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
 235 int lprocfs_filter_rd_blacklist(char *page, char **start, off_t off, int count,
 236                                 int *eof, void *data)
 237 {
 238         int rc;
 239
 240         rc = blacklist_display(page, count);
 241         *eof = 1;
 242         return rc;
 243 }
 244
 245 static
 246 int lprocfs_filter_wr_blacklist(struct file *file, const char *buffer,
 247                                 unsigned long count, void *data)
 248 {
 249         int add;
 250         uid_t uid = -1;
 251
 252         if (count < 2)
 253                 return count;
 254         if (buffer[0] == '+')
 255                 add = 1;
 256         else if (buffer[0] == '-')
 257                 add = 0;
 258         else
 259                 return count;
 260
 261         sscanf(buffer + 1, "%u", &uid);
 262         if (add)
 263                 blacklist_add(uid);
 264         else
 265                 blacklist_del(uid);
 266         return count;
 267 }
 268
 269 static struct lprocfs_vars lprocfs_obd_vars[] = {
 270         { "uuid",         lprocfs_rd_uuid,          0, 0 },
 271         { "blocksize",    lprocfs_rd_blksize,       0, 0 },
 272         { "kbytestotal",  lprocfs_rd_kbytestotal,   0, 0 },
 273         { "kbytesfree",   lprocfs_rd_kbytesfree,    0, 0 },
 274         { "kbytesavail",  lprocfs_rd_kbytesavail,   0, 0 },
 275         { "filestotal",   lprocfs_rd_filestotal,    0, 0 },
 276         { "filesfree",    lprocfs_rd_filesfree,     0, 0 },
 277         { "filegroups",   lprocfs_filter_rd_groups, 0, 0 },
 278         { "fstype",       lprocfs_rd_fstype,        0, 0 },
 279         { "mntdev",       lprocfs_filter_rd_mntdev, 0, 0 },
 280         { "last_id",      lprocfs_filter_rd_last_id,0, 0 },
 281         { "tot_dirty",    lprocfs_filter_rd_tot_dirty,   0, 0 },
 282         { "tot_pending",  lprocfs_filter_rd_tot_pending, 0, 0 },
 283         { "tot_granted",  lprocfs_filter_rd_tot_granted, 0, 0 },
 284         { "recovery_status", lprocfs_obd_rd_recovery_status, 0, 0 },
 285         { "evict_client", 0, lprocfs_wr_evict_client, 0 },
 286         { "num_exports",  lprocfs_rd_num_exports,   0, 0 },
 287         { "readcache_max_filesize",
 288                           lprocfs_filter_rd_readcache,
 289                           lprocfs_filter_wr_readcache, 0 },
 290 #ifdef HAVE_QUOTA_SUPPORT
 291         { "quota_bunit_sz", lprocfs_rd_bunit, lprocfs_wr_bunit, 0},
 292         { "quota_btune_sz", lprocfs_rd_btune, lprocfs_wr_btune, 0},
 293         { "quota_iunit_sz", lprocfs_rd_iunit, lprocfs_wr_iunit, 0},
 294         { "quota_itune_sz", lprocfs_rd_itune, lprocfs_wr_itune, 0},
 295         { "quota_type",     lprocfs_rd_type, lprocfs_wr_type, 0},
 296 #endif
 297         { "client_cache_count", lprocfs_filter_rd_fmd_max_num,
 298                           lprocfs_filter_wr_fmd_max_num, 0 },
 299         { "client_cache_seconds", lprocfs_filter_rd_fmd_max_age,
 300                           lprocfs_filter_wr_fmd_max_age, 0 },
 301         { "capa",         lprocfs_filter_rd_capa,
 302                           lprocfs_filter_wr_capa, 0 },
 303         { "capa_count",   lprocfs_filter_rd_capa_count, 0, 0 },
 304         { "blacklist",    lprocfs_filter_rd_blacklist,
 305                           lprocfs_filter_wr_blacklist, 0 },
 306         { 0 }
 307 };
 308
 309 static struct lprocfs_vars lprocfs_module_vars[] = {
 310         { "num_refs",     lprocfs_rd_numrefs,       0, 0 },
 311         { 0 }
 312 };
 313
 314 void filter_tally(struct obd_export *exp, struct page **pages, int nr_pages,
 315                   unsigned long *blocks, int blocks_per_page, int wr)
