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_direct_io_default(char *page, char **start, off_t off,
 580                                      int count, int *eof, void *data)
 581 {
 582         struct super_block *sb = data;
 583
 584         *eof = 1;
 585         return snprintf(page, count, "%u\n",
 586                         ll_s2sbi(sb)->ll_direct_io_default);
 587 }
 588
 589 static int ll_wr_direct_io_default(struct file *file, const char *buffer,
 590                                      unsigned long count, void *data)
 591 {
 592         struct super_block *sb = data;
 593         struct ll_sb_info *sbi = ll_s2sbi(sb);
 594
 595         return lprocfs_write_helper(buffer, count,
 596                                     &sbi->ll_direct_io_default)
 597                 ?: count;
 598 }
 599
 600 static int ll_rd_statahead_max(char *page, char **start, off_t off,
 601                                int count, int *eof, void *data)
 602 {
 603         struct super_block *sb = data;
 604         struct ll_sb_info *sbi = ll_s2sbi(sb);
 605
 606         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
 607 }
 608
 609 static int ll_wr_statahead_max(struct file *file, const char *buffer,
 610                                unsigned long count, void *data)
 611 {
 612         struct super_block *sb = data;
 613         struct ll_sb_info *sbi = ll_s2sbi(sb);
 614         int val, rc;
 615
 616         rc = lprocfs_write_helper(buffer, count, &val);
 617         if (rc)
 618                 return rc;
 619
 620         if (val >= 0 && val <= LL_SA_RPC_MAX)
 621                 sbi->ll_sa_max = val;
 622         else
 623                 CERROR("Bad statahead_max value %d. Valid values are in the "
 624                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
 625
 626         return count;
 627 }
 628
 629 static int ll_rd_statahead_stats(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,
 636                         "statahead wrong: %u\n"
 637                         "statahead total: %u\n"
 638                         "ls blocked:      %llu\n"
 639                         "ls cached:       %llu\n"
 640                         "hit count:       %llu\n"
 641                         "miss count:      %llu\n",
 642                         sbi->ll_sa_wrong,
 643                         sbi->ll_sa_total,
 644                         sbi->ll_sa_blocked,
 645                         sbi->ll_sa_cached,
 646                         sbi->ll_sa_hit,
 647                         sbi->ll_sa_miss);
 648 }
 649
 650 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
 651                             int count, int *eof, void *data)
 652 {
 653         struct super_block *sb = data;
 654         struct ll_sb_info *sbi = ll_s2sbi(sb);
 655
 656         return snprintf(page, count, "%u\n",
 657                         (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 658 }
 659
 660 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
 661                             unsigned long count, void *data)
 662 {
 663         struct super_block *sb = data;
 664         struct ll_sb_info *sbi = ll_s2sbi(sb);
 665         int val, rc;
 666
 667         rc = lprocfs_write_helper(buffer, count, &val);
 668         if (rc)
 669                 return rc;
 670
 671         if (val)
 672                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 673         else
 674                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 675
 676         return count;
 677 }
 678
 679 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 680         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 681         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 682         { "fstype",       ll_rd_fstype,           0, 0 },
 683         { "blocksize",    ll_rd_blksize,          0, 0 },
 684         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 685         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 686         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 687         { "filestotal",   ll_rd_filestotal,       0, 0 },
 688         { "filesfree",    ll_rd_filesfree,        0, 0 },
 689         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 690         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 691                                ll_wr_max_readahead_mb, 0 },
 692         { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
 693                                         ll_wr_max_readahead_per_file_mb, 0 },
 694         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 695                                      ll_wr_max_read_ahead_whole_mb, 0 },
 696         { "max_cached_mb",  ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 697         { "pgcache_balance",ll_rd_pgcache_balance, 0, 0 },
 698         { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
 699         { "max_rw_chunk",   ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 700         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 701         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 702         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 703         { "contention_seconds", ll_rd_contention_time,
