lustre/llite/lproc_llite.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright  2008 Sun Microsystems, Inc. All rights reserved
  30  * Use is subject to license terms.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  */
  36 #define DEBUG_SUBSYSTEM S_LLITE
  37
  38 #include <linux/version.h>
  39 #include <lustre_lite.h>
  40 #include <lprocfs_status.h>
  41 #include <linux/seq_file.h>
  42 #include <obd_support.h>
  43
  44 #include "llite_internal.h"
  45
  46 struct proc_dir_entry *proc_lustre_fs_root;
  47
  48 #ifdef LPROCFS
  49 /* /proc/lustre/llite mount point registration */
  50 struct file_operations llite_dump_pgcache_fops;
  51 struct file_operations ll_rw_extents_stats_fops;
  52 struct file_operations ll_rw_extents_stats_pp_fops;
  53 struct file_operations ll_rw_offset_stats_fops;
  54
  55 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
  56                          int *eof, void *data)
  57 {
  58         struct super_block *sb = (struct super_block *)data;
  59         struct obd_statfs osfs;
  60         int rc;
  61
  62         LASSERT(sb != NULL);
  63         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  64                                 OBD_STATFS_NODELAY);
  65         if (!rc) {
  66               *eof = 1;
  67               rc = snprintf(page, count, "%u\n", osfs.os_bsize);
  68         }
  69
  70         return rc;
  71 }
  72
  73 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
  74                              int *eof, void *data)
  75 {
  76         struct super_block *sb = (struct super_block *)data;
  77         struct obd_statfs osfs;
  78         int rc;
  79
  80         LASSERT(sb != NULL);
  81         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  82                                 OBD_STATFS_NODELAY);
  83         if (!rc) {
  84                 __u32 blk_size = osfs.os_bsize >> 10;
  85                 __u64 result = osfs.os_blocks;
  86
  87                 while (blk_size >>= 1)
  88                         result <<= 1;
  89
  90                 *eof = 1;
  91                 rc = snprintf(page, count, LPU64"\n", result);
  92         }
  93         return rc;
  94
  95 }
  96
  97 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
  98                             int *eof, void *data)
  99 {
 100         struct super_block *sb = (struct super_block *)data;
 101         struct obd_statfs osfs;
 102         int rc;
 103
 104         LASSERT(sb != NULL);
 105         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 106                                 OBD_STATFS_NODELAY);
 107         if (!rc) {
 108                 __u32 blk_size = osfs.os_bsize >> 10;
 109                 __u64 result = osfs.os_bfree;
 110
 111                 while (blk_size >>= 1)
 112                         result <<= 1;
 113
 114                 *eof = 1;
 115                 rc = snprintf(page, count, LPU64"\n", result);
 116         }
 117         return rc;
 118 }
 119
 120 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
 121                              int *eof, void *data)
 122 {
 123         struct super_block *sb = (struct super_block *)data;
 124         struct obd_statfs osfs;
 125         int rc;
 126
 127         LASSERT(sb != NULL);
 128         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 129                                 OBD_STATFS_NODELAY);
 130         if (!rc) {
 131                 __u32 blk_size = osfs.os_bsize >> 10;
 132                 __u64 result = osfs.os_bavail;
 133
 134                 while (blk_size >>= 1)
 135                         result <<= 1;
 136
 137                 *eof = 1;
 138                 rc = snprintf(page, count, LPU64"\n", result);
 139         }
 140         return rc;
 141 }
 142
 143 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
 144                             int *eof, void *data)
 145 {
 146         struct super_block *sb = (struct super_block *)data;
 147         struct obd_statfs osfs;
 148         int rc;
 149
 150         LASSERT(sb != NULL);
 151         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 152                                 OBD_STATFS_NODELAY);
 153         if (!rc) {
 154                  *eof = 1;
 155                  rc = snprintf(page, count, LPU64"\n", osfs.os_files);
 156         }
 157         return rc;
 158 }
 159
 160 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
 161                            int *eof, void *data)
 162 {
 163         struct super_block *sb = (struct super_block *)data;
 164         struct obd_statfs osfs;
 165         int rc;
 166
 167         LASSERT(sb != NULL);
 168         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 169                                 OBD_STATFS_NODELAY);
 170         if (!rc) {
 171                  *eof = 1;
 172                  rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
 173         }
 174         return rc;
 175
 176 }
 177
 178 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
 179                         int *eof, void *data)
 180 {
 181         struct super_block *sb = (struct super_block*)data;
 182
 183         LASSERT(sb != NULL);
 184         *eof = 1;
 185         return snprintf(page, count, "%s\n", sb->s_type->name);
 186 }
 187
 188 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
 189                          int *eof, void *data)
 190 {
 191         struct super_block *sb = (struct super_block *)data;
 192
 193         LASSERT(sb != NULL);
 194         *eof = 1;
 195         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
 196 }
 197
 198 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
 199                                    int count, int *eof, void *data)
 200 {
 201         struct super_block *sb = data;
 202         struct ll_sb_info *sbi = ll_s2sbi(sb);
 203         long pages_number;
 204         int mult;
 205
 206         spin_lock(&sbi->ll_lock);
 207         pages_number = sbi->ll_ra_info.ra_max_pages;
 208         spin_unlock(&sbi->ll_lock);
 209
 210         mult = 1 << (20 - CFS_PAGE_SHIFT);
 211         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 212 }
 213
 214 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
 215                                    unsigned long count, void *data)
 216 {
 217         struct super_block *sb = data;
 218         struct ll_sb_info *sbi = ll_s2sbi(sb);
 219         int mult, rc, pages_number;
 220
 221         mult = 1 << (20 - CFS_PAGE_SHIFT);
 222         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 223         if (rc)
 224                 return rc;
 225
 226         if (pages_number < 0 || pages_number > num_physpages / 2) {
 227                 CERROR("can't set file readahead more than %lu MB\n",
 228                         num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
 229                 return -ERANGE;
 230         }
 231
 232         spin_lock(&sbi->ll_lock);
 233         sbi->ll_ra_info.ra_max_pages = pages_number;
 234         spin_unlock(&sbi->ll_lock);
 235
 236         return count;
 237 }
 238
 239 static int ll_rd_max_readahead_per_file_mb(char *page, char **start, off_t off,
 240                                           int count, int *eof, void *data)
 241 {
 242         struct super_block *sb = data;
 243         struct ll_sb_info *sbi = ll_s2sbi(sb);
 244         long pages_number;
 245         int mult;
 246
 247         spin_lock(&sbi->ll_lock);
 248         pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
 249         spin_unlock(&sbi->ll_lock);
 250
 251         mult = 1 << (20 - CFS_PAGE_SHIFT);
 252         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 253 }
 254
 255 static int ll_wr_max_readahead_per_file_mb(struct file *file, const char *buffer,
 256                                           unsigned long count, void *data)
 257 {
 258         struct super_block *sb = data;
 259         struct ll_sb_info *sbi = ll_s2sbi(sb);
