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