lustre/llite/lproc_llite.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_LLITE
  37
  38 #include <linux/version.h>
  39 #include <lustre_lite.h>
  40 #include <lprocfs_status.h>
  41 #include <linux/seq_file.h>
  42 #include <obd_support.h>
  43
  44 #include "llite_internal.h"
  45
  46 struct proc_dir_entry *proc_lustre_fs_root;
  47
  48 #ifdef LPROCFS
  49 /* /proc/lustre/llite mount point registration */
  50 struct file_operations llite_dump_pgcache_fops;
  51 struct file_operations ll_rw_extents_stats_fops;
  52 struct file_operations ll_rw_extents_stats_pp_fops;
  53 struct file_operations ll_rw_offset_stats_fops;
  54
  55 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
  56                          int *eof, void *data)
  57 {
  58         struct super_block *sb = (struct super_block *)data;
  59         struct obd_statfs osfs;
  60         int rc;
  61
  62         LASSERT(sb != NULL);
  63         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  64                                 OBD_STATFS_NODELAY);
  65         if (!rc) {
  66               *eof = 1;
  67               rc = snprintf(page, count, "%u\n", osfs.os_bsize);
  68         }
  69
  70         return rc;
  71 }
  72
  73 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
  74                              int *eof, void *data)
  75 {
  76         struct super_block *sb = (struct super_block *)data;
  77         struct obd_statfs osfs;
  78         int rc;
  79
  80         LASSERT(sb != NULL);
  81         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  82                                 OBD_STATFS_NODELAY);
  83         if (!rc) {
  84                 __u32 blk_size = osfs.os_bsize >> 10;
  85                 __u64 result = osfs.os_blocks;
  86
  87                 while (blk_size >>= 1)
  88                         result <<= 1;
  89
  90                 *eof = 1;
  91                 rc = snprintf(page, count, LPU64"\n", result);
  92         }
  93         return rc;
  94
  95 }
  96
  97 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
  98                             int *eof, void *data)
  99 {
 100         struct super_block *sb = (struct super_block *)data;
 101         struct obd_statfs osfs;
 102         int rc;
 103
 104         LASSERT(sb != NULL);
 105         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 106                                 OBD_STATFS_NODELAY);
 107         if (!rc) {
 108                 __u32 blk_size = osfs.os_bsize >> 10;
 109                 __u64 result = osfs.os_bfree;
 110
 111                 while (blk_size >>= 1)
 112                         result <<= 1;
 113
 114                 *eof = 1;
 115                 rc = snprintf(page, count, LPU64"\n", result);
 116         }
 117         return rc;
 118 }
 119
 120 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
 121                              int *eof, void *data)
 122 {
 123         struct super_block *sb = (struct super_block *)data;
 124         struct obd_statfs osfs;
 125         int rc;
 126
 127         LASSERT(sb != NULL);
 128         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 129                                 OBD_STATFS_NODELAY);
 130         if (!rc) {
 131                 __u32 blk_size = osfs.os_bsize >> 10;
 132                 __u64 result = osfs.os_bavail;
 133
 134                 while (blk_size >>= 1)
 135                         result <<= 1;
 136
 137                 *eof = 1;
 138                 rc = snprintf(page, count, LPU64"\n", result);
 139         }
 140         return rc;
 141 }
 142
 143 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
 144                             int *eof, void *data)
 145 {
 146         struct super_block *sb = (struct super_block *)data;
 147         struct obd_statfs osfs;
 148         int rc;
 149
 150         LASSERT(sb != NULL);
 151         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 152                                 OBD_STATFS_NODELAY);
 153         if (!rc) {
 154                  *eof = 1;
 155                  rc = snprintf(page, count, LPU64"\n", osfs.os_files);
 156         }
 157         return rc;
 158 }
 159
 160 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
 161                            int *eof, void *data)
 162 {
 163         struct super_block *sb = (struct super_block *)data;
 164         struct obd_statfs osfs;
 165         int rc;
 166
 167         LASSERT(sb != NULL);
 168         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 169                                 OBD_STATFS_NODELAY);
 170         if (!rc) {
 171                  *eof = 1;
 172                  rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
 173         }
 174         return rc;
 175
 176 }
 177
 178 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
 179                         int *eof, void *data)
 180 {
 181         struct super_block *sb = (struct super_block*)data;
 182
 183         LASSERT(sb != NULL);
 184         *eof = 1;
 185         return snprintf(page, count, "%s\n", sb->s_type->name);
 186 }
 187
 188 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
 189                          int *eof, void *data)
 190 {
 191         struct super_block *sb = (struct super_block *)data;
 192
 193         LASSERT(sb != NULL);
 194         *eof = 1;
 195         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
 196 }
 197
 198 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
 199                                    int count, int *eof, void *data)
 200 {
 201         struct super_block *sb = data;
 202         struct ll_sb_info *sbi = ll_s2sbi(sb);
 203         long pages_number;
 204         int mult;
 205
 206         spin_lock(&sbi->ll_lock);
 207         pages_number = sbi->ll_ra_info.ra_max_pages;
 208         spin_unlock(&sbi->ll_lock);
 209
 210         mult = 1 << (20 - CFS_PAGE_SHIFT);
 211         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 212 }
 213
 214 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
 215                                    unsigned long count, void *data)
 216 {
 217         struct super_block *sb = data;
 218         struct ll_sb_info *sbi = ll_s2sbi(sb);
 219         int mult, rc, pages_number;
 220
 221         mult = 1 << (20 - CFS_PAGE_SHIFT);
 222         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 223         if (rc)
 224                 return rc;
 225
 226         if (pages_number < 0 || pages_number > num_physpages / 2) {
 227                 CERROR("can't set file readahead more than %lu MB\n",
 228                         num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
 229                 return -ERANGE;
 230         }
 231
 232         spin_lock(&sbi->ll_lock);
 233         sbi->ll_ra_info.ra_max_pages = pages_number;
 234         spin_unlock(&sbi->ll_lock);
 235
 236         return count;
 237 }
 238
 239 static int ll_rd_max_readahead_per_file_mb(char *page, char **start, off_t off,
 240                                           int count, int *eof, void *data)
 241 {
 242         struct super_block *sb = data;
 243         struct ll_sb_info *sbi = ll_s2sbi(sb);
 244         long pages_number;
 245         int mult;
 246
 247         spin_lock(&sbi->ll_lock);
 248         pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
 249         spin_unlock(&sbi->ll_lock);
 250
 251         mult = 1 << (20 - CFS_PAGE_SHIFT);
 252         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 253 }
 254
 255 static int ll_wr_max_readahead_per_file_mb(struct file *file, const char *buffer,
 256                                           unsigned long count, void *data)
 257 {