 704                                 ll_wr_contention_time, 0},
 705         { "lockless_truncate", ll_rd_lockless_truncate,
 706                                ll_wr_lockless_truncate, 0},
 707         { "direct_io_default", ll_rd_direct_io_default,
 708                                ll_wr_direct_io_default, 0},
 709         { "statahead_max",      ll_rd_statahead_max, ll_wr_statahead_max, 0 },
 710         { "statahead_stats",    ll_rd_statahead_stats, 0, 0 },
 711         { "lazystatfs",         ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
 712         { 0 }
 713 };
 714
 715 #define MAX_STRING_SIZE 128
 716
 717 struct llite_file_opcode {
 718         __u32       opcode;
 719         __u32       type;
 720         const char *opname;
 721 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 722         /* file operation */
 723         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 724         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 725         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 726                                    "writeback_from_writepage" },
 727         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 728                                    "writeback_from_pressure" },
 729         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 730                                    "writeback_ok_pages" },
 731         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 732                                    "writeback_failed_pages" },
 733         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 734                                    "read_bytes" },
 735         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 736                                    "write_bytes" },
 737         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 738                                    "brw_read" },
 739         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 740                                    "brw_write" },
 741
 742         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 743         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 744         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 745         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 746         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 747         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 748         /* inode operation */
 749         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 750         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 751         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate" },
 752         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 753         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 754         /* special inode operation */
 755         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 756         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 757         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 758         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 759         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 760         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 761         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 762         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 763                                    "direct_read" },
 764         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 765                                    "direct_write" },
 766         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 767                                    "lockless_read_bytes" },
 768         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 769                                    "lockless_write_bytes" },
 770
 771 };
 772
 773 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 774 {
 775         if (!sbi->ll_stats)
 776                 return;
 777         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 778                 lprocfs_counter_add(sbi->ll_stats, op, count);
 779         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 780                  sbi->ll_stats_track_id == current->pid)
 781                 lprocfs_counter_add(sbi->ll_stats, op, count);
 782         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 783                  sbi->ll_stats_track_id == current->parent->pid)
 784                 lprocfs_counter_add(sbi->ll_stats, op, count);
 785         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 786                  sbi->ll_stats_track_id == cfs_curproc_gid())
 787                 lprocfs_counter_add(sbi->ll_stats, op, count);
 788 }
 789 EXPORT_SYMBOL(ll_stats_ops_tally);
 790
 791 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 792                                 struct super_block *sb, char *osc, char *mdc)
 793 {
 794         struct lprocfs_vars lvars[2];
 795         struct lustre_sb_info *lsi = s2lsi(sb);
 796         struct ll_sb_info *sbi = ll_s2sbi(sb);
 797         struct obd_device *obd;