 260         int mult, rc, pages_number;
 261
 262         mult = 1 << (20 - CFS_PAGE_SHIFT);
 263         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 264         if (rc)
 265                 return rc;
 266
 267         if (pages_number < 0 ||
 268                 pages_number > sbi->ll_ra_info.ra_max_pages) {
 269                 CERROR("can't set file readahead more than"
 270                        "max_read_ahead_mb %lu MB\n", sbi->ll_ra_info.ra_max_pages);
 271                 return -ERANGE;
 272         }
 273
 274         spin_lock(&sbi->ll_lock);
 275         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 276         spin_unlock(&sbi->ll_lock);
 277
 278         return count;
 279 }
 280
 281 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
 282                                        int count, int *eof, void *data)
 283 {
 284         struct super_block *sb = data;
 285         struct ll_sb_info *sbi = ll_s2sbi(sb);
 286         long pages_number;
 287         int mult;
 288
 289         spin_lock(&sbi->ll_lock);
 290         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
 291         spin_unlock(&sbi->ll_lock);
 292
 293         mult = 1 << (20 - CFS_PAGE_SHIFT);
 294         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 295 }
 296
 297 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
 298                                        unsigned long count, void *data)
 299 {
 300         struct super_block *sb = data;
 301         struct ll_sb_info *sbi = ll_s2sbi(sb);
 302         int mult, rc, pages_number;
 303
 304         mult = 1 << (20 - CFS_PAGE_SHIFT);
 305         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 306         if (rc)
 307                 return rc;
 308
 309         /* Cap this at the current max readahead window size, the readahead
 310          * algorithm does this anyway so it's pointless to set it larger. */
 311         if (pages_number < 0 ||
 312             pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 313                 CERROR("can't set max_read_ahead_whole_mb more than "
 314                        "max_read_ahead_per_file_mb: %lu\n",
 315                         sbi->ll_ra_info.ra_max_pages_per_file >> (20 - CFS_PAGE_SHIFT));
 316                 return -ERANGE;
 317         }
 318
 319         spin_lock(&sbi->ll_lock);
 320         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 321         spin_unlock(&sbi->ll_lock);
 322
 323         return count;
 324 }
 325
 326 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
 327                                int count, int *eof, void *data)
 328 {
 329         struct super_block *sb = data;
 330         struct ll_sb_info *sbi = ll_s2sbi(sb);
 331         long pages_number;
 332         int mult;
 333
 334         spin_lock(&sbi->ll_lock);
 335         pages_number = sbi->ll_async_page_max;
 336         spin_unlock(&sbi->ll_lock);
 337
 338         mult = 1 << (20 - CFS_PAGE_SHIFT);
 339         return lprocfs_read_frac_helper(page, count, pages_number, mult);;
 340 }
 341
 342 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
 343                                   unsigned long count, void *data)
 344 {
 345         struct super_block *sb = data;
 346         struct ll_sb_info *sbi = ll_s2sbi(sb);
 347         unsigned long budget;
 348         int mult, rc, pages_number, cpu;
 349
 350         mult = 1 << (20 - CFS_PAGE_SHIFT);
 351         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 352         if (rc)
 353                 return rc;
 354
 355         if (pages_number < 0 || pages_number > num_physpages) {
 356                 CERROR("can't set max cache more than %lu MB\n",
 357                         num_physpages >> (20 - CFS_PAGE_SHIFT));
 358                 return -ERANGE;
 359         }
 360
 361         spin_lock(&sbi->ll_lock);
 362         sbi->ll_async_page_max = pages_number ;
 363         spin_unlock(&sbi->ll_lock);
 364
 365         if (!sbi->ll_osc_exp)
 366                 /* Not set up yet, don't call llap_shrink_cache */
 367                 return count;
 368
 369         spin_lock(&sbi->ll_async_page_reblnc_lock);
 370         budget = sbi->ll_async_page_max / num_online_cpus();
 371         for_each_online_cpu(cpu)
 372                 LL_PGLIST_DATA_CPU(sbi, cpu)->llpd_budget = budget;
 373         spin_unlock(&sbi->ll_async_page_reblnc_lock);
 374
 375         if (lcounter_read_positive(&sbi->ll_async_page_count) >=
 376             sbi->ll_async_page_max)
 377                 llap_shrink_cache(sbi, -1);
 378
 379         return count;
 380 }
 381
 382 static int ll_rd_pgcache_balance(char *page, char **start, off_t off,
 383                                  int count, int *eof, void *data)
 384 {
 385         struct super_block *sb = data;
 386         struct ll_sb_info *sbi = ll_s2sbi(sb);
 387         struct ll_pglist_data *pd;
 388         unsigned long total_budget = 0;
 389         int n = 0, cpu;
 390
 391         n += snprintf(page +n, count - n, "cpu\tpage count\tbudget"
 392                       "\t\treblnc count\tgen\thit\tmiss\tcross\n");
 393         for_each_online_cpu(cpu) {
 394                 pd = LL_PGLIST_DATA_CPU(sbi, cpu);
 395                 n += snprintf(page + n, count - n,
 396                               "%d\t%-8lu\t%-8lu\t%-8lu\t%lu\t%lu\t%lu\t%lu\n",
 397                               cpu, pd->llpd_count, pd->llpd_budget,
 398                               pd->llpd_reblnc_count, pd->llpd_gen,
 399                               pd->llpd_hit, pd->llpd_miss, pd->llpd_cross);
 400                 total_budget += pd->llpd_budget;
 401         }
 402         n += snprintf(page + n, count - n,
 403                       "Total budget: %lu, page max: %lu, rebalance cnt: %lu\n",
 404                       total_budget, sbi->ll_async_page_max,
 405                       sbi->ll_async_page_reblnc_count);
 406         *eof = 1;
 407         return n;
 408 }
 409
 410 static int ll_rd_checksum(char *page, char **start, off_t off,
 411                           int count, int *eof, void *data)
 412 {
 413         struct super_block *sb = data;
 414         struct ll_sb_info *sbi = ll_s2sbi(sb);
 415
 416         return snprintf(page, count, "%u\n",
 417                         (sbi->ll_flags & LL_SBI_LLITE_CHECKSUM) ? 1 : 0);
 418 }
 419
 420 static int ll_wr_checksum(struct file *file, const char *buffer,
 421                           unsigned long count, void *data)
 422 {
 423         struct super_block *sb = data;
 424         struct ll_sb_info *sbi = ll_s2sbi(sb);
 425         int val, rc;
 426
 427         if (!sbi->ll_osc_exp)
 428                 /* Not set up yet */
 429                 return -EAGAIN;
 430
 431         rc = lprocfs_write_helper(buffer, count, &val);
 432         if (rc)
 433                 return rc;
 434         if (val)
 435                 sbi->ll_flags |=  (LL_SBI_LLITE_CHECKSUM|LL_SBI_DATA_CHECKSUM);
 436         else
 437                 sbi->ll_flags &= ~(LL_SBI_LLITE_CHECKSUM|LL_SBI_DATA_CHECKSUM);
 438
 439         rc = obd_set_info_async(sbi->ll_osc_exp, sizeof(KEY_CHECKSUM), KEY_CHECKSUM,
 440                                 sizeof(val), &val, NULL);
 441         if (rc)
 442                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 443
 444         return count;
 445 }
 446
 447 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
 448                           int count, int *eof, void *data)
 449 {
 450         struct super_block *sb = data;
 451
 452         return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
 453 }
 454
 455 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
 456                           unsigned long count, void *data)
 457 {
 458         struct super_block *sb = data;
 459         int rc, val;
 460
 461         rc = lprocfs_write_helper(buffer, count, &val);
 462         if (rc)
 463                 return rc;
 464         ll_s2sbi(sb)->ll_max_rw_chunk = val;