 258         struct super_block *sb = data;
 259         struct ll_sb_info *sbi = ll_s2sbi(sb);
 260         int mult, rc, pages_number;
 261
 262         mult = 1 << (20 - CFS_PAGE_SHIFT);
 263         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 264         if (rc)
 265                 return rc;
 266
 267         if (pages_number < 0 ||
 268                 pages_number > sbi->ll_ra_info.ra_max_pages) {
 269                 CERROR("can't set file readahead more than"
 270                        "max_read_ahead_mb %lu MB\n", sbi->ll_ra_info.ra_max_pages);
 271                 return -ERANGE;
 272         }
 273
 274         spin_lock(&sbi->ll_lock);
 275         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 276         spin_unlock(&sbi->ll_lock);
 277
 278         return count;
 279 }
 280
 281 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
 282                                        int count, int *eof, void *data)
 283 {
 284         struct super_block *sb = data;
 285         struct ll_sb_info *sbi = ll_s2sbi(sb);
 286         long pages_number;
 287         int mult;
 288
 289         spin_lock(&sbi->ll_lock);
 290         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
 291         spin_unlock(&sbi->ll_lock);
 292
 293         mult = 1 << (20 - CFS_PAGE_SHIFT);
 294         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 295 }
 296
 297 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
 298                                        unsigned long count, void *data)
 299 {
 300         struct super_block *sb = data;
 301         struct ll_sb_info *sbi = ll_s2sbi(sb);
 302         int mult, rc, pages_number;
 303
 304         mult = 1 << (20 - CFS_PAGE_SHIFT);
 305         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 306         if (rc)
 307                 return rc;
 308
 309         /* Cap this at the current max readahead window size, the readahead
 310          * algorithm does this anyway so it's pointless to set it larger. */
 311         if (pages_number < 0 ||
 312             pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 313                 CERROR("can't set max_read_ahead_whole_mb more than "
 314                        "max_read_ahead_per_file_mb: %lu\n",
 315                         sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
 316                 return -ERANGE;
 317         }
 318
 319         spin_lock(&sbi->ll_lock);
 320         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 321         spin_unlock(&sbi->ll_lock);
 322
 323         return count;
 324 }
 325
 326 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
 327                                int count, int *eof, void *data)
 328 {
 329         struct super_block *sb = data;
 330         struct ll_sb_info *sbi = ll_s2sbi(sb);
 331         long pages_number;
 332         int mult;
 333
 334         spin_lock(&sbi->ll_lock);
 335         pages_number = sbi->ll_async_page_max;
 336         spin_unlock(&sbi->ll_lock);
 337
 338         mult = 1 << (20 - CFS_PAGE_SHIFT);
 339         return lprocfs_read_frac_helper(page, count, pages_number, mult);;
 340 }
 341
 342 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
 343                                   unsigned long count, void *data)
 344 {
 345         struct super_block *sb = data;
 346         struct ll_sb_info *sbi = ll_s2sbi(sb);
 347         unsigned long budget;
 348         int mult, rc, pages_number, cpu;
 349
 350         mult = 1 << (20 - CFS_PAGE_SHIFT);
 351         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 352         if (rc)
 353                 return rc;
 354
 355         if (pages_number < 0 || pages_number > num_physpages) {
 356                 CERROR("can't set max cache more than %lu MB\n",
 357                         num_physpages >> (20 - CFS_PAGE_SHIFT));
 358                 return -ERANGE;
 359         }
 360
 361         spin_lock(&sbi->ll_lock);
 362         sbi->ll_async_page_max = pages_number ;
 363         spin_unlock(&sbi->ll_lock);
 364
 365         if (!sbi->ll_osc_exp)
 366                 /* Not set up yet, don't call llap_shrink_cache */
 367                 return count;
 368
 369         spin_lock(&sbi->ll_async_page_reblnc_lock);
 370         budget = sbi->ll_async_page_max / num_online_cpus();
 371         for_each_online_cpu(cpu)
 372                 LL_PGLIST_DATA_CPU(sbi, cpu)->llpd_budget = budget;
 373         spin_unlock(&sbi->ll_async_page_reblnc_lock);
 374
 375         if (lcounter_read(&sbi->ll_async_page_count) >= sbi->ll_async_page_max)
 376                 llap_shrink_cache(sbi, -1);
 377
 378         return count;
 379 }
 380
 381 static int ll_rd_pgcache_balance(char *page, char **start, off_t off,
 382                                  int count, int *eof, void *data)
 383 {
 384         struct super_block *sb = data;
 385         struct ll_sb_info *sbi = ll_s2sbi(sb);
 386         struct ll_pglist_data *pd;
 387         unsigned long total_budget = 0;
 388         int n = 0, cpu;
 389
 390         n += snprintf(page +n, count - n, "cpu\tpage count\tbudget"
 391                       "\t\treblnc count\tgen\thit\tmiss\tcross\n");
 392         for_each_online_cpu(cpu) {
 393                 pd = LL_PGLIST_DATA_CPU(sbi, cpu);
 394                 n += snprintf(page + n, count - n,
 395                               "%d\t%-8lu\t%-8lu\t%-8lu\t%lu\t%lu\t%lu\t%lu\n",
 396                               cpu, pd->llpd_count, pd->llpd_budget,
 397                               pd->llpd_reblnc_count, pd->llpd_gen,
 398                               pd->llpd_hit, pd->llpd_miss, pd->llpd_cross);
 399                 total_budget += pd->llpd_budget;
 400         }
 401         n += snprintf(page + n, count - n,
 402                       "Total budget: %lu, page max: %lu, rebalance cnt: %lu\n",
 403                       total_budget, sbi->ll_async_page_max,
 404                       sbi->ll_async_page_reblnc_count);
 405         *eof = 1;
 406         return n;
 407 }
 408
 409 static int ll_rd_checksum(char *page, char **start, off_t off,
 410                           int count, int *eof, void *data)
 411 {
 412         struct super_block *sb = data;
 413         struct ll_sb_info *sbi = ll_s2sbi(sb);
 414
 415         return snprintf(page, count, "%u\n",
 416                         (sbi->ll_flags & LL_SBI_LLITE_CHECKSUM) ? 1 : 0);
 417 }
 418
 419 static int ll_wr_checksum(struct file *file, const char *buffer,
 420                           unsigned long count, void *data)
 421 {
 422         struct super_block *sb = data;
 423         struct ll_sb_info *sbi = ll_s2sbi(sb);
 424         int val, rc;
 425
 426         if (!sbi->ll_osc_exp)
 427                 /* Not set up yet */
 428                 return -EAGAIN;
 429
 430         rc = lprocfs_write_helper(buffer, count, &val);
 431         if (rc)
 432                 return rc;
 433         if (val)
 434                 sbi->ll_flags |=  (LL_SBI_LLITE_CHECKSUM|LL_SBI_DATA_CHECKSUM);
 435         else
 436                 sbi->ll_flags &= ~(LL_SBI_LLITE_CHECKSUM|LL_SBI_DATA_CHECKSUM);
 437
 438         rc = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 439                                 sizeof(val), &val, NULL);
 440         if (rc)
 441                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 442
 443         return count;
 444 }
 445
 446 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
 447                           int count, int *eof, void *data)
 448 {
 449         struct super_block *sb = data;
 450
 451         return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
 452 }
 453
 454 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