 798         char name[MAX_STRING_SIZE + 1], *ptr;
 799         int err, id, len;
 800         struct proc_dir_entry *entry;
 801         static const char *ra_stats_string[] = LL_RA_STAT_STRINGS;
 802         ENTRY;
 803
 804         memset(lvars, 0, sizeof(lvars));
 805
 806         name[MAX_STRING_SIZE] = '\0';
 807         lvars[0].name = name;
 808
 809         LASSERT(sbi != NULL);
 810         LASSERT(mdc != NULL);
 811         LASSERT(osc != NULL);
 812
 813         /* Get fsname */
 814         len = strlen(lsi->lsi_lmd->lmd_profile);
 815         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 816         if (ptr && (strcmp(ptr, "-client") == 0))
 817                 len -= 7;
 818
 819         /* Mount info */
 820         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 821                  lsi->lsi_lmd->lmd_profile, sb);
 822
 823         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 824         if (IS_ERR(sbi->ll_proc_root)) {
 825                 err = PTR_ERR(sbi->ll_proc_root);
 826                 sbi->ll_proc_root = NULL;
 827                 RETURN(err);
 828         }
 829
 830         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 831         if (entry == NULL)
 832                 GOTO(out, err = -ENOMEM);
 833         entry->proc_fops = &llite_dump_pgcache_fops;
 834         entry->data = sbi;
 835
 836         sbi->ll_ra_stats = lprocfs_alloc_stats(LL_RA_STAT,
 837                                                LPROCFS_STATS_FLAG_NONE);
 838         for (id = 0; id < LL_RA_STAT; id++)
 839                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
 840                         ra_stats_string[id], "pages");
 841         lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
 842                 sbi->ll_ra_stats);
 843
 844         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 845         if (entry == NULL)
 846                  GOTO(out, err = -ENOMEM);
 847         entry->proc_fops = &ll_rw_extents_stats_fops;
 848         entry->data = sbi;
 849
 850         entry = create_proc_entry("extents_stats_per_process", 0644,
 851                                   sbi->ll_proc_root);
 852         if (entry == NULL)
 853                  GOTO(out, err = -ENOMEM);
 854         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 855         entry->data = sbi;
 856
 857         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 858         if (entry == NULL)
 859                 GOTO(out, err = -ENOMEM);
 860         entry->proc_fops = &ll_rw_offset_stats_fops;
 861         entry->data = sbi;
 862
 863         /* File operations stats */
 864         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 865                                             LPROCFS_STATS_FLAG_NONE);
 866         if (sbi->ll_stats == NULL)
 867                 GOTO(out, err = -ENOMEM);
 868         /* do counter init */
 869         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 870                 __u32 type = llite_opcode_table[id].type;
 871                 void *ptr = NULL;
 872                 if (type & LPROCFS_TYPE_REGS)
 873                         ptr = "regs";
 874                 else if (type & LPROCFS_TYPE_BYTES)
 875                         ptr = "bytes";
 876                 else if (type & LPROCFS_TYPE_PAGES)
 877                         ptr = "pages";
 878                 lprocfs_counter_init(sbi->ll_stats,
 879                                      llite_opcode_table[id].opcode,
 880                                      (type & LPROCFS_CNTR_AVGMINMAX),
 881                                      llite_opcode_table[id].opname, ptr);
 882         }
 883         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 884         if (err)
 885                 GOTO(out, err);
 886
 887         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 888         if (err)
 889                 GOTO(out, err);
 890
 891         /* MDC info */
 892         obd = class_name2obd(mdc);
 893
 894         LASSERT(obd != NULL);
 895         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 896         LASSERT(obd->obd_type->typ_name != NULL);
 897
 898         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 899                  obd->obd_type->typ_name);
 900         lvars[0].read_fptr = lprocfs_rd_name;
 901         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 902         if (err)
 903                 GOTO(out, err);
 904
 905         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 906         lvars[0].read_fptr = lprocfs_rd_uuid;
 907         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 908         if (err)
 909                 GOTO(out, err);
 910
 911         /* OSC */
 912         obd = class_name2obd(osc);
 913
 914         LASSERT(obd != NULL);
 915         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 916         LASSERT(obd->obd_type->typ_name != NULL);
 917
 918         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 919                  obd->obd_type->typ_name);
 920         lvars[0].read_fptr = lprocfs_rd_name;
 921         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 922         if (err)
