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_lockless_direct_io(char *page, char **start, off_t off,
 601                                     int count, int *eof, void *data)
 602 {
 603         struct super_block *sb = data;
 604
 605         *eof = 1;
 606         return snprintf(page, count, "%u\n",
 607                         ll_s2sbi(sb)->ll_lockless_direct_io);
 608 }
 609
 610 static int ll_wr_lockless_direct_io(struct file *file, const char *buffer,
 611                                     unsigned long count, void *data)
 612 {
 613         struct super_block *sb = data;
 614         struct ll_sb_info *sbi = ll_s2sbi(sb);
 615
 616         return lprocfs_write_helper(buffer, count,
 617                                     &sbi->ll_lockless_direct_io)
 618                 ?: count;
 619 }
 620
 621 static int ll_rd_statahead_max(char *page, char **start, off_t off,
 622                                int count, int *eof, void *data)
 623 {
 624         struct super_block *sb = data;
 625         struct ll_sb_info *sbi = ll_s2sbi(sb);
 626
 627         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
 628 }
 629
 630 static int ll_wr_statahead_max(struct file *file, const char *buffer,
 631                                unsigned long count, void *data)
 632 {
 633         struct super_block *sb = data;
 634         struct ll_sb_info *sbi = ll_s2sbi(sb);
 635         int val, rc;
 636
 637         rc = lprocfs_write_helper(buffer, count, &val);
 638         if (rc)
 639                 return rc;
 640
 641         if (val >= 0 && val <= LL_SA_RPC_MAX)
 642                 sbi->ll_sa_max = val;
 643         else
 644                 CERROR("Bad statahead_max value %d. Valid values are in the "
 645                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
 646
 647         return count;
 648 }
 649
 650 static int ll_rd_statahead_stats(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,
 657                         "statahead wrong: %u\n"
 658                         "statahead total: %u\n"
 659                         "ls blocked:      %llu\n"
 660                         "ls cached:       %llu\n"
 661                         "hit count:       %llu\n"
 662                         "miss count:      %llu\n",
 663                         sbi->ll_sa_wrong,
 664                         sbi->ll_sa_total,
 665                         sbi->ll_sa_blocked,
 666                         sbi->ll_sa_cached,
 667                         sbi->ll_sa_hit,
 668                         sbi->ll_sa_miss);
 669 }
 670
 671 static int ll_rd_lazystatfs(char *page, char **start, off_t off,
 672                             int count, int *eof, void *data)
 673 {
 674         struct super_block *sb = data;
 675         struct ll_sb_info *sbi = ll_s2sbi(sb);
 676
 677         return snprintf(page, count, "%u\n",
 678                         (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 679 }
 680
 681 static int ll_wr_lazystatfs(struct file *file, const char *buffer,
 682                             unsigned long count, void *data)
 683 {
 684         struct super_block *sb = data;
 685         struct ll_sb_info *sbi = ll_s2sbi(sb);
 686         int val, rc;
 687
 688         rc = lprocfs_write_helper(buffer, count, &val);
 689         if (rc)
 690                 return rc;
 691
 692         if (val)
 693                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 694         else
 695                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 696
 697         return count;
 698 }
 699
 700 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 701         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 702         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 703         { "fstype",       ll_rd_fstype,           0, 0 },
 704         { "blocksize",    ll_rd_blksize,          0, 0 },
 705         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 706         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 707         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 708         { "filestotal",   ll_rd_filestotal,       0, 0 },
 709         { "filesfree",    ll_rd_filesfree,        0, 0 },
 710         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 711         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 712                                ll_wr_max_readahead_mb, 0 },
 713         { "max_read_ahead_per_file_mb", ll_rd_max_readahead_per_file_mb,
 714                                         ll_wr_max_readahead_per_file_mb, 0 },
 715         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 716                                      ll_wr_max_read_ahead_whole_mb, 0 },
 717         { "max_cached_mb",  ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 718         { "pgcache_balance",ll_rd_pgcache_balance, 0, 0 },
 719         { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
 720         { "max_rw_chunk",   ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 721         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 722         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 723         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 724         { "contention_seconds", ll_rd_contention_time,
 725                                 ll_wr_contention_time, 0},
 726         { "lockless_truncate", ll_rd_lockless_truncate,
 727                                ll_wr_lockless_truncate, 0},
 728         { "lockless_direct_io", ll_rd_lockless_direct_io,
 729                                ll_wr_lockless_direct_io, 0},
 730         { "direct_io_default", ll_rd_direct_io_default,