 455                           unsigned long count, void *data)
 456 {
 457         struct super_block *sb = data;
 458         int rc, val;
 459
 460         rc = lprocfs_write_helper(buffer, count, &val);
 461         if (rc)
 462                 return rc;
 463         ll_s2sbi(sb)->ll_max_rw_chunk = val;
 464         return count;
 465 }
 466
 467 static int ll_rd_track_id(char *page, int count, void *data,
 468                           enum stats_track_type type)
 469 {
 470         struct super_block *sb = data;
 471
 472         if (ll_s2sbi(sb)->ll_stats_track_type == type) {
 473                 return snprintf(page, count, "%d\n",
 474                                 ll_s2sbi(sb)->ll_stats_track_id);
 475
 476         } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
 477                 return snprintf(page, count, "0 (all)\n");
 478         } else {
 479                 return snprintf(page, count, "untracked\n");
 480         }
 481 }
 482
 483 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
 484                           enum stats_track_type type)
 485 {
 486         struct super_block *sb = data;
 487         int rc, pid;
 488
 489         rc = lprocfs_write_helper(buffer, count, &pid);
 490         if (rc)
 491                 return rc;
 492         ll_s2sbi(sb)->ll_stats_track_id = pid;
 493         if (pid == 0)
 494                 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
 495         else
 496                 ll_s2sbi(sb)->ll_stats_track_type = type;
 497         lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
 498         return count;
 499 }
 500
 501 static int ll_rd_track_pid(char *page, char **start, off_t off,
 502                           int count, int *eof, void *data)
 503 {
 504         return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
 505 }
 506
 507 static int ll_wr_track_pid(struct file *file, const char *buffer,
 508                           unsigned long count, void *data)
 509 {
 510         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
 511 }
 512
 513 static int ll_rd_track_ppid(char *page, char **start, off_t off,
 514                           int count, int *eof, void *data)
 515 {
 516         return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
 517 }
 518
 519 static int ll_wr_track_ppid(struct file *file, const char *buffer,
 520                           unsigned long count, void *data)
 521 {
 522         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
 523 }
 524
 525 static int ll_rd_track_gid(char *page, char **start, off_t off,
 526                           int count, int *eof, void *data)
 527 {
 528         return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
 529 }
 530
 531 static int ll_wr_track_gid(struct file *file, const char *buffer,
 532                           unsigned long count, void *data)
 533 {
 534         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
 535 }
 536
 537 static int ll_rd_contention_time(char *page, char **start, off_t off,
 538                                  int count, int *eof, void *data)
 539 {
 540         struct super_block *sb = data;
 541
 542         *eof = 1;
 543         return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
 544
 545 }
 546
 547 static int ll_wr_contention_time(struct file *file, const char *buffer,
 548                                  unsigned long count, void *data)
 549 {
 550         struct super_block *sb = data;
 551         struct ll_sb_info *sbi = ll_s2sbi(sb);
 552
 553         return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
 554                 count;
 555 }
 556
 557 static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
 558                                    int count, int *eof, void *data)
 559 {
 560         struct super_block *sb = data;
 561
 562         *eof = 1;
 563         return snprintf(page, count, "%u\n",
 564                         ll_s2sbi(sb)->ll_lockless_truncate_enable);
 565 }
 566
 567 static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
 568                                    unsigned long count, void *data)
 569 {
 570         struct super_block *sb = data;
 571         struct ll_sb_info *sbi = ll_s2sbi(sb);
 572
 573         return lprocfs_write_helper(buffer, count,
 574                                     &sbi->ll_lockless_truncate_enable)
 575                 ?: count;
 576 }
 577
 578 static int ll_rd_statahead_max(char *page, char **start, off_t off,
 579                                int count, int *eof, void *data)
 580 {
 581         struct super_block *sb = data;
 582         struct ll_sb_info *sbi = ll_s2sbi(sb);
 583
 584         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
 585 }
 586
 587 static int ll_wr_statahead_max(struct file *file, const char *buffer,
 588                                unsigned long count, void *data)
 589 {
 590         struct super_block *sb = data;
 591         struct ll_sb_info *sbi = ll_s2sbi(sb);
 592         int val, rc;
 593
 594         rc = lprocfs_write_helper(buffer, count, &val);
 595         if (rc)
 596                 return rc;
 597
 598         if (val >= 0 && val <= LL_SA_RPC_MAX)
 599                 sbi->ll_sa_max = val;
 600         else
 601                 CERROR("Bad statahead_max value %d. Valid values are in the "
 602                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
 603
 604         return count;
 605 }
 606
 607 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
 608                                  int count, int *eof, void *data)
 609 {
 610         struct super_block *sb = data;
 611         struct ll_sb_info *sbi = ll_s2sbi(sb);
 612
 613         return snprintf(page, count,
 614                         "statahead wrong: %u\n"
 615                         "statahead total: %u\n"
 616                         "ls blocked:      %llu\n"
 617                         "ls cached:       %llu\n"
 618                         "hit count:       %llu\n"
 619                         "miss count:      %llu\n",
 620                         sbi->ll_sa_wrong,
 621                         sbi->ll_sa_total,
 622                         sbi->ll_sa_blocked,
 623                         sbi->ll_sa_cached,
 624                         sbi->ll_sa_hit,
 625                         sbi->ll_sa_miss);
 626 }
 627
 628 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 629         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 630         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 631         { "fstype",       ll_rd_fstype,           0, 0 },
 632         { "blocksize",    ll_rd_blksize,          0, 0 },
 633         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 634         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 635         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 636         { "filestotal",   ll_rd_filestotal,       0, 0 },
 637         { "filesfree",    ll_rd_filesfree,        0, 0 },
 638         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 639         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 640                                ll_wr_max_readahead_mb, 0 },
 641         { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
 642                                         ll_wr_max_readahead_per_file_mb, 0 },
 643         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 644                                      ll_wr_max_read_ahead_whole_mb, 0 },
 645         { "max_cached_mb",  ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 646         { "pgcache_balance",ll_rd_pgcache_balance, 0, 0 },
 647         { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
