 144 }
 145
 146 int lprocfs_filter_wr_fmd_max_num(struct file *file, const char *buffer,
 147                                   unsigned long count, void *data)
 148 {
 149         struct obd_device *obd = data;
 150         int val;
 151         int rc;
 152
 153         rc = lprocfs_write_helper(buffer, count, &val);
 154         if (rc)
 155                 return rc;
 156
 157         if (val > 65536 || val < 1)
 158                 return -EINVAL;
 159
 160         obd->u.filter.fo_fmd_max_num = val;
 161         return count;
 162 }
 163
 164 int lprocfs_filter_rd_fmd_max_age(char *page, char **start, off_t off,
 165                                   int count, int *eof, void *data)
 166 {
 167         struct obd_device *obd = data;
 168         int rc;
 169
 170         rc = snprintf(page, count, "%u\n", obd->u.filter.fo_fmd_max_age / HZ);
 171         return rc;
 172 }
 173
 174 int lprocfs_filter_wr_fmd_max_age(struct file *file, const char *buffer,
 175                                   unsigned long count, void *data)
 176 {
 177         struct obd_device *obd = data;
 178         int val;
 179         int rc;
 180
 181         rc = lprocfs_write_helper(buffer, count, &val);
 182         if (rc)
 183                 return rc;
 184
 185         if (val > 65536 || val < 1)
 186                 return -EINVAL;
 187
 188         obd->u.filter.fo_fmd_max_age = val * HZ;
 189         return count;
 190 }
 191
 192 static int lprocfs_filter_rd_capa(char *page, char **start, off_t off,
 193                                   int count, int *eof, void *data)
 194 {
 195         struct obd_device *obd = data;
 196         int rc;
 197
 198         rc = snprintf(page, count, "capability on: %s\n",
 199                       obd->u.filter.fo_fl_oss_capa ? "oss" : "");
 200         return rc;
 201 }
 202
 203 static int lprocfs_filter_wr_capa(struct file *file, const char *buffer,
 204                                   unsigned long count, void *data)
 205 {
 206         struct obd_device *obd = data;
 207         int val, rc;
 208
 209         rc = lprocfs_write_helper(buffer, count, &val);
 210         if (rc)
 211                 return rc;
 212
 213         if (val & ~0x1) {
 214                 CERROR("invalid capability mode, only 0/1 are accepted.\n"
 215                        " 1: enable oss fid capability\n"
 216                        " 0: disable oss fid capability\n");
 217                 return -EINVAL;
 218         }
 219
 220         obd->u.filter.fo_fl_oss_capa = val;
 221         LCONSOLE_INFO("OSS %s %s fid capability.\n", obd->obd_name,
 222                       val ? "enabled" : "disabled");
 223         return count;
 224 }
 225
 226 static int lprocfs_filter_rd_capa_count(char *page, char **start, off_t off,
 227                                         int count, int *eof, void *data)
 228 {
 229         return snprintf(page, count, "%d %d\n",
 230                         capa_count[CAPA_SITE_CLIENT],
 231                         capa_count[CAPA_SITE_SERVER]);
 232 }
 233
 234 static struct lprocfs_vars lprocfs_filter_obd_vars[] = {
 235         { "uuid",         lprocfs_rd_uuid,          0, 0 },
 236         { "blocksize",    lprocfs_rd_blksize,       0, 0 },
 237         { "kbytestotal",  lprocfs_rd_kbytestotal,   0, 0 },
 238         { "kbytesfree",   lprocfs_rd_kbytesfree,    0, 0 },
 239         { "kbytesavail",  lprocfs_rd_kbytesavail,   0, 0 },
 240         { "filestotal",   lprocfs_rd_filestotal,    0, 0 },
 241         { "filesfree",    lprocfs_rd_filesfree,     0, 0 },
 242         { "filegroups",   lprocfs_filter_rd_groups, 0, 0 },
 243         { "fstype",       lprocfs_rd_fstype,        0, 0 },
 244         { "mntdev",       lprocfs_filter_rd_mntdev, 0, 0 },
 245         { "last_id",      lprocfs_filter_rd_last_id,0, 0 },
 246         { "tot_dirty",    lprocfs_filter_rd_tot_dirty,   0, 0 },
 247         { "tot_pending",  lprocfs_filter_rd_tot_pending, 0, 0 },
 248         { "tot_granted",  lprocfs_filter_rd_tot_granted, 0, 0 },
 249         { "recovery_status", lprocfs_obd_rd_recovery_status, 0, 0 },
 250         { "recovery_maxtime", lprocfs_obd_rd_recovery_maxtime,
 251                               lprocfs_obd_wr_recovery_maxtime, 0},
 252         { "evict_client", 0, lprocfs_wr_evict_client, 0,
 253                                 &lprocfs_evict_client_fops},
 254         { "num_exports",  lprocfs_rd_num_exports,   0, 0 },
 255         { "readcache_max_filesize",
 256                           lprocfs_filter_rd_readcache,
 257                           lprocfs_filter_wr_readcache, 0 },
 258 #ifdef HAVE_QUOTA_SUPPORT
 259         { "quota_bunit_sz", lprocfs_rd_bunit, lprocfs_wr_bunit, 0},
 260         { "quota_btune_sz", lprocfs_rd_btune, lprocfs_wr_btune, 0},
 261         { "quota_iunit_sz", lprocfs_rd_iunit, lprocfs_wr_iunit, 0},
 262         { "quota_itune_sz", lprocfs_rd_itune, lprocfs_wr_itune, 0},