 923                 GOTO(out, err);
 924
 925         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 926         lvars[0].read_fptr = lprocfs_rd_uuid;
 927         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 928 out:
 929         if (err) {
 930                 lprocfs_remove(&sbi->ll_proc_root);
 931                 lprocfs_free_stats(&sbi->ll_stats);
 932         }
 933         RETURN(err);
 934 }
 935
 936 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 937 {
 938         if (sbi->ll_proc_root) {
 939                 lprocfs_remove(&sbi->ll_proc_root);
 940                 lprocfs_free_stats(&sbi->ll_ra_stats);
 941                 lprocfs_free_stats(&sbi->ll_stats);
 942         }
 943 }
 944 #undef MAX_STRING_SIZE
 945
 946 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 947                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 948                         if (!has_flags)                                 \
 949                                 has_flags = 1;                          \
 950                         else                                            \
 951                                 seq_putc(seq, '|');                     \
 952                         seq_puts(seq, #flag);                           \
 953                 }                                                       \
 954         } while(0)
 955
 956 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 957 {
 958         struct ll_async_page *dummy_llap = seq->private;
 959
 960         if (dummy_llap->llap_magic == 2)
 961                 return NULL;
 962
 963         return (void *)1;
 964 }
 965
 966 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 967 {
 968         struct ll_async_page *llap, *dummy_llap = seq->private;
 969         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 970         struct ll_pglist_data *pd;
 971         int cpu = dummy_llap->llap_pglist_cpu;
 972
 973         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 974          * it in our own state */
 975         if (dummy_llap->llap_magic == 0) {
 976                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
 977                                 "inode index count [ page flags ]\n");
 978                 return 0;
 979         }
 980
 981         pd = ll_pglist_cpu_lock(sbi, cpu);
 982         llap = llite_pglist_next_llap(&pd->llpd_list,
 983                                       &dummy_llap->llap_pglist_item);
 984         if (llap != NULL)  {
 985                 int has_flags = 0, i;
 986                 struct page *page = llap->llap_page;
 987                 unsigned long gen = 0UL;
 988
 989                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 990                          llap->llap_origin);
 991
 992                 for_each_online_cpu(i)
 993                          gen += LL_PGLIST_DATA_CPU(sbi, i)->llpd_gen;
 994
 995                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
 996                            "%lu %u [",
 997                            gen,
 998                            llap, llap->llap_cookie,
 999                            llap_origins[llap->llap_origin],
1000                            llap->llap_write_queued ? "wq" : "- ",
1001                            llap->llap_defer_uptodate ? "du" : "- ",
1002                            PageWriteback(page) ? "wb" : "-",
1003                            page, page->mapping->host->i_ino,
1004                            page->mapping->host->i_generation,
1005                            page->mapping->host, page->index,
1006                            page_count(page));
1007                 seq_page_flag(seq, page, locked, has_flags);
1008                 seq_page_flag(seq, page, error, has_flags);
1009                 seq_page_flag(seq, page, referenced, has_flags);
1010                 seq_page_flag(seq, page, uptodate, has_flags);
1011                 seq_page_flag(seq, page, dirty, has_flags);
1012 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
1013                 seq_page_flag(seq, page, highmem, has_flags);
1014 #endif
1015                 seq_page_flag(seq, page, writeback, has_flags);
1016                 if (!has_flags)
1017                         seq_puts(seq, "-]\n");
1018                 else
1019                         seq_puts(seq, "]\n");
1020         }
1021         ll_pglist_cpu_unlock(sbi, cpu);
1022
1023         return 0;
1024 }
1025
1026 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
1027                                          loff_t *pos)
1028 {
1029         struct ll_async_page *llap, *dummy_llap = seq->private;
1030         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1031         struct ll_pglist_data *pd, *next;
1032         int cpu = dummy_llap->llap_pglist_cpu;
1033
1034         /* bail if we just displayed the banner */
1035         if (dummy_llap->llap_magic == 0) {
1036                 dummy_llap->llap_magic = 1;
1037                 return dummy_llap;
1038         }
1039
1040         /* we've just displayed the llap that is after us in the list.