 648         { "max_rw_chunk",   ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 649         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 650         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 651         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 652         { "contention_seconds", ll_rd_contention_time,
 653                                 ll_wr_contention_time, 0},
 654         { "lockless_truncate", ll_rd_lockless_truncate,
 655                                ll_wr_lockless_truncate, 0},
 656         { "statahead_max",      ll_rd_statahead_max, ll_wr_statahead_max, 0 },
 657         { "statahead_stats",    ll_rd_statahead_stats, 0, 0 },
 658         { 0 }
 659 };
 660
 661 #define MAX_STRING_SIZE 128
 662
 663 struct llite_file_opcode {
 664         __u32       opcode;
 665         __u32       type;
 666         const char *opname;
 667 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 668         /* file operation */
 669         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 670         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 671         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 672                                    "writeback_from_writepage" },
 673         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 674                                    "writeback_from_pressure" },
 675         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 676                                    "writeback_ok_pages" },
 677         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 678                                    "writeback_failed_pages" },
 679         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 680                                    "read_bytes" },
 681         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 682                                    "write_bytes" },
 683         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 684                                    "brw_read" },
 685         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 686                                    "brw_write" },
 687
 688         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 689         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 690         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 691         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 692         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 693         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 694         /* inode operation */
 695         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 696         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 697         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate" },
 698         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 699         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 700         /* special inode operation */
 701         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 702         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 703         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 704         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 705         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 706         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 707         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 708         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 709                                    "direct_read" },
 710         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 711                                    "direct_write" },
 712         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 713                                    "lockless_read_bytes" },
 714         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 715                                    "lockless_write_bytes" },
 716
 717 };
 718
 719 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 720 {
 721         if (!sbi->ll_stats)
 722                 return;
 723         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 724                 lprocfs_counter_add(sbi->ll_stats, op, count);
 725         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 726                  sbi->ll_stats_track_id == current->pid)
 727                 lprocfs_counter_add(sbi->ll_stats, op, count);
 728         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 729                  sbi->ll_stats_track_id == current->p_pptr->pid)
 730                 lprocfs_counter_add(sbi->ll_stats, op, count);
 731         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 732                  sbi->ll_stats_track_id == current->gid)
 733                 lprocfs_counter_add(sbi->ll_stats, op, count);
 734 }
 735 EXPORT_SYMBOL(ll_stats_ops_tally);
 736
 737 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 738                                 struct super_block *sb, char *osc, char *mdc)
 739 {
 740         struct lprocfs_vars lvars[2];
 741         struct lustre_sb_info *lsi = s2lsi(sb);
 742         struct ll_sb_info *sbi = ll_s2sbi(sb);
 743         struct obd_device *obd;
 744         char name[MAX_STRING_SIZE + 1], *ptr;
 745         int err, id, len;
 746         struct proc_dir_entry *entry;
 747         static const char *ra_stats_string[] = LL_RA_STAT_STRINGS;
 748         ENTRY;
 749
 750         memset(lvars, 0, sizeof(lvars));
 751
 752         name[MAX_STRING_SIZE] = '\0';
 753         lvars[0].name = name;
 754
 755         LASSERT(sbi != NULL);
 756         LASSERT(mdc != NULL);
 757         LASSERT(osc != NULL);
 758
 759         /* Get fsname */
 760         len = strlen(lsi->lsi_lmd->lmd_profile);
 761         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 762         if (ptr && (strcmp(ptr, "-client") == 0))
 763                 len -= 7;
 764
 765         /* Mount info */
 766         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 767                  lsi->lsi_lmd->lmd_profile, sb);
 768
 769         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 770         if (IS_ERR(sbi->ll_proc_root)) {
 771                 err = PTR_ERR(sbi->ll_proc_root);
 772                 sbi->ll_proc_root = NULL;
 773                 RETURN(err);
 774         }
 775
 776         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 777         if (entry == NULL)
 778                 GOTO(out, err = -ENOMEM);
 779         entry->proc_fops = &llite_dump_pgcache_fops;
 780         entry->data = sbi;
 781
 782         sbi->ll_ra_stats = lprocfs_alloc_stats(LL_RA_STAT,
 783                                                LPROCFS_STATS_FLAG_NONE);
 784         for (id = 0; id < LL_RA_STAT; id++)
 785                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
 786                         ra_stats_string[id], "pages");
 787         lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
 788                 sbi->ll_ra_stats);
 789
 790         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 791         if (entry == NULL)
 792                  GOTO(out, err = -ENOMEM);
 793         entry->proc_fops = &ll_rw_extents_stats_fops;
 794         entry->data = sbi;
 795
 796         entry = create_proc_entry("extents_stats_per_process", 0644,
 797                                   sbi->ll_proc_root);
 798         if (entry == NULL)
 799                  GOTO(out, err = -ENOMEM);
 800         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 801         entry->data = sbi;
 802
 803         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 804         if (entry == NULL)
 805                 GOTO(out, err = -ENOMEM);
 806         entry->proc_fops = &ll_rw_offset_stats_fops;