 465         return count;
 466 }
 467
 468 static int ll_rd_track_id(char *page, int count, void *data,
 469                           enum stats_track_type type)
 470 {
 471         struct super_block *sb = data;
 472
 473         if (ll_s2sbi(sb)->ll_stats_track_type == type) {
 474                 return snprintf(page, count, "%d\n",
 475                                 ll_s2sbi(sb)->ll_stats_track_id);
 476
 477         } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
 478                 return snprintf(page, count, "0 (all)\n");
 479         } else {
 480                 return snprintf(page, count, "untracked\n");
 481         }
 482 }
 483
 484 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
 485                           enum stats_track_type type)
 486 {
 487         struct super_block *sb = data;
 488         int rc, pid;
 489
 490         rc = lprocfs_write_helper(buffer, count, &pid);
 491         if (rc)
 492                 return rc;
 493         ll_s2sbi(sb)->ll_stats_track_id = pid;
 494         if (pid == 0)
 495                 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
 496         else
 497                 ll_s2sbi(sb)->ll_stats_track_type = type;
 498         lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
 499         return count;
 500 }
 501
 502 static int ll_rd_track_pid(char *page, char **start, off_t off,
 503                           int count, int *eof, void *data)
 504 {
 505         return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
 506 }
 507
 508 static int ll_wr_track_pid(struct file *file, const char *buffer,
 509                           unsigned long count, void *data)
 510 {
 511         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
 512 }
 513
 514 static int ll_rd_track_ppid(char *page, char **start, off_t off,
 515                           int count, int *eof, void *data)
 516 {
 517         return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
 518 }
 519
 520 static int ll_wr_track_ppid(struct file *file, const char *buffer,
 521                           unsigned long count, void *data)
 522 {
 523         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
 524 }
 525
 526 static int ll_rd_track_gid(char *page, char **start, off_t off,
 527                           int count, int *eof, void *data)
 528 {
 529         return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
 530 }
 531
 532 static int ll_wr_track_gid(struct file *file, const char *buffer,
 533                           unsigned long count, void *data)
 534 {
 535         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
 536 }
 537
 538 static int ll_rd_contention_time(char *page, char **start, off_t off,
 539                                  int count, int *eof, void *data)
 540 {
 541         struct super_block *sb = data;
 542
 543         *eof = 1;
 544         return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
 545
 546 }
 547
 548 static int ll_wr_contention_time(struct file *file, const char *buffer,
 549                                  unsigned long count, void *data)
 550 {
 551         struct super_block *sb = data;
 552         struct ll_sb_info *sbi = ll_s2sbi(sb);
 553
 554         return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
 555                 count;
 556 }
 557
 558 static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
 559                                    int count, int *eof, void *data)
 560 {
 561         struct super_block *sb = data;
 562
 563         *eof = 1;
 564         return snprintf(page, count, "%u\n",
 565                         ll_s2sbi(sb)->ll_lockless_truncate_enable);
 566 }
 567
 568 static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
 569                                    unsigned long count, void *data)
 570 {
 571         struct super_block *sb = data;
 572         struct ll_sb_info *sbi = ll_s2sbi(sb);
 573
 574         return lprocfs_write_helper(buffer, count,
 575                                     &sbi->ll_lockless_truncate_enable)
 576                 ?: count;
 577 }
 578
 579 static int ll_rd_statahead_max(char *page, char **start, off_t off,
 580                                int count, int *eof, void *data)
 581 {
 582         struct super_block *sb = data;
 583         struct ll_sb_info *sbi = ll_s2sbi(sb);
 584
 585         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
 586 }
 587
 588 static int ll_wr_statahead_max(struct file *file, const char *buffer,
 589                                unsigned long count, void *data)
 590 {
 591         struct super_block *sb = data;
 592         struct ll_sb_info *sbi = ll_s2sbi(sb);
 593         int val, rc;
 594
 595         rc = lprocfs_write_helper(buffer, count, &val);
 596         if (rc)
 597                 return rc;
 598
 599         if (val >= 0 && val <= LL_SA_RPC_MAX)
 600                 sbi->ll_sa_max = val;
 601         else
 602                 CERROR("Bad statahead_max value %d. Valid values are in the "
 603                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
 604
 605         return count;
 606 }
 607
 608 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
 609                                  int count, int *eof, void *data)
 610 {
 611         struct super_block *sb = data;
 612         struct ll_sb_info *sbi = ll_s2sbi(sb);
 613
 614         return snprintf(page, count,
 615                         "statahead wrong: %u\n"
 616                         "statahead total: %u\n"
 617                         "ls blocked:      %llu\n"
 618                         "ls cached:       %llu\n"
 619                         "hit count:       %llu\n"
 620                         "miss count:      %llu\n",
 621                         sbi->ll_sa_wrong,
 622                         sbi->ll_sa_total,
 623                         sbi->ll_sa_blocked,
 624                         sbi->ll_sa_cached,
 625                         sbi->ll_sa_hit,
 626                         sbi->ll_sa_miss);
 627 }
 628
 629 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
 630                             int count, int *eof, void *data)
 631 {
 632         struct super_block *sb = data;
 633         struct ll_sb_info *sbi = ll_s2sbi(sb);
 634
 635         return snprintf(page, count, "%u\n",
 636                         (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 637 }
 638
 639 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
 640                             unsigned long count, void *data)
 641 {
 642         struct super_block *sb = data;
 643         struct ll_sb_info *sbi = ll_s2sbi(sb);
 644         int val, rc;
 645
 646         rc = lprocfs_write_helper(buffer, count, &val);
 647         if (rc)
 648                 return rc;
 649
 650         if (val)
 651                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 652         else
 653                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 654
 655         return count;
 656 }
 657
 658 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 659         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 660         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 661         { "fstype",       ll_rd_fstype,           0, 0 },
 662         { "blocksize",    ll_rd_blksize,          0, 0 },
 663         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 664         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 665         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 666         { "filestotal",   ll_rd_filestotal,       0, 0 },
 667         { "filesfree",    ll_rd_filesfree,        0, 0 },
 668         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 669         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 670                                ll_wr_max_readahead_mb, 0 },
 671         { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