1041          * we advance to a position beyond it, returning null if there
1042          * isn't another llap in the list beyond that new position. */
1043         pd = ll_pglist_cpu_lock(sbi, cpu);
1044         llap = llite_pglist_next_llap(&pd->llpd_list,
1045                         &dummy_llap->llap_pglist_item);
1046         list_del_init(&dummy_llap->llap_pglist_item);
1047         if (llap) {
1048                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
1049                 llap = llite_pglist_next_llap(&pd->llpd_list,
1050                                 &dummy_llap->llap_pglist_item);
1051         }
1052         if (llap == NULL) {
1053                 int i = cpu + 1;
1054                 for (next = NULL; i < num_possible_cpus(); i++, next = NULL) {
1055                         next = ll_pglist_cpu_lock(sbi, i);
1056                         if (!list_empty(&next->llpd_list))
1057                                 break;
1058                         ll_pglist_cpu_unlock(sbi, i);
1059                 }
1060                 if (next != NULL) {
1061                         list_move(&dummy_llap->llap_pglist_item,
1062                                   &next->llpd_list);
1063                         dummy_llap->llap_pglist_cpu = i;
1064                         ll_pglist_cpu_unlock(sbi, cpu);
1065                         llap = llite_pglist_next_llap(&next->llpd_list,
1066                                         &dummy_llap->llap_pglist_item);
1067                         LASSERT(llap);
1068                         cpu = i;
1069                 }
1070         }
1071         ll_pglist_cpu_unlock(sbi, cpu);
1072
1073         ++*pos;
1074         if (llap == NULL) {
1075                 dummy_llap->llap_magic = 2;
1076                 return NULL;
1077         }
1078         return dummy_llap;
1079 }
1080
1081 static void null_stop(struct seq_file *seq, void *v)
1082 {
1083 }
1084
1085 struct seq_operations llite_dump_pgcache_seq_sops = {
1086         .start = llite_dump_pgcache_seq_start,
1087         .stop = null_stop,
1088         .next = llite_dump_pgcache_seq_next,
1089         .show = llite_dump_pgcache_seq_show,
1090 };
1091
1092 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
1093  * while we walk the entire list, and we don't want to have to seek into
1094  * the right position in the list as an app advances with many syscalls.  we
1095  * allocate a dummy llap and hang it off file->private.  its position in
1096  * the list records where the app is currently displaying.  this way our
1097  * seq .start and .stop don't actually do anything.  .next returns null
1098  * when the dummy hits the end of the list which eventually leads to .release
1099  * where we tear down.  this kind of displaying is super-racey, so we put
1100  * a generation counter on the list so the output shows when the list
1101  * changes between reads.
1102  */
1103 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
1104 {
1105         struct proc_dir_entry *dp = PDE(inode);
1106         struct ll_async_page *dummy_llap;
1107         struct seq_file *seq;
1108         struct ll_sb_info *sbi = dp->data;
1109         struct ll_pglist_data *pd;
1110         int rc = -ENOMEM;
1111
1112         LPROCFS_ENTRY_AND_CHECK(dp);
1113
1114         OBD_ALLOC_PTR_WAIT(dummy_llap);
1115         if (dummy_llap == NULL)
1116                 GOTO(out, rc);
1117
1118         dummy_llap->llap_page = NULL;
1119         dummy_llap->llap_cookie = sbi;
1120         dummy_llap->llap_magic = 0;
1121         dummy_llap->llap_pglist_cpu = 0;
1122
1123         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
1124         if (rc) {
1125                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1126                 GOTO(out, rc);
1127         }
1128         seq = file->private_data;
1129         seq->private = dummy_llap;
1130
1131         pd = ll_pglist_cpu_lock(sbi, 0);
1132         list_add(&dummy_llap->llap_pglist_item, &pd->llpd_list);
1133         ll_pglist_cpu_unlock(sbi, 0);
1134
1135 out:
1136         if (rc)
1137                 LPROCFS_EXIT();
1138         return rc;
1139 }
1140
1141 static int llite_dump_pgcache_seq_release(struct inode *inode,
1142                                           struct file *file)
1143 {
1144         struct seq_file *seq = file->private_data;
1145         struct ll_async_page *dummy_llap = seq->private;
1146         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1147         int cpu = dummy_llap->llap_pglist_cpu;
1148
1149         ll_pglist_cpu_lock(sbi, cpu);
1150         if (!list_empty(&dummy_llap->llap_pglist_item))
1151                 list_del_init(&dummy_llap->llap_pglist_item);
1152         ll_pglist_cpu_unlock(sbi, cpu);
1153         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1154
1155         return lprocfs_seq_release(inode, file);
1156 }
1157
1158 struct file_operations llite_dump_pgcache_fops = {
1159         .owner   = THIS_MODULE,
1160         .open    = llite_dump_pgcache_seq_open,
1161         .read    = seq_read,
1162         .release = llite_dump_pgcache_seq_release,
1163 };
1164
1165 #define pct(a,b) (b ? a * 100 / b : 0)
1166
1167 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1168                                    struct seq_file *seq, int which)
1169 {
1170         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1171         unsigned long start, end, r, w;
1172         char *unitp = "KMGTPEZY";