 807         entry->data = sbi;
 808
 809         /* File operations stats */
 810         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 811                                             LPROCFS_STATS_FLAG_NONE);
 812         if (sbi->ll_stats == NULL)
 813                 GOTO(out, err = -ENOMEM);
 814         /* do counter init */
 815         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 816                 __u32 type = llite_opcode_table[id].type;
 817                 void *ptr = NULL;
 818                 if (type & LPROCFS_TYPE_REGS)
 819                         ptr = "regs";
 820                 else if (type & LPROCFS_TYPE_BYTES)
 821                         ptr = "bytes";
 822                 else if (type & LPROCFS_TYPE_PAGES)
 823                         ptr = "pages";
 824                 lprocfs_counter_init(sbi->ll_stats,
 825                                      llite_opcode_table[id].opcode,
 826                                      (type & LPROCFS_CNTR_AVGMINMAX),
 827                                      llite_opcode_table[id].opname, ptr);
 828         }
 829         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 830         if (err)
 831                 GOTO(out, err);
 832
 833         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 834         if (err)
 835                 GOTO(out, err);
 836
 837         /* MDC info */
 838         obd = class_name2obd(mdc);
 839
 840         LASSERT(obd != NULL);
 841         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 842         LASSERT(obd->obd_type->typ_name != NULL);
 843
 844         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 845                  obd->obd_type->typ_name);
 846         lvars[0].read_fptr = lprocfs_rd_name;
 847         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 848         if (err)
 849                 GOTO(out, err);
 850
 851         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 852         lvars[0].read_fptr = lprocfs_rd_uuid;
 853         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 854         if (err)
 855                 GOTO(out, err);
 856
 857         /* OSC */
 858         obd = class_name2obd(osc);
 859
 860         LASSERT(obd != NULL);
 861         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 862         LASSERT(obd->obd_type->typ_name != NULL);
 863
 864         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 865                  obd->obd_type->typ_name);
 866         lvars[0].read_fptr = lprocfs_rd_name;
 867         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 868         if (err)
 869                 GOTO(out, err);
 870
 871         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 872         lvars[0].read_fptr = lprocfs_rd_uuid;
 873         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 874 out:
 875         if (err) {
 876                 lprocfs_remove(&sbi->ll_proc_root);
 877                 lprocfs_free_stats(&sbi->ll_stats);
 878         }
 879         RETURN(err);
 880 }
 881
 882 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 883 {
 884         if (sbi->ll_proc_root) {
 885                 lprocfs_remove(&sbi->ll_proc_root);
 886                 lprocfs_free_stats(&sbi->ll_ra_stats);
 887                 lprocfs_free_stats(&sbi->ll_stats);
 888         }
 889 }
 890 #undef MAX_STRING_SIZE
 891
 892 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 893                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 894                         if (!has_flags)                                 \
 895                                 has_flags = 1;                          \
 896                         else                                            \
 897                                 seq_putc(seq, '|');                     \
 898                         seq_puts(seq, #flag);                           \
 899                 }                                                       \
 900         } while(0);
 901
 902 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 903 {
 904         struct ll_async_page *dummy_llap = seq->private;
 905
 906         if (dummy_llap->llap_magic == 2)
 907                 return NULL;
 908
 909         return (void *)1;
 910 }
 911
 912 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 913 {
 914         struct ll_async_page *llap, *dummy_llap = seq->private;
 915         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 916         struct ll_pglist_data *pd;
 917         int cpu = dummy_llap->llap_pglist_cpu;
 918
 919         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 920          * it in our own state */
 921         if (dummy_llap->llap_magic == 0) {
 922                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
 923                                 "inode index count [ page flags ]\n");
 924                 return 0;
 925         }
 926
 927         pd = ll_pglist_cpu_lock(sbi, cpu);
 928         llap = llite_pglist_next_llap(&pd->llpd_list,
 929                                       &dummy_llap->llap_pglist_item);
 930         if (llap != NULL)  {
 931                 int has_flags = 0, i;
 932                 struct page *page = llap->llap_page;
 933                 unsigned long gen = 0UL;
 934
 935                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 936                          llap->llap_origin);
 937
 938                 for_each_online_cpu(i)
 939                          gen += LL_PGLIST_DATA_CPU(sbi, i)->llpd_gen;
 940
 941                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
 942                            "%lu %u [",
 943                            gen,
 944                            llap, llap->llap_cookie,
 945                            llap_origins[llap->llap_origin],
 946                            llap->llap_write_queued ? "wq" : "- ",
 947                            llap->llap_defer_uptodate ? "du" : "- ",
 948                            PageWriteback(page) ? "wb" : "-",
 949                            page, page->mapping->host->i_ino,
 950                            page->mapping->host->i_generation,
 951                            page->mapping->host, page->index,
 952                            page_count(page));
 953                 seq_page_flag(seq, page, locked, has_flags);
 954                 seq_page_flag(seq, page, error, has_flags);
 955                 seq_page_flag(seq, page, referenced, has_flags);
 956                 seq_page_flag(seq, page, uptodate, has_flags);
 957                 seq_page_flag(seq, page, dirty, has_flags);
 958 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 959                 seq_page_flag(seq, page, highmem, has_flags);
 960 #endif
 961                 seq_page_flag(seq, page, writeback, has_flags);
 962                 if (!has_flags)
 963                         seq_puts(seq, "-]\n");
 964                 else
 965                         seq_puts(seq, "]\n");
 966         }
 967         ll_pglist_cpu_unlock(sbi, cpu);
 968
 969         return 0;
 970 }
 971
 972 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
 973                                          loff_t *pos)
 974 {
 975         struct ll_async_page *llap, *dummy_llap = seq->private;
 976         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 977         struct ll_pglist_data *pd, *next;
 978         int cpu = dummy_llap->llap_pglist_cpu;
 979
 980         /* bail if we just displayed the banner */
 981         if (dummy_llap->llap_magic == 0) {
 982                 dummy_llap->llap_magic = 1;
 983                 return dummy_llap;
 984         }
 985
 986         /* we've just displayed the llap that is after us in the list.