 672                                         ll_wr_max_readahead_per_file_mb, 0 },
 673         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 674                                      ll_wr_max_read_ahead_whole_mb, 0 },
 675         { "max_cached_mb",  ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 676         { "pgcache_balance",ll_rd_pgcache_balance, 0, 0 },
 677         { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
 678         { "max_rw_chunk",   ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 679         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 680         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 681         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 682         { "contention_seconds", ll_rd_contention_time,
 683                                 ll_wr_contention_time, 0},
 684         { "lockless_truncate", ll_rd_lockless_truncate,
 685                                ll_wr_lockless_truncate, 0},
 686         { "statahead_max",      ll_rd_statahead_max, ll_wr_statahead_max, 0 },
 687         { "statahead_stats",    ll_rd_statahead_stats, 0, 0 },
 688         { "lazystatfs",         ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
 689         { 0 }
 690 };
 691
 692 #define MAX_STRING_SIZE 128
 693
 694 struct llite_file_opcode {
 695         __u32       opcode;
 696         __u32       type;
 697         const char *opname;
 698 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 699         /* file operation */
 700         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 701         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 702         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 703                                    "writeback_from_writepage" },
 704         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 705                                    "writeback_from_pressure" },
 706         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 707                                    "writeback_ok_pages" },
 708         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 709                                    "writeback_failed_pages" },
 710         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 711                                    "read_bytes" },
 712         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 713                                    "write_bytes" },
 714         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 715                                    "brw_read" },
 716         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 717                                    "brw_write" },
 718
 719         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 720         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 721         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 722         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 723         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 724         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 725         /* inode operation */
 726         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 727         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 728         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate" },
 729         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 730         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 731         /* special inode operation */
 732         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 733         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 734         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 735         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 736         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 737         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 738         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 739         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 740                                    "direct_read" },
 741         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 742                                    "direct_write" },
 743         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 744                                    "lockless_read_bytes" },
 745         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 746                                    "lockless_write_bytes" },
 747
 748 };
 749
 750 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 751 {
 752         if (!sbi->ll_stats)
 753                 return;
 754         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 755                 lprocfs_counter_add(sbi->ll_stats, op, count);
 756         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 757                  sbi->ll_stats_track_id == current->pid)
 758                 lprocfs_counter_add(sbi->ll_stats, op, count);
 759         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 760                  sbi->ll_stats_track_id == current->p_pptr->pid)
 761                 lprocfs_counter_add(sbi->ll_stats, op, count);
 762         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 763                  sbi->ll_stats_track_id == current->gid)
 764                 lprocfs_counter_add(sbi->ll_stats, op, count);
 765 }
 766 EXPORT_SYMBOL(ll_stats_ops_tally);
 767
 768 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 769                                 struct super_block *sb, char *osc, char *mdc)
 770 {
 771         struct lprocfs_vars lvars[2];
 772         struct lustre_sb_info *lsi = s2lsi(sb);
 773         struct ll_sb_info *sbi = ll_s2sbi(sb);
 774         struct obd_device *obd;
 775         char name[MAX_STRING_SIZE + 1], *ptr;
 776         int err, id, len;
 777         struct proc_dir_entry *entry;
 778         static const char *ra_stats_string[] = LL_RA_STAT_STRINGS;
 779         ENTRY;
 780
 781         memset(lvars, 0, sizeof(lvars));
 782
 783         name[MAX_STRING_SIZE] = '\0';
 784         lvars[0].name = name;
 785
 786         LASSERT(sbi != NULL);
 787         LASSERT(mdc != NULL);
 788         LASSERT(osc != NULL);
 789
 790         /* Get fsname */
 791         len = strlen(lsi->lsi_lmd->lmd_profile);
 792         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 793         if (ptr && (strcmp(ptr, "-client") == 0))
 794                 len -= 7;
 795
 796         /* Mount info */
 797         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 798                  lsi->lsi_lmd->lmd_profile, sb);
 799
 800         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 801         if (IS_ERR(sbi->ll_proc_root)) {
 802                 err = PTR_ERR(sbi->ll_proc_root);
 803                 sbi->ll_proc_root = NULL;
 804                 RETURN(err);
 805         }
 806
 807         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 808         if (entry == NULL)
 809                 GOTO(out, err = -ENOMEM);
 810         entry->proc_fops = &llite_dump_pgcache_fops;
 811         entry->data = sbi;
 812
 813         sbi->ll_ra_stats = lprocfs_alloc_stats(LL_RA_STAT,
 814                                                LPROCFS_STATS_FLAG_NONE);
 815         for (id = 0; id < LL_RA_STAT; id++)
 816                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
 817                         ra_stats_string[id], "pages");
 818         lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
 819                 sbi->ll_ra_stats);
 820
 821         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 822         if (entry == NULL)
 823                  GOTO(out, err = -ENOMEM);
 824         entry->proc_fops = &ll_rw_extents_stats_fops;
 825         entry->data = sbi;
 826
 827         entry = create_proc_entry("extents_stats_per_process", 0644,
 828                                   sbi->ll_proc_root);
 829         if (entry == NULL)
 830                  GOTO(out, err = -ENOMEM);
 831         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 832         entry->data = sbi;