 316 {
 317         struct filter_obd *filter = &exp->exp_obd->u.filter;
 318         struct filter_export_data *fed = &exp->exp_filter_data;
 319         struct page *last_page = NULL;
 320         unsigned long *last_block = NULL;
 321         unsigned long discont_pages = 0;
 322         unsigned long discont_blocks = 0;
 323         int i;
 324
 325         if (nr_pages == 0)
 326                 return;
 327
 328         lprocfs_oh_tally_log2(&filter->fo_filter_stats.hist[BRW_R_PAGES + wr],
 329                               nr_pages);
 330         lprocfs_oh_tally_log2(&fed->fed_brw_stats.hist[BRW_R_PAGES + wr],
 331                               nr_pages);
 332
 333         while (nr_pages-- > 0) {
 334                 if (last_page && (*pages)->index != (last_page->index + 1))
 335                         discont_pages++;
 336                 last_page = *pages;
 337                 pages++;
 338                 for (i = 0; i < blocks_per_page; i++) {
 339                         if (last_block && *blocks != (*last_block + 1))
 340                                 discont_blocks++;
 341                         last_block = blocks++;
 342                 }
 343         }
 344
 345         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_PAGES +wr],
 346                          discont_pages);
 347         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_PAGES + wr],
 348                          discont_pages);
 349         lprocfs_oh_tally(&filter->fo_filter_stats.hist[BRW_R_DISCONT_BLOCKS+wr],
 350                          discont_blocks);
 351         lprocfs_oh_tally(&fed->fed_brw_stats.hist[BRW_R_DISCONT_BLOCKS + wr],
 352                          discont_blocks);
 353 }
 354
 355 #define pct(a,b) (b ? a * 100 / b : 0)
 356
 357 static void display_brw_stats(struct seq_file *seq, char *name, char *units,
 358         struct obd_histogram *read, struct obd_histogram *write, int log2)
 359 {
 360         unsigned long read_tot, write_tot, r, w, read_cum = 0, write_cum = 0;
 361         int i;
 362
 363         seq_printf(seq, "\n%26s read      |     write\n", " ");
 364         seq_printf(seq, "%-22s %-5s %% cum %% |  %-5s %% cum %%\n",
 365                    name, units, units);
 366
 367         read_tot = lprocfs_oh_sum(read);
 368         write_tot = lprocfs_oh_sum(write);
 369         for (i = 0; i < OBD_HIST_MAX; i++) {
 370                 r = read->oh_buckets[i];
 371                 w = write->oh_buckets[i];
 372                 read_cum += r;
 373                 write_cum += w;
 374                 if (read_cum == 0 && write_cum == 0)
 375                         continue;
 376
 377                 if (!log2)
 378                         seq_printf(seq, "%u", i);
 379                 else if (i < 10)
 380                         seq_printf(seq, "%u", 1<<i);
 381                 else if (i < 20)
 382                         seq_printf(seq, "%uK", 1<<(i-10));
 383                 else
 384                         seq_printf(seq, "%uM", 1<<(i-20));
 385
 386                 seq_printf(seq, ":\t\t%10lu %3lu %3lu   | %4lu %3lu %3lu\n",
 387                            r, pct(r, read_tot), pct(read_cum, read_tot),
 388                            w, pct(w, write_tot), pct(write_cum, write_tot));
 389
 390                 if (read_cum == read_tot && write_cum == write_tot)
 391                         break;
 392         }
 393 }
 394
 395 static void brw_stats_show(struct seq_file *seq, struct brw_stats *brw_stats)
 396 {
 397         struct timeval now;
 398
 399         /* this sampling races with updates */
 400         do_gettimeofday(&now);