 987          * we advance to a position beyond it, returning null if there
 988          * isn't another llap in the list beyond that new position. */
 989         pd = ll_pglist_cpu_lock(sbi, cpu);
 990         llap = llite_pglist_next_llap(&pd->llpd_list,
 991                         &dummy_llap->llap_pglist_item);
 992         list_del_init(&dummy_llap->llap_pglist_item);
 993         if (llap) {
 994                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
 995                 llap = llite_pglist_next_llap(&pd->llpd_list,
 996                                 &dummy_llap->llap_pglist_item);
 997         }
 998         if (llap == NULL) {
 999                 int i = cpu + 1;
1000                 for (next = NULL; i < num_possible_cpus(); i++, next = NULL) {
1001                         next = ll_pglist_cpu_lock(sbi, i);
1002                         if (!list_empty(&next->llpd_list))
1003                                 break;
1004                         ll_pglist_cpu_unlock(sbi, i);
1005                 }
1006                 if (next != NULL) {
1007                         list_move(&dummy_llap->llap_pglist_item,
1008                                   &next->llpd_list);
1009                         dummy_llap->llap_pglist_cpu = i;
1010                         ll_pglist_cpu_unlock(sbi, cpu);
1011                         llap = llite_pglist_next_llap(&next->llpd_list,
1012                                         &dummy_llap->llap_pglist_item);
1013                         LASSERT(llap);
1014                         cpu = i;
1015                 }
1016         }
1017         ll_pglist_cpu_unlock(sbi, cpu);
1018
1019         ++*pos;
1020         if (llap == NULL) {
1021                 dummy_llap->llap_magic = 2;
1022                 return NULL;
1023         }
1024         return dummy_llap;
1025 }
1026
1027 static void null_stop(struct seq_file *seq, void *v)
1028 {
1029 }
1030
1031 struct seq_operations llite_dump_pgcache_seq_sops = {
1032         .start = llite_dump_pgcache_seq_start,
1033         .stop = null_stop,
1034         .next = llite_dump_pgcache_seq_next,
1035         .show = llite_dump_pgcache_seq_show,
1036 };
1037
1038 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
1039  * while we walk the entire list, and we don't want to have to seek into
1040  * the right position in the list as an app advances with many syscalls.  we
1041  * allocate a dummy llap and hang it off file->private.  its position in
1042  * the list records where the app is currently displaying.  this way our
1043  * seq .start and .stop don't actually do anything.  .next returns null
1044  * when the dummy hits the end of the list which eventually leads to .release
1045  * where we tear down.  this kind of displaying is super-racey, so we put
1046  * a generation counter on the list so the output shows when the list
1047  * changes between reads.
1048  */
1049 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
1050 {
1051         struct proc_dir_entry *dp = PDE(inode);
1052         struct ll_async_page *dummy_llap;
1053         struct seq_file *seq;
1054         struct ll_sb_info *sbi = dp->data;
1055         struct ll_pglist_data *pd;
1056         int rc = -ENOMEM;
1057
1058         LPROCFS_ENTRY_AND_CHECK(dp);
1059
1060         OBD_ALLOC_PTR_WAIT(dummy_llap);
1061         if (dummy_llap == NULL)
1062                 GOTO(out, rc);
1063
1064         dummy_llap->llap_page = NULL;
1065         dummy_llap->llap_cookie = sbi;
1066         dummy_llap->llap_magic = 0;
1067         dummy_llap->llap_pglist_cpu = 0;
1068
1069         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
1070         if (rc) {
1071                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1072                 GOTO(out, rc);
1073         }
1074         seq = file->private_data;
1075         seq->private = dummy_llap;
1076
1077         pd = ll_pglist_cpu_lock(sbi, 0);
1078         list_add(&dummy_llap->llap_pglist_item, &pd->llpd_list);
1079         ll_pglist_cpu_unlock(sbi, 0);
1080
1081 out:
1082         if (rc)
1083                 LPROCFS_EXIT();
1084         return rc;
1085 }
1086
1087 static int llite_dump_pgcache_seq_release(struct inode *inode,
1088                                           struct file *file)
1089 {
1090         struct seq_file *seq = file->private_data;
1091         struct ll_async_page *dummy_llap = seq->private;
1092         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1093         int cpu = dummy_llap->llap_pglist_cpu;
1094
1095         ll_pglist_cpu_lock(sbi, cpu);
1096         if (!list_empty(&dummy_llap->llap_pglist_item))
1097                 list_del_init(&dummy_llap->llap_pglist_item);
1098         ll_pglist_cpu_unlock(sbi, cpu);
1099         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1100
1101         return lprocfs_seq_release(inode, file);
1102 }
1103
1104 struct file_operations llite_dump_pgcache_fops = {
1105         .owner   = THIS_MODULE,
1106         .open    = llite_dump_pgcache_seq_open,
1107         .read    = seq_read,
1108         .release = llite_dump_pgcache_seq_release,
1109 };
1110
1111 #define pct(a,b) (b ? a * 100 / b : 0)
1112
1113 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1114                                    struct seq_file *seq, int which)
1115 {
1116         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1117         unsigned long start, end, r, w;
1118         char *unitp = "KMGTPEZY";
1119         int i, units = 10;
1120         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1121
1122         read_cum = 0;
1123         write_cum = 0;
1124         start = 0;
1125
1126         for(i = 0; i < LL_HIST_MAX; i++) {
1127                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1128                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1129         }
1130
1131         for(i = 0; i < LL_HIST_MAX; i++) {
1132                 r = pp_info->pp_r_hist.oh_buckets[i];
1133                 w = pp_info->pp_w_hist.oh_buckets[i];
1134                 read_cum += r;
1135                 write_cum += w;
1136                 end = 1 << (i + LL_HIST_START - units);
1137                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1138                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1139                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1140                            r, pct(r, read_tot), pct(read_cum, read_tot),
1141                            w, pct(w, write_tot), pct(write_cum, write_tot));
1142                 start = end;
1143                 if (start == 1<<10) {
1144                         start = 1;
1145                         units += 10;
1146                         unitp++;
1147                 }
1148                 if (read_cum == read_tot && write_cum == write_tot)
1149                         break;
1150         }
1151 }
1152
1153 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1154 {
1155         struct timeval now;
1156         struct ll_sb_info *sbi = seq->private;
1157         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1158         int k;
1159
1160         do_gettimeofday(&now);
1161
1162         if (!sbi->ll_rw_stats_on) {
1163                 seq_printf(seq, "disabled\n"