 833
 834         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 835         if (entry == NULL)
 836                 GOTO(out, err = -ENOMEM);
 837         entry->proc_fops = &ll_rw_offset_stats_fops;
 838         entry->data = sbi;
 839
 840         /* File operations stats */
 841         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 842                                             LPROCFS_STATS_FLAG_NONE);
 843         if (sbi->ll_stats == NULL)
 844                 GOTO(out, err = -ENOMEM);
 845         /* do counter init */
 846         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 847                 __u32 type = llite_opcode_table[id].type;
 848                 void *ptr = NULL;
 849                 if (type & LPROCFS_TYPE_REGS)
 850                         ptr = "regs";
 851                 else if (type & LPROCFS_TYPE_BYTES)
 852                         ptr = "bytes";
 853                 else if (type & LPROCFS_TYPE_PAGES)
 854                         ptr = "pages";
 855                 lprocfs_counter_init(sbi->ll_stats,
 856                                      llite_opcode_table[id].opcode,
 857                                      (type & LPROCFS_CNTR_AVGMINMAX),
 858                                      llite_opcode_table[id].opname, ptr);
 859         }
 860         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 861         if (err)
 862                 GOTO(out, err);
 863
 864         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 865         if (err)
 866                 GOTO(out, err);
 867
 868         /* MDC info */
 869         obd = class_name2obd(mdc);
 870
 871         LASSERT(obd != NULL);
 872         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 873         LASSERT(obd->obd_type->typ_name != NULL);
 874
 875         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 876                  obd->obd_type->typ_name);
 877         lvars[0].read_fptr = lprocfs_rd_name;
 878         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 879         if (err)
 880                 GOTO(out, err);
 881
 882         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 883         lvars[0].read_fptr = lprocfs_rd_uuid;
 884         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 885         if (err)
 886                 GOTO(out, err);
 887
 888         /* OSC */
 889         obd = class_name2obd(osc);
 890
 891         LASSERT(obd != NULL);
 892         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 893         LASSERT(obd->obd_type->typ_name != NULL);
 894
 895         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 896                  obd->obd_type->typ_name);
 897         lvars[0].read_fptr = lprocfs_rd_name;
 898         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 899         if (err)
 900                 GOTO(out, err);
 901
 902         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 903         lvars[0].read_fptr = lprocfs_rd_uuid;
 904         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 905 out:
 906         if (err) {
 907                 lprocfs_remove(&sbi->ll_proc_root);
 908                 lprocfs_free_stats(&sbi->ll_stats);
 909         }
 910         RETURN(err);
 911 }
 912
 913 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 914 {
 915         if (sbi->ll_proc_root) {
 916                 lprocfs_remove(&sbi->ll_proc_root);
 917                 lprocfs_free_stats(&sbi->ll_ra_stats);
 918                 lprocfs_free_stats(&sbi->ll_stats);
 919         }
 920 }
 921 #undef MAX_STRING_SIZE
 922
 923 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 924                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 925                         if (!has_flags)                                 \
 926                                 has_flags = 1;                          \
 927                         else                                            \
 928                                 seq_putc(seq, '|');                     \
 929                         seq_puts(seq, #flag);                           \
 930                 }                                                       \
 931         } while(0);
 932
 933 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 934 {
 935         struct ll_async_page *dummy_llap = seq->private;
 936
 937         if (dummy_llap->llap_magic == 2)
 938                 return NULL;
 939
 940         return (void *)1;
 941 }
 942
 943 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 944 {
 945         struct ll_async_page *llap, *dummy_llap = seq->private;
 946         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 947         struct ll_pglist_data *pd;
 948         int cpu = dummy_llap->llap_pglist_cpu;
 949
 950         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 951          * it in our own state */
 952         if (dummy_llap->llap_magic == 0) {
 953                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
 954                                 "inode index count [ page flags ]\n");
 955                 return 0;
 956         }
 957
 958         pd = ll_pglist_cpu_lock(sbi, cpu);
 959         llap = llite_pglist_next_llap(&pd->llpd_list,
 960                                       &dummy_llap->llap_pglist_item);
 961         if (llap != NULL)  {
 962                 int has_flags = 0, i;
 963                 struct page *page = llap->llap_page;
 964                 unsigned long gen = 0UL;
 965
 966                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 967                          llap->llap_origin);
 968
 969                 for_each_online_cpu(i)
 970                          gen += LL_PGLIST_DATA_CPU(sbi, i)->llpd_gen;
 971
 972                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
 973                            "%lu %u [",
 974                            gen,
 975                            llap, llap->llap_cookie,
 976                            llap_origins[llap->llap_origin],
 977                            llap->llap_write_queued ? "wq" : "- ",
 978                            llap->llap_defer_uptodate ? "du" : "- ",
 979                            PageWriteback(page) ? "wb" : "-",
 980                            page, page->mapping->host->i_ino,
 981                            page->mapping->host->i_generation,
 982                            page->mapping->host, page->index,
 983                            page_count(page));
 984                 seq_page_flag(seq, page, locked, has_flags);
 985                 seq_page_flag(seq, page, error, has_flags);
 986                 seq_page_flag(seq, page, referenced, has_flags);
 987                 seq_page_flag(seq, page, uptodate, has_flags);
 988                 seq_page_flag(seq, page, dirty, has_flags);
 989 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 990                 seq_page_flag(seq, page, highmem, has_flags);
 991 #endif
 992                 seq_page_flag(seq, page, writeback, has_flags);
 993                 if (!has_flags)
 994                         seq_puts(seq, "-]\n");
 995                 else
 996                         seq_puts(seq, "]\n");
 997         }
 998         ll_pglist_cpu_unlock(sbi, cpu);
 999
1000         return 0;
1001 }
1002
1003 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
1004                                          loff_t *pos)
1005 {
1006         struct ll_async_page *llap, *dummy_llap = seq->private;
1007         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1008         struct ll_pglist_data *pd, *next;
1009         int cpu = dummy_llap->llap_pglist_cpu;
1010
1011         /* bail if we just displayed the banner */
1012         if (dummy_llap->llap_magic == 0) {
1013                 dummy_llap->llap_magic = 1;
1014                 return dummy_llap;
1015         }
1016
1017         /* we've just displayed the llap that is after us in the list.