1173         int i, units = 10;
1174         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1175
1176         read_cum = 0;
1177         write_cum = 0;
1178         start = 0;
1179
1180         for(i = 0; i < LL_HIST_MAX; i++) {
1181                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1182                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1183         }
1184
1185         for(i = 0; i < LL_HIST_MAX; i++) {
1186                 r = pp_info->pp_r_hist.oh_buckets[i];
1187                 w = pp_info->pp_w_hist.oh_buckets[i];
1188                 read_cum += r;
1189                 write_cum += w;
1190                 end = 1 << (i + LL_HIST_START - units);
1191                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1192                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1193                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1194                            r, pct(r, read_tot), pct(read_cum, read_tot),
1195                            w, pct(w, write_tot), pct(write_cum, write_tot));
1196                 start = end;
1197                 if (start == 1<<10) {
1198                         start = 1;
1199                         units += 10;
1200                         unitp++;
1201                 }
1202                 if (read_cum == read_tot && write_cum == write_tot)
1203                         break;
1204         }
1205 }
1206
1207 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1208 {
1209         struct timeval now;
1210         struct ll_sb_info *sbi = seq->private;
1211         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1212         int k;
1213
1214         do_gettimeofday(&now);
1215
1216         if (!sbi->ll_rw_stats_on) {
1217                 seq_printf(seq, "disabled\n"
1218                                 "write anything in this file to activate, "
1219                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1220                 return 0;
1221         }
1222         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1223                    now.tv_sec, now.tv_usec);
1224         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1225         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1226                    "extents", "calls", "%", "cum%",
1227                    "calls", "%", "cum%");
1228         spin_lock(&sbi->ll_pp_extent_lock);
1229         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1230                 if(io_extents->pp_extents[k].pid != 0) {
1231                         seq_printf(seq, "\nPID: %d\n",
1232                                    io_extents->pp_extents[k].pid);
1233                         ll_display_extents_info(io_extents, seq, k);
1234                 }
1235         }
1236         spin_unlock(&sbi->ll_pp_extent_lock);
1237         return 0;
1238 }
1239
1240 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1241                                                 const char *buf, size_t len,
1242                                                 loff_t *off)
1243 {
1244         struct seq_file *seq = file->private_data;
1245         struct ll_sb_info *sbi = seq->private;
1246         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1247         int i;
1248         int value = 1, rc = 0;
1249
1250         rc = lprocfs_write_helper(buf, len, &value);
1251         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1252                        strcmp(buf, "Disabled") == 0))
1253                 value = 0;
1254
1255         if (value == 0)
1256                 sbi->ll_rw_stats_on = 0;
1257         else
1258                 sbi->ll_rw_stats_on = 1;
1259
1260         spin_lock(&sbi->ll_pp_extent_lock);
1261         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1262                 io_extents->pp_extents[i].pid = 0;
1263                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1264                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1265         }
1266         spin_unlock(&sbi->ll_pp_extent_lock);
1267         return len;
1268 }
1269
1270 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1271
1272 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1273 {
1274         struct timeval now;
1275         struct ll_sb_info *sbi = seq->private;
1276         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1277
1278         do_gettimeofday(&now);
1279
1280         if (!sbi->ll_rw_stats_on) {
1281                 seq_printf(seq, "disabled\n"
1282                                 "write anything in this file to activate, "
1283                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1284                 return 0;
1285         }
1286         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1287                    now.tv_sec, now.tv_usec);
1288
1289         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1290         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1291                    "extents", "calls", "%", "cum%",
1292                    "calls", "%", "cum%");
1293         spin_lock(&sbi->ll_lock);
1294         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1295         spin_unlock(&sbi->ll_lock);
1296
1297         return 0;
1298 }
1299
1300 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1301                                         size_t len, loff_t *off)
1302 {
1303         struct seq_file *seq = file->private_data;