1164                                 "write anything in this file to activate, "
1165                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1166                 return 0;
1167         }
1168         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1169                    now.tv_sec, now.tv_usec);
1170         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1171         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1172                    "extents", "calls", "%", "cum%",
1173                    "calls", "%", "cum%");
1174         spin_lock(&sbi->ll_pp_extent_lock);
1175         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1176                 if(io_extents->pp_extents[k].pid != 0) {
1177                         seq_printf(seq, "\nPID: %d\n",
1178                                    io_extents->pp_extents[k].pid);
1179                         ll_display_extents_info(io_extents, seq, k);
1180                 }
1181         }
1182         spin_unlock(&sbi->ll_pp_extent_lock);
1183         return 0;
1184 }
1185
1186 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1187                                                 const char *buf, size_t len,
1188                                                 loff_t *off)
1189 {
1190         struct seq_file *seq = file->private_data;
1191         struct ll_sb_info *sbi = seq->private;
1192         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1193         int i;
1194         int value = 1, rc = 0;
1195
1196         rc = lprocfs_write_helper(buf, len, &value);
1197         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1198                        strcmp(buf, "Disabled") == 0))
1199                 value = 0;
1200
1201         if (value == 0)
1202                 sbi->ll_rw_stats_on = 0;
1203         else
1204                 sbi->ll_rw_stats_on = 1;
1205
1206         spin_lock(&sbi->ll_pp_extent_lock);
1207         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1208                 io_extents->pp_extents[i].pid = 0;
1209                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1210                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1211         }
1212         spin_unlock(&sbi->ll_pp_extent_lock);
1213         return len;
1214 }
1215
1216 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1217
1218 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1219 {
1220         struct timeval now;
1221         struct ll_sb_info *sbi = seq->private;
1222         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1223
1224         do_gettimeofday(&now);
1225
1226         if (!sbi->ll_rw_stats_on) {
1227                 seq_printf(seq, "disabled\n"
1228                                 "write anything in this file to activate, "
1229                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1230                 return 0;
1231         }
1232         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1233                    now.tv_sec, now.tv_usec);
1234
1235         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1236         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1237                    "extents", "calls", "%", "cum%",
1238                    "calls", "%", "cum%");
1239         spin_lock(&sbi->ll_lock);
1240         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1241         spin_unlock(&sbi->ll_lock);
1242
1243         return 0;
1244 }
1245
1246 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1247                                         size_t len, loff_t *off)
1248 {
1249         struct seq_file *seq = file->private_data;
1250         struct ll_sb_info *sbi = seq->private;
1251         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1252         int i;
1253         int value = 1, rc = 0;
1254
1255         rc = lprocfs_write_helper(buf, len, &value);
1256         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1257                        strcmp(buf, "Disabled") == 0))
1258                 value = 0;
1259
1260         if (value == 0)
1261                 sbi->ll_rw_stats_on = 0;
1262         else
1263                 sbi->ll_rw_stats_on = 1;
1264         spin_lock(&sbi->ll_pp_extent_lock);
1265         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1266         {
1267                 io_extents->pp_extents[i].pid = 0;
1268                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1269                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1270         }
1271         spin_unlock(&sbi->ll_pp_extent_lock);
1272
1273         return len;
1274 }
1275
1276 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1277
1278 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1279                                *file, size_t count, int rw)
1280 {
1281         int i, cur = -1;
1282         struct ll_rw_process_info *process;
1283         struct ll_rw_process_info *offset;
1284         int *off_count = &sbi->ll_rw_offset_entry_count;
1285         int *process_count = &sbi->ll_offset_process_count;
1286         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1287
1288         if(!sbi->ll_rw_stats_on)
1289                 return;
1290         process = sbi->ll_rw_process_info;
1291         offset = sbi->ll_rw_offset_info;
1292
1293         spin_lock(&sbi->ll_pp_extent_lock);
1294         /* Extent statistics */
1295         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1296                 if(io_extents->pp_extents[i].pid == pid) {
1297                         cur = i;
1298                         break;
1299                 }
1300         }
1301
1302         if (cur == -1) {
1303                 /* new process */
1304                 sbi->ll_extent_process_count =
1305                         (sbi->ll_extent_process_count+1) % LL_PROCESS_HIST_MAX;
1306                 cur = sbi->ll_extent_process_count;
1307                 io_extents->pp_extents[cur].pid = pid;
1308                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1309                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1310         }
1311
1312         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1313              (i < (LL_HIST_MAX - 1)); i++);
1314         if (rw == 0) {
1315                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1316                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1317         } else {
1318                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1319                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1320         }
1321         spin_unlock(&sbi->ll_pp_extent_lock);
1322
1323         spin_lock(&sbi->ll_process_lock);
1324         /* Offset statistics */
1325         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1326                 if (process[i].rw_pid == pid) {
1327                         if (process[i].rw_last_file != file) {
1328                                 process[i].rw_range_start = file->f_pos;
1329                                 process[i].rw_last_file_pos =
1330                                                         file->f_pos + count;
1331                                 process[i].rw_smallest_extent = count;
1332                                 process[i].rw_largest_extent = count;