1018          * we advance to a position beyond it, returning null if there
1019          * isn't another llap in the list beyond that new position. */
1020         pd = ll_pglist_cpu_lock(sbi, cpu);
1021         llap = llite_pglist_next_llap(&pd->llpd_list,
1022                         &dummy_llap->llap_pglist_item);
1023         list_del_init(&dummy_llap->llap_pglist_item);
1024         if (llap) {
1025                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
1026                 llap = llite_pglist_next_llap(&pd->llpd_list,
1027                                 &dummy_llap->llap_pglist_item);
1028         }
1029         if (llap == NULL) {
1030                 int i = cpu + 1;
1031                 for (next = NULL; i < num_possible_cpus(); i++, next = NULL) {
1032                         next = ll_pglist_cpu_lock(sbi, i);
1033                         if (!list_empty(&next->llpd_list))
1034                                 break;
1035                         ll_pglist_cpu_unlock(sbi, i);
1036                 }
1037                 if (next != NULL) {
1038                         list_move(&dummy_llap->llap_pglist_item,
1039                                   &next->llpd_list);
1040                         dummy_llap->llap_pglist_cpu = i;
1041                         ll_pglist_cpu_unlock(sbi, cpu);
1042                         llap = llite_pglist_next_llap(&next->llpd_list,
1043                                         &dummy_llap->llap_pglist_item);
1044                         LASSERT(llap);
1045                         cpu = i;
1046                 }
1047         }
1048         ll_pglist_cpu_unlock(sbi, cpu);
1049
1050         ++*pos;
1051         if (llap == NULL) {
1052                 dummy_llap->llap_magic = 2;
1053                 return NULL;
1054         }
1055         return dummy_llap;
1056 }
1057
1058 static void null_stop(struct seq_file *seq, void *v)
1059 {
1060 }
1061
1062 struct seq_operations llite_dump_pgcache_seq_sops = {
1063         .start = llite_dump_pgcache_seq_start,
1064         .stop = null_stop,
1065         .next = llite_dump_pgcache_seq_next,
1066         .show = llite_dump_pgcache_seq_show,
1067 };
1068
1069 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
1070  * while we walk the entire list, and we don't want to have to seek into
1071  * the right position in the list as an app advances with many syscalls.  we
1072  * allocate a dummy llap and hang it off file->private.  its position in
1073  * the list records where the app is currently displaying.  this way our
1074  * seq .start and .stop don't actually do anything.  .next returns null
1075  * when the dummy hits the end of the list which eventually leads to .release
1076  * where we tear down.  this kind of displaying is super-racey, so we put
1077  * a generation counter on the list so the output shows when the list
1078  * changes between reads.
1079  */
1080 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
1081 {
1082         struct proc_dir_entry *dp = PDE(inode);
1083         struct ll_async_page *dummy_llap;
1084         struct seq_file *seq;
1085         struct ll_sb_info *sbi = dp->data;
1086         struct ll_pglist_data *pd;
1087         int rc = -ENOMEM;
1088
1089         LPROCFS_ENTRY_AND_CHECK(dp);
1090
1091         OBD_ALLOC_PTR_WAIT(dummy_llap);
1092         if (dummy_llap == NULL)
1093                 GOTO(out, rc);
1094
1095         dummy_llap->llap_page = NULL;
1096         dummy_llap->llap_cookie = sbi;
1097         dummy_llap->llap_magic = 0;
1098         dummy_llap->llap_pglist_cpu = 0;
1099
1100         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
1101         if (rc) {
1102                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1103                 GOTO(out, rc);
1104         }
1105         seq = file->private_data;
1106         seq->private = dummy_llap;
1107
1108         pd = ll_pglist_cpu_lock(sbi, 0);
1109         list_add(&dummy_llap->llap_pglist_item, &pd->llpd_list);
1110         ll_pglist_cpu_unlock(sbi, 0);
1111
1112 out:
1113         if (rc)
1114                 LPROCFS_EXIT();
1115         return rc;
1116 }
1117
1118 static int llite_dump_pgcache_seq_release(struct inode *inode,
1119                                           struct file *file)
1120 {
1121         struct seq_file *seq = file->private_data;
1122         struct ll_async_page *dummy_llap = seq->private;
1123         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1124         int cpu = dummy_llap->llap_pglist_cpu;
1125
1126         ll_pglist_cpu_lock(sbi, cpu);
1127         if (!list_empty(&dummy_llap->llap_pglist_item))
1128                 list_del_init(&dummy_llap->llap_pglist_item);
1129         ll_pglist_cpu_unlock(sbi, cpu);
1130         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1131
1132         return lprocfs_seq_release(inode, file);
1133 }
1134
1135 struct file_operations llite_dump_pgcache_fops = {
1136         .owner   = THIS_MODULE,
1137         .open    = llite_dump_pgcache_seq_open,
1138         .read    = seq_read,
1139         .release = llite_dump_pgcache_seq_release,
1140 };
1141
1142 #define pct(a,b) (b ? a * 100 / b : 0)
1143
1144 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1145                                    struct seq_file *seq, int which)
1146 {
1147         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1148         unsigned long start, end, r, w;
1149         char *unitp = "KMGTPEZY";
1150         int i, units = 10;
1151         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1152
1153         read_cum = 0;
1154         write_cum = 0;
1155         start = 0;
1156
1157         for(i = 0; i < LL_HIST_MAX; i++) {
1158                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1159                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1160         }
1161
1162         for(i = 0; i < LL_HIST_MAX; i++) {
1163                 r = pp_info->pp_r_hist.oh_buckets[i];
1164                 w = pp_info->pp_w_hist.oh_buckets[i];
1165                 read_cum += r;
1166                 write_cum += w;
1167                 end = 1 << (i + LL_HIST_START - units);
1168                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1169                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1170                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1171                            r, pct(r, read_tot), pct(read_cum, read_tot),
1172                            w, pct(w, write_tot), pct(write_cum, write_tot));
1173                 start = end;
1174                 if (start == 1<<10) {
1175                         start = 1;
1176                         units += 10;
1177                         unitp++;
1178                 }
1179                 if (read_cum == read_tot && write_cum == write_tot)
1180                         break;
1181         }
1182 }
1183
1184 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1185 {
1186         struct timeval now;
1187         struct ll_sb_info *sbi = seq->private;
1188         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1189         int k;
1190
1191         do_gettimeofday(&now);
1192
1193         if (!sbi->ll_rw_stats_on) {
1194                 seq_printf(seq, "disabled\n"