 263         { "quota_type",     lprocfs_rd_type, lprocfs_wr_type, 0},
 264 #endif
 265         { "client_cache_count", lprocfs_filter_rd_fmd_max_num,
 266                           lprocfs_filter_wr_fmd_max_num, 0 },
 267         { "client_cache_seconds", lprocfs_filter_rd_fmd_max_age,
 268                           lprocfs_filter_wr_fmd_max_age, 0 },
 269         { "capa",         lprocfs_filter_rd_capa,
 270                           lprocfs_filter_wr_capa, 0 },
 271         { "capa_count",   lprocfs_filter_rd_capa_count, 0, 0 },
 272         { 0 }
 273 };
 274
 275 static struct lprocfs_vars lprocfs_filter_module_vars[] = {
 276         { "num_refs",     lprocfs_rd_numrefs,       0, 0 },
 277         { 0 }
 278 };
 279
 280 void filter_tally(struct obd_export *exp, struct page **pages, int nr_pages,
 281                   unsigned long *blocks, int blocks_per_page, int wr)
 282 {
 283         struct filter_obd *filter = &exp->exp_obd->u.filter;
 284         struct filter_export_data *fed = &exp->exp_filter_data;
 285         struct page *last_page = NULL;
 286         unsigned long *last_block = NULL;
 287         unsigned long discont_pages = 0;
 288         unsigned long discont_blocks = 0;
 289         int i;
 290
 291         if (nr_pages == 0)
 292                 return;
 293
 294         lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_R_PAGES + wr],
 295                               nr_pages);
 296         lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_R_PAGES + wr],
 297                               nr_pages);
 298         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats)
 299                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_PAGES + wr],
 300                                       nr_pages);
 301
 302         while (nr_pages-- > 0) {
 303                 if (last_page && (*pages)->index != (last_page->index + 1))
 304                         discont_pages++;
 305                 last_page = *pages;
 306                 pages++;
 307                 for (i = 0; i < blocks_per_page; i++) {
 308                         if (last_block && *blocks != (*last_block + 1))
 309                                 discont_blocks++;
 310                         last_block = blocks++;
 311                 }
 312         }
 313
 314         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_PAGES +wr],
 315                          discont_pages);
 316         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_PAGES + wr],
 317                          discont_pages);
 318         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_BLOCKS+wr],
 319                          discont_blocks);
 320         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_BLOCKS + wr],
 321                          discont_blocks);
 322
 323         if (exp->exp_nid_stats && exp->exp_nid_stats->nid_brw_stats) {
 324                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_DISCONT_PAGES + wr],
 325                                       discont_pages);
 326                 lprocfs_oh_tally_log2(&exp->exp_nid_stats->nid_brw_stats->hist[BRW_W_DISCONT_BLOCKS + wr],
 327                                       discont_blocks);
 328         }
 329 }
 330
 331 #define pct(a,b) (b ? a * 100 / b : 0)
 332
 333 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
 334         struct obd_histogram *read, struct obd_histogram *write, int log2)
 335 {
 336         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 337         int i;
 338
 339         seq_printf(seq, "\n%26s read      |     write\n", " ");
 340         seq_printf(seq, "%-22s %-5s %% cum %% |  %-5s %% cum %%\n",
 341                    name, units, units);
 342
 343         read_tot = lprocfs_oh_sum(read);
 344         write_tot = lprocfs_oh_sum(write);
 345         for (i = 0; i < OBD_HIST_MAX; i++) {
 346                 r = read->oh_buckets[i];
 347                 w = write->oh_buckets[i];
 348                 read_cum += r;
 349                 write_cum += w;
 350                 if (read_cum == 0 && write_cum == 0)
 351                         continue;
 352
 353                 if (!log2)
 354                         seq_printf(seq, "%u", i);
 355                 else if (i < 10)
 356                         seq_printf(seq, "%u", 1<<i);
 357                 else if (i < 20)
 358                         seq_printf(seq, "%uK", 1<<(i-10));
 359                 else
 360                         seq_printf(seq, "%uM", 1<<(i-20));
 361
 362                 seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
 363                            r, pct(r, read_tot), pct(read_cum, read_tot),
 364                            w, pct(w, write_tot), pct(write_cum, write_tot));
 365
 366                 if (read_cum == read_tot && write_cum == write_tot)
 367                         break;