1304         struct ll_sb_info *sbi = seq->private;
1305         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1306         int i;
1307         int value = 1, rc = 0;
1308
1309         rc = lprocfs_write_helper(buf, len, &value);
1310         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1311                        strcmp(buf, "Disabled") == 0))
1312                 value = 0;
1313
1314         if (value == 0)
1315                 sbi->ll_rw_stats_on = 0;
1316         else
1317                 sbi->ll_rw_stats_on = 1;
1318         spin_lock(&sbi->ll_pp_extent_lock);
1319         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1320         {
1321                 io_extents->pp_extents[i].pid = 0;
1322                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1323                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1324         }
1325         spin_unlock(&sbi->ll_pp_extent_lock);
1326
1327         return len;
1328 }
1329
1330 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1331
1332 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1333                                *file, size_t count, int rw)
1334 {
1335         int i, cur = -1;
1336         struct ll_rw_process_info *process;
1337         struct ll_rw_process_info *offset;
1338         int *off_count = &sbi->ll_rw_offset_entry_count;
1339         int *process_count = &sbi->ll_offset_process_count;
1340         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1341
1342         if(!sbi->ll_rw_stats_on)
1343                 return;
1344         process = sbi->ll_rw_process_info;
1345         offset = sbi->ll_rw_offset_info;
1346
1347         spin_lock(&sbi->ll_pp_extent_lock);
1348         /* Extent statistics */
1349         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1350                 if(io_extents->pp_extents[i].pid == pid) {
1351                         cur = i;
1352                         break;
1353                 }
1354         }
1355
1356         if (cur == -1) {
1357                 /* new process */
1358                 sbi->ll_extent_process_count =
1359                         (sbi->ll_extent_process_count+1) % LL_PROCESS_HIST_MAX;
1360                 cur = sbi->ll_extent_process_count;
1361                 io_extents->pp_extents[cur].pid = pid;
1362                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1363                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1364         }
1365
1366         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1367              (i < (LL_HIST_MAX - 1)); i++);
1368         if (rw == 0) {
1369                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1370                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1371         } else {
1372                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1373                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1374         }
1375         spin_unlock(&sbi->ll_pp_extent_lock);
1376
1377         spin_lock(&sbi->ll_process_lock);
1378         /* Offset statistics */
1379         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1380                 if (process[i].rw_pid == pid) {
1381                         if (process[i].rw_last_file != file) {
1382                                 process[i].rw_range_start = file->f_pos;
1383                                 process[i].rw_last_file_pos =
1384                                                         file->f_pos + count;
1385                                 process[i].rw_smallest_extent = count;
1386                                 process[i].rw_largest_extent = count;
1387                                 process[i].rw_offset = 0;
1388                                 process[i].rw_last_file = file;
1389                                 spin_unlock(&sbi->ll_process_lock);
1390                                 return;
1391                         }
1392                         if (process[i].rw_last_file_pos != file->f_pos) {
1393                                 *off_count =
1394                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1395                                 offset[*off_count].rw_op = process[i].rw_op;
1396                                 offset[*off_count].rw_pid = pid;
1397                                 offset[*off_count].rw_range_start =
1398                                         process[i].rw_range_start;
1399                                 offset[*off_count].rw_range_end =
1400                                         process[i].rw_last_file_pos;
1401                                 offset[*off_count].rw_smallest_extent =
1402                                         process[i].rw_smallest_extent;
1403                                 offset[*off_count].rw_largest_extent =
1404                                         process[i].rw_largest_extent;
1405                                 offset[*off_count].rw_offset =
1406                                         process[i].rw_offset;
1407                                 process[i].rw_op = rw;
1408                                 process[i].rw_range_start = file->f_pos;
1409                                 process[i].rw_smallest_extent = count;
1410                                 process[i].rw_largest_extent = count;
1411                                 process[i].rw_offset = file->f_pos -
1412                                         process[i].rw_last_file_pos;