1195                                 "write anything in this file to activate, "
1196                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1197                 return 0;
1198         }
1199         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1200                    now.tv_sec, now.tv_usec);
1201         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1202         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1203                    "extents", "calls", "%", "cum%",
1204                    "calls", "%", "cum%");
1205         spin_lock(&sbi->ll_pp_extent_lock);
1206         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1207                 if(io_extents->pp_extents[k].pid != 0) {
1208                         seq_printf(seq, "\nPID: %d\n",
1209                                    io_extents->pp_extents[k].pid);
1210                         ll_display_extents_info(io_extents, seq, k);
1211                 }
1212         }
1213         spin_unlock(&sbi->ll_pp_extent_lock);
1214         return 0;
1215 }
1216
1217 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1218                                                 const char *buf, size_t len,
1219                                                 loff_t *off)
1220 {
1221         struct seq_file *seq = file->private_data;
1222         struct ll_sb_info *sbi = seq->private;
1223         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1224         int i;
1225         int value = 1, rc = 0;
1226
1227         rc = lprocfs_write_helper(buf, len, &value);
1228         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1229                        strcmp(buf, "Disabled") == 0))
1230                 value = 0;
1231
1232         if (value == 0)
1233                 sbi->ll_rw_stats_on = 0;
1234         else
1235                 sbi->ll_rw_stats_on = 1;
1236
1237         spin_lock(&sbi->ll_pp_extent_lock);
1238         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1239                 io_extents->pp_extents[i].pid = 0;
1240                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1241                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1242         }
1243         spin_unlock(&sbi->ll_pp_extent_lock);
1244         return len;
1245 }
1246
1247 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1248
1249 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1250 {
1251         struct timeval now;
1252         struct ll_sb_info *sbi = seq->private;
1253         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1254
1255         do_gettimeofday(&now);
1256
1257         if (!sbi->ll_rw_stats_on) {
1258                 seq_printf(seq, "disabled\n"
1259                                 "write anything in this file to activate, "
1260                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1261                 return 0;
1262         }
1263         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1264                    now.tv_sec, now.tv_usec);
1265
1266         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1267         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1268                    "extents", "calls", "%", "cum%",
1269                    "calls", "%", "cum%");
1270         spin_lock(&sbi->ll_lock);
1271         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1272         spin_unlock(&sbi->ll_lock);
1273
1274         return 0;
1275 }
1276
1277 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1278                                         size_t len, loff_t *off)
1279 {
1280         struct seq_file *seq = file->private_data;
1281         struct ll_sb_info *sbi = seq->private;
1282         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1283         int i;
1284         int value = 1, rc = 0;
1285
1286         rc = lprocfs_write_helper(buf, len, &value);
1287         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1288                        strcmp(buf, "Disabled") == 0))
1289                 value = 0;
1290
1291         if (value == 0)
1292                 sbi->ll_rw_stats_on = 0;
1293         else
1294                 sbi->ll_rw_stats_on = 1;
1295         spin_lock(&sbi->ll_pp_extent_lock);
1296         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1297         {
1298                 io_extents->pp_extents[i].pid = 0;
1299                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1300                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1301         }
1302         spin_unlock(&sbi->ll_pp_extent_lock);
1303
1304         return len;
1305 }
1306
1307 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1308
1309 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1310                                *file, size_t count, int rw)
1311 {
1312         int i, cur = -1;
1313         struct ll_rw_process_info *process;
1314         struct ll_rw_process_info *offset;
1315         int *off_count = &sbi->ll_rw_offset_entry_count;
1316         int *process_count = &sbi->ll_offset_process_count;
1317         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1318
1319         if(!sbi->ll_rw_stats_on)
1320                 return;
1321         process = sbi->ll_rw_process_info;
1322         offset = sbi->ll_rw_offset_info;
1323
1324         spin_lock(&sbi->ll_pp_extent_lock);
1325         /* Extent statistics */
1326         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1327                 if(io_extents->pp_extents[i].pid == pid) {
1328                         cur = i;
1329                         break;
1330                 }
1331         }
1332
1333         if (cur == -1) {
1334                 /* new process */
1335                 sbi->ll_extent_process_count =
1336                         (sbi->ll_extent_process_count+1) % LL_PROCESS_HIST_MAX;
1337                 cur = sbi->ll_extent_process_count;
1338                 io_extents->pp_extents[cur].pid = pid;
1339                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1340                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1341         }
1342
1343         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1344              (i < (LL_HIST_MAX - 1)); i++);
1345         if (rw == 0) {
1346                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1347                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1348         } else {
1349                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1350                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1351         }
1352         spin_unlock(&sbi->ll_pp_extent_lock);
1353
1354         spin_lock(&sbi->ll_process_lock);
1355         /* Offset statistics */
1356         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1357                 if (process[i].rw_pid == pid) {
1358                         if (process[i].rw_last_file != file) {
1359                                 process[i].rw_range_start = file->f_pos;
1360                                 process[i].rw_last_file_pos =
1361                                                         file->f_pos + count;
1362                                 process[i].rw_smallest_extent = count;
1363                                 process[i].rw_largest_extent = count;
