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