 368         }
 369 }
 370
 371 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
 372 {
 373         struct timeval now;
 374
 375         /* this sampling races with updates */
 376         do_gettimeofday(&now);
 377         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
 378                    now.tv_sec, now.tv_usec);
 379
 380         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
 381                           &brw_stats->hist[BRW_R_PAGES],
 382                           &brw_stats->hist[BRW_W_PAGES], 1);
 383
 384         display_brw_stats(seq, "discontiguous pages", "rpcs",
 385                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
 386                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
 387
 388         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 389                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
 390                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
 391
 392         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 393                           &brw_stats->hist[BRW_R_DIO_FRAGS],
 394                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
 395
 396         display_brw_stats(seq, "disk I/Os in flight", "ios",
 397                           &brw_stats->hist[BRW_R_RPC_HIST],
 398                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 399
 400         {
 401                 char title[24];
 402                 sprintf(title, "I/O time (1/%ds)", HZ);
 403                 display_brw_stats(seq, title, "ios",
 404                                   &brw_stats->hist[BRW_R_IO_TIME],
 405                                   &brw_stats->hist[BRW_W_IO_TIME], 1);
 406         }
 407
 408         display_brw_stats(seq, "disk I/O size", "ios",
 409                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
 410                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 411 }
 412
 413 #undef pct
 414
 415 static int filter_brw_stats_seq_show(struct seq_file *seq, void *v)
 416 {
 417         struct obd_device *dev = seq->private;
 418         struct filter_obd *filter = &dev->u.filter;
 419
 420         brw_stats_show(seq, &filter->fo_filter_stats);
 421
 422         return 0;
 423 }
 424
 425 static ssize_t filter_brw_stats_seq_write(struct file *file, const char *buf,
 426                                        size_t len, loff_t *off)
 427 {
 428         struct seq_file *seq = file->private_data;
 429         struct obd_device *dev = seq->private;
 430         struct filter_obd *filter = &dev->u.filter;
 431         int i;
 432
 433         for (i = 0; i < BRW_LAST; i++)
 434                 lprocfs_oh_clear(&filter->fo_filter_stats.hist[i]);
 435
 436         return len;
 437 }
 438
 439 LPROC_SEQ_FOPS(filter_brw_stats);
 440
 441 int lproc_filter_attach_seqstat(struct obd_device *dev)
 442 {
 443         return lprocfs_obd_seq_create(dev, "brw_stats", 0444,
 444                                       &filter_brw_stats_fops, dev);
 445 }
 446
 447 static int filter_per_export_stats_seq_show(struct seq_file *seq, void *v)
 448 {
 449         struct filter_export_data *fed = seq->private;
 450
 451         brw_stats_show(seq, &fed->fed_brw_stats);
 452
 453         return 0;
 454 }
 455
 456 static ssize_t filter_per_export_stats_seq_write(struct file *file,
 457                                        const char *buf, size_t len, loff_t *off)
 458 {
 459         struct seq_file *seq = file->private_data;
 460         struct filter_export_data *fed = seq->private;
 461         int i;
 462
 463         for (i = 0; i < BRW_LAST; i++)
 464                 lprocfs_oh_clear(&fed->fed_brw_stats.hist[i]);
 465
 466         return len;
 467 }
 468
 469 LPROC_SEQ_FOPS(filter_per_export_stats);
 470
 471 void lprocfs_filter_init_vars(struct lprocfs_static_vars *lvars)
 472 {
 473     lvars->module_vars  = lprocfs_filter_module_vars;
 474     lvars->obd_vars     = lprocfs_filter_obd_vars;
 475 }
 476
 477 static int filter_per_nid_stats_seq_show(struct seq_file *seq, void *v)
 478 {
 479         nid_stat_t *tmp = seq->private;
 480
 481         if (tmp->nid_brw_stats)
 482                 brw_stats_show(seq, tmp->nid_brw_stats);
 483
 484         return 0;
 485 }
 486
 487 static ssize_t filter_per_nid_stats_seq_write(struct file *file,
 488                                               const char *buf, size_t len,
 489                                               loff_t *off)
 490 {
 491         struct seq_file *seq = file->private_data;
 492         nid_stat_t *tmp = seq->private;
 493         int i;
 494
 495         if (tmp->nid_brw_stats)
 496                 for (i = 0; i < BRW_LAST; i++)
 497                         lprocfs_oh_clear(&tmp->nid_brw_stats->hist[i]);
 498
 499         return len;
 500 }
 501
 502 LPROC_SEQ_FOPS(filter_per_nid_stats);
 503 #endif /* LPROCFS */