1413                         }
1414                         if(process[i].rw_smallest_extent > count)
1415                                 process[i].rw_smallest_extent = count;
1416                         if(process[i].rw_largest_extent < count)
1417                                 process[i].rw_largest_extent = count;
1418                         process[i].rw_last_file_pos = file->f_pos + count;
1419                         spin_unlock(&sbi->ll_process_lock);
1420                         return;
1421                 }
1422         }
1423         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1424         process[*process_count].rw_pid = pid;
1425         process[*process_count].rw_op = rw;
1426         process[*process_count].rw_range_start = file->f_pos;
1427         process[*process_count].rw_last_file_pos = file->f_pos + count;
1428         process[*process_count].rw_smallest_extent = count;
1429         process[*process_count].rw_largest_extent = count;
1430         process[*process_count].rw_offset = 0;
1431         process[*process_count].rw_last_file = file;
1432         spin_unlock(&sbi->ll_process_lock);
1433 }
1434
1435 char lpszt[] = LPSZ;
1436
1437 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1438 {
1439         struct timeval now;
1440         struct ll_sb_info *sbi = seq->private;
1441         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1442         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1443         char format[50];
1444         int i;
1445
1446         do_gettimeofday(&now);
1447
1448         if (!sbi->ll_rw_stats_on) {
1449                 seq_printf(seq, "disabled\n"
1450                                 "write anything in this file to activate, "
1451                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1452                 return 0;
1453         }
1454         spin_lock(&sbi->ll_process_lock);
1455
1456         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1457                    now.tv_sec, now.tv_usec);
1458         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1459                    "R/W", "PID", "RANGE START", "RANGE END",
1460                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1461         sprintf(format, "%s%s%s%s%s\n",
1462                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1463         /* We stored the discontiguous offsets here; print them first */
1464         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1465                 if (offset[i].rw_pid != 0)
1466                         /* Is there a way to snip the '%' off of LPSZ? */
1467                         seq_printf(seq, format,
1468                                    offset[i].rw_op ? 'W' : 'R',
1469                                    offset[i].rw_pid,
1470                                    offset[i].rw_range_start,
1471                                    offset[i].rw_range_end,
1472                                    offset[i].rw_smallest_extent,
1473                                    offset[i].rw_largest_extent,
1474                                    offset[i].rw_offset);
1475         }
1476         /* Then print the current offsets for each process */
1477         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1478                 if (process[i].rw_pid != 0)
1479                         seq_printf(seq, format,
1480                                    process[i].rw_op ? 'W' : 'R',
1481                                    process[i].rw_pid,
1482                                    process[i].rw_range_start,
1483                                    process[i].rw_last_file_pos,
1484                                    process[i].rw_smallest_extent,
1485                                    process[i].rw_largest_extent,
1486                                    process[i].rw_offset);
1487         }
1488         spin_unlock(&sbi->ll_process_lock);
1489
1490         return 0;
1491 }
1492
1493 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1494                                        size_t len, loff_t *off)
1495 {
1496         struct seq_file *seq = file->private_data;
1497         struct ll_sb_info *sbi = seq->private;
1498         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1499         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1500         int value = 1, rc = 0;
1501
1502         rc = lprocfs_write_helper(buf, len, &value);
1503
1504         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1505                            strcmp(buf, "Disabled") == 0))
1506                 value = 0;
1507
1508         if (value == 0)
1509                 sbi->ll_rw_stats_on = 0;
1510         else
1511                 sbi->ll_rw_stats_on = 1;
1512
1513         spin_lock(&sbi->ll_process_lock);
1514         sbi->ll_offset_process_count = 0;
1515         sbi->ll_rw_offset_entry_count = 0;
1516         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1517                LL_PROCESS_HIST_MAX);
1518         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1519                LL_OFFSET_HIST_MAX);
1520         spin_unlock(&sbi->ll_process_lock);
1521
1522         return len;
1523 }
1524
1525 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1526
1527 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1528 {
1529     lvars->module_vars  = NULL;
1530     lvars->obd_vars     = lprocfs_llite_obd_vars;
1531 }
1532 #endif /* LPROCFS */