 401         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
 402                    now.tv_sec, now.tv_usec);
 403
 404         display_brw_stats(seq, "pages per bulk r/w", "rpcs",
 405                           &brw_stats->hist[BRW_R_PAGES],
 406                           &brw_stats->hist[BRW_W_PAGES], 1);
 407
 408         display_brw_stats(seq, "discontiguous pages", "rpcs",
 409                           &brw_stats->hist[BRW_R_DISCONT_PAGES],
 410                           &brw_stats->hist[BRW_W_DISCONT_PAGES], 0);
 411
 412         display_brw_stats(seq, "discontiguous blocks", "rpcs",
 413                           &brw_stats->hist[BRW_R_DISCONT_BLOCKS],
 414                           &brw_stats->hist[BRW_W_DISCONT_BLOCKS], 0);
 415
 416         display_brw_stats(seq, "disk fragmented I/Os", "ios",
 417                           &brw_stats->hist[BRW_R_DIO_FRAGS],
 418                           &brw_stats->hist[BRW_W_DIO_FRAGS], 0);
 419
 420 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 421         display_brw_stats(seq, "disk I/Os in flight", "ios",
 422                           &brw_stats->hist[BRW_R_RPC_HIST],
 423                           &brw_stats->hist[BRW_W_RPC_HIST], 0);
 424
 425         {
 426                 char title[24];
 427                 sprintf(title, "I/O time (1/%ds)", HZ);
 428                 display_brw_stats(seq, title, "ios",
 429                                   &brw_stats->hist[BRW_R_IO_TIME],
 430                                   &brw_stats->hist[BRW_W_IO_TIME], 1);
 431         }
 432
 433         display_brw_stats(seq, "disk I/O size", "ios",
 434                           &brw_stats->hist[BRW_R_DISK_IOSIZE],
 435                           &brw_stats->hist[BRW_W_DISK_IOSIZE], 1);
 436 #endif
 437 }
 438
 439 static int filter_brw_stats_seq_show(struct seq_file *seq, void *v)
 440 {
 441         struct obd_device *dev = seq->private;
 442         struct filter_obd *filter = &dev->u.filter;
 443
 444         brw_stats_show(seq, &filter->fo_filter_stats);
 445
 446         return 0;
 447 }
 448
 449 static ssize_t filter_brw_stats_seq_write(struct file *file, const char *buf,
 450                                        size_t len, loff_t *off)
 451 {
 452         struct seq_file *seq = file->private_data;
 453         struct obd_device *dev = seq->private;
 454         struct filter_obd *filter = &dev->u.filter;
 455         int i;
 456
 457         for (i = 0; i < BRW_LAST; i++)
 458                 lprocfs_oh_clear(&filter->fo_filter_stats.hist[i]);
 459
 460         return len;
 461 }
 462
 463 LPROC_SEQ_FOPS(filter_brw_stats);
 464
 465 int lproc_filter_attach_seqstat(struct obd_device *dev)
 466 {
 467         return lprocfs_obd_seq_create(dev, "brw_stats", 0444,
 468                                       &filter_brw_stats_fops, dev);
 469 }
 470
 471 static int filter_per_export_stats_seq_show(struct seq_file *seq, void *v)
 472 {
 473         struct filter_export_data *fed = seq->private;
 474
 475         brw_stats_show(seq, &fed->fed_brw_stats);
 476
 477         return 0;
 478 }
 479
 480 static ssize_t filter_per_export_stats_seq_write(struct file *file,
 481                                        const char *buf, size_t len, loff_t *off)
 482 {
 483         struct seq_file *seq = file->private_data;
 484         struct filter_export_data *fed = seq->private;
 485         int i;
 486
 487         for (i = 0; i < BRW_LAST; i++)
 488                 lprocfs_oh_clear(&fed->fed_brw_stats.hist[i]);
 489
 490         return len;
 491 }
 492
 493 LPROC_SEQ_FOPS(filter_per_export_stats);
 494
 495 LPROCFS_INIT_VARS(filter, lprocfs_module_vars, lprocfs_obd_vars)
 496 #endif /* LPROCFS */