1364                                 process[i].rw_offset = 0;
1365                                 process[i].rw_last_file = file;
1366                                 spin_unlock(&sbi->ll_process_lock);
1367                                 return;
1368                         }
1369                         if (process[i].rw_last_file_pos != file->f_pos) {
1370                                 *off_count =
1371                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1372                                 offset[*off_count].rw_op = process[i].rw_op;
1373                                 offset[*off_count].rw_pid = pid;
1374                                 offset[*off_count].rw_range_start =
1375                                         process[i].rw_range_start;
1376                                 offset[*off_count].rw_range_end =
1377                                         process[i].rw_last_file_pos;
1378                                 offset[*off_count].rw_smallest_extent =
1379                                         process[i].rw_smallest_extent;
1380                                 offset[*off_count].rw_largest_extent =
1381                                         process[i].rw_largest_extent;
1382                                 offset[*off_count].rw_offset =
1383                                         process[i].rw_offset;
1384                                 process[i].rw_op = rw;
1385                                 process[i].rw_range_start = file->f_pos;
1386                                 process[i].rw_smallest_extent = count;
1387                                 process[i].rw_largest_extent = count;
1388                                 process[i].rw_offset = file->f_pos -
1389                                         process[i].rw_last_file_pos;
1390                         }
1391                         if(process[i].rw_smallest_extent > count)
1392                                 process[i].rw_smallest_extent = count;
1393                         if(process[i].rw_largest_extent < count)
1394                                 process[i].rw_largest_extent = count;
1395                         process[i].rw_last_file_pos = file->f_pos + count;
1396                         spin_unlock(&sbi->ll_process_lock);
1397                         return;
1398                 }
1399         }
1400         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1401         process[*process_count].rw_pid = pid;
1402         process[*process_count].rw_op = rw;
1403         process[*process_count].rw_range_start = file->f_pos;
1404         process[*process_count].rw_last_file_pos = file->f_pos + count;
1405         process[*process_count].rw_smallest_extent = count;
1406         process[*process_count].rw_largest_extent = count;
1407         process[*process_count].rw_offset = 0;
1408         process[*process_count].rw_last_file = file;
1409         spin_unlock(&sbi->ll_process_lock);
1410 }
1411
1412 char lpszt[] = LPSZ;
1413
1414 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1415 {
1416         struct timeval now;
1417         struct ll_sb_info *sbi = seq->private;
1418         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1419         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1420         char format[50];
1421         int i;
1422
1423         do_gettimeofday(&now);
1424
1425         if (!sbi->ll_rw_stats_on) {
1426                 seq_printf(seq, "disabled\n"
1427                                 "write anything in this file to activate, "
1428                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1429                 return 0;
1430         }
1431         spin_lock(&sbi->ll_process_lock);
1432
1433         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1434                    now.tv_sec, now.tv_usec);
1435         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1436                    "R/W", "PID", "RANGE START", "RANGE END",
1437                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1438         sprintf(format, "%s%s%s%s%s\n",
1439                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1440         /* We stored the discontiguous offsets here; print them first */
1441         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1442                 if (offset[i].rw_pid != 0)
1443                         /* Is there a way to snip the '%' off of LPSZ? */
1444                         seq_printf(seq, format,
1445                                    offset[i].rw_op ? 'W' : 'R',
1446                                    offset[i].rw_pid,
1447                                    offset[i].rw_range_start,
1448                                    offset[i].rw_range_end,
1449                                    offset[i].rw_smallest_extent,
1450                                    offset[i].rw_largest_extent,
1451                                    offset[i].rw_offset);
1452         }
1453         /* Then print the current offsets for each process */
1454         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1455                 if (process[i].rw_pid != 0)
1456                         seq_printf(seq, format,
1457                                    process[i].rw_op ? 'W' : 'R',
1458                                    process[i].rw_pid,
1459                                    process[i].rw_range_start,
1460                                    process[i].rw_last_file_pos,
1461                                    process[i].rw_smallest_extent,
1462                                    process[i].rw_largest_extent,
1463                                    process[i].rw_offset);
1464         }
1465         spin_unlock(&sbi->ll_process_lock);
1466
1467         return 0;
1468 }
1469
1470 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1471                                        size_t len, loff_t *off)
1472 {
1473         struct seq_file *seq = file->private_data;
1474         struct ll_sb_info *sbi = seq->private;
1475         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1476         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1477         int value = 1, rc = 0;
1478
1479         rc = lprocfs_write_helper(buf, len, &value);
1480
1481         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1482                            strcmp(buf, "Disabled") == 0))
1483                 value = 0;
1484
1485         if (value == 0)
1486                 sbi->ll_rw_stats_on = 0;
1487         else
1488                 sbi->ll_rw_stats_on = 1;
1489
1490         spin_lock(&sbi->ll_process_lock);
1491         sbi->ll_offset_process_count = 0;
1492         sbi->ll_rw_offset_entry_count = 0;
1493         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1494                LL_PROCESS_HIST_MAX);
1495         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1496                LL_OFFSET_HIST_MAX);
1497         spin_unlock(&sbi->ll_process_lock);
1498
1499         return len;
1500 }
1501
1502 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1503
1504 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1505 {
1506     lvars->module_vars  = NULL;
1507     lvars->obd_vars     = lprocfs_llite_obd_vars;
1508 }
1509 #endif /* LPROCFS */