1333                                 process[i].rw_offset = 0;
1334                                 process[i].rw_last_file = file;
1335                                 spin_unlock(&sbi->ll_process_lock);
1336                                 return;
1337                         }
1338                         if (process[i].rw_last_file_pos != file->f_pos) {
1339                                 *off_count =
1340                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1341                                 offset[*off_count].rw_op = process[i].rw_op;
1342                                 offset[*off_count].rw_pid = pid;
1343                                 offset[*off_count].rw_range_start =
1344                                         process[i].rw_range_start;
1345                                 offset[*off_count].rw_range_end =
1346                                         process[i].rw_last_file_pos;
1347                                 offset[*off_count].rw_smallest_extent =
1348                                         process[i].rw_smallest_extent;
1349                                 offset[*off_count].rw_largest_extent =
1350                                         process[i].rw_largest_extent;
1351                                 offset[*off_count].rw_offset =
1352                                         process[i].rw_offset;
1353                                 process[i].rw_op = rw;
1354                                 process[i].rw_range_start = file->f_pos;
1355                                 process[i].rw_smallest_extent = count;
1356                                 process[i].rw_largest_extent = count;
1357                                 process[i].rw_offset = file->f_pos -
1358                                         process[i].rw_last_file_pos;
1359                         }
1360                         if(process[i].rw_smallest_extent > count)
1361                                 process[i].rw_smallest_extent = count;
1362                         if(process[i].rw_largest_extent < count)
1363                                 process[i].rw_largest_extent = count;
1364                         process[i].rw_last_file_pos = file->f_pos + count;
1365                         spin_unlock(&sbi->ll_process_lock);
1366                         return;
1367                 }
1368         }
1369         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1370         process[*process_count].rw_pid = pid;
1371         process[*process_count].rw_op = rw;
1372         process[*process_count].rw_range_start = file->f_pos;
1373         process[*process_count].rw_last_file_pos = file->f_pos + count;
1374         process[*process_count].rw_smallest_extent = count;
1375         process[*process_count].rw_largest_extent = count;
1376         process[*process_count].rw_offset = 0;
1377         process[*process_count].rw_last_file = file;
1378         spin_unlock(&sbi->ll_process_lock);
1379 }
1380
1381 char lpszt[] = LPSZ;
1382
1383 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1384 {
1385         struct timeval now;
1386         struct ll_sb_info *sbi = seq->private;
1387         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1388         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1389         char format[50];
1390         int i;
1391
1392         do_gettimeofday(&now);
1393
1394         if (!sbi->ll_rw_stats_on) {
1395                 seq_printf(seq, "disabled\n"
1396                                 "write anything in this file to activate, "
1397                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1398                 return 0;
1399         }
1400         spin_lock(&sbi->ll_process_lock);
1401
1402         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1403                    now.tv_sec, now.tv_usec);
1404         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1405                    "R/W", "PID", "RANGE START", "RANGE END",
1406                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1407         sprintf(format, "%s%s%s%s%s\n",
1408                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1409         /* We stored the discontiguous offsets here; print them first */
1410         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1411                 if (offset[i].rw_pid != 0)
1412                         /* Is there a way to snip the '%' off of LPSZ? */
1413                         seq_printf(seq, format,
1414                                    offset[i].rw_op ? 'W' : 'R',
1415                                    offset[i].rw_pid,
1416                                    offset[i].rw_range_start,
1417                                    offset[i].rw_range_end,
1418                                    offset[i].rw_smallest_extent,
1419                                    offset[i].rw_largest_extent,
1420                                    offset[i].rw_offset);
1421         }
1422         /* Then print the current offsets for each process */
1423         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1424                 if (process[i].rw_pid != 0)
1425                         seq_printf(seq, format,
1426                                    process[i].rw_op ? 'W' : 'R',
1427                                    process[i].rw_pid,
1428                                    process[i].rw_range_start,
1429                                    process[i].rw_last_file_pos,
1430                                    process[i].rw_smallest_extent,
1431                                    process[i].rw_largest_extent,
1432                                    process[i].rw_offset);
1433         }
1434         spin_unlock(&sbi->ll_process_lock);
1435
1436         return 0;
1437 }
1438
1439 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1440                                        size_t len, loff_t *off)
1441 {
1442         struct seq_file *seq = file->private_data;
1443         struct ll_sb_info *sbi = seq->private;
1444         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1445         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1446         int value = 1, rc = 0;
1447
1448         rc = lprocfs_write_helper(buf, len, &value);
1449
1450         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1451                            strcmp(buf, "Disabled") == 0))
1452                 value = 0;
1453
1454         if (value == 0)
1455                 sbi->ll_rw_stats_on = 0;
1456         else
1457                 sbi->ll_rw_stats_on = 1;
1458
1459         spin_lock(&sbi->ll_process_lock);
1460         sbi->ll_offset_process_count = 0;
1461         sbi->ll_rw_offset_entry_count = 0;
1462         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1463                LL_PROCESS_HIST_MAX);
1464         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1465                LL_OFFSET_HIST_MAX);
1466         spin_unlock(&sbi->ll_process_lock);
1467
1468         return len;
1469 }
1470
1471 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1472
1473 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1474 {
1475     lvars->module_vars  = NULL;
1476     lvars->obd_vars     = lprocfs_llite_obd_vars;
1477 }
1478 #endif /* LPROCFS */