 731                                ll_wr_direct_io_default, 0},
 732         { "statahead_max",      ll_rd_statahead_max, ll_wr_statahead_max, 0 },
 733         { "statahead_stats",    ll_rd_statahead_stats, 0, 0 },
 734         { "lazystatfs",         ll_rd_lazystatfs, ll_wr_lazystatfs, 0 },
 735         { 0 }
 736 };
 737
 738 #define MAX_STRING_SIZE 128
 739
 740 struct llite_file_opcode {
 741         __u32       opcode;
 742         __u32       type;
 743         const char *opname;
 744 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 745         /* file operation */
 746         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 747         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 748         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 749                                    "writeback_from_writepage" },
 750         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 751                                    "writeback_from_pressure" },
 752         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 753                                    "writeback_ok_pages" },
 754         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 755                                    "writeback_failed_pages" },
 756         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 757                                    "read_bytes" },
 758         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 759                                    "write_bytes" },
 760         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 761                                    "brw_read" },
 762         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 763                                    "brw_write" },
 764
 765         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 766         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 767         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 768         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 769         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 770         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 771         /* inode operation */
 772         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 773         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 774         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate" },
 775         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 776         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 777         /* special inode operation */
 778         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 779         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 780         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 781         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 782         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 783         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 784         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 785         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 786                                    "direct_read" },
 787         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 788                                    "direct_write" },
 789         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 790                                    "lockless_read_bytes" },
 791         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 792                                    "lockless_write_bytes" },
 793
 794 };
 795
 796 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 797 {
 798         if (!sbi->ll_stats)
 799                 return;
 800         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 801                 lprocfs_counter_add(sbi->ll_stats, op, count);
 802         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 803                  sbi->ll_stats_track_id == current->pid)
 804                 lprocfs_counter_add(sbi->ll_stats, op, count);
 805         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 806                  sbi->ll_stats_track_id == current->parent->pid)
 807                 lprocfs_counter_add(sbi->ll_stats, op, count);
 808         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 809                  sbi->ll_stats_track_id == cfs_curproc_gid())
 810                 lprocfs_counter_add(sbi->ll_stats, op, count);
 811 }
 812 EXPORT_SYMBOL(ll_stats_ops_tally);
 813
 814 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 815                                 struct super_block *sb, char *osc, char *mdc)
 816 {
 817         struct lprocfs_vars lvars[2];
 818         struct lustre_sb_info *lsi = s2lsi(sb);
 819         struct ll_sb_info *sbi = ll_s2sbi(sb);
 820         struct obd_device *obd;
 821         char name[MAX_STRING_SIZE + 1], *ptr;
 822         int err, id, len;
 823         struct proc_dir_entry *entry;
 824         static const char *ra_stats_string[] = LL_RA_STAT_STRINGS;
 825         ENTRY;
 826
 827         memset(lvars, 0, sizeof(lvars));
 828
 829         name[MAX_STRING_SIZE] = '\0';
 830         lvars[0].name = name;
 831
 832         LASSERT(sbi != NULL);
 833         LASSERT(mdc != NULL);
 834         LASSERT(osc != NULL);
 835
 836         /* Get fsname */
 837         len = strlen(lsi->lsi_lmd->lmd_profile);
 838         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 839         if (ptr && (strcmp(ptr, "-client") == 0))
 840                 len -= 7;
 841
 842         /* Mount info */
 843         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 844                  lsi->lsi_lmd->lmd_profile, sb);
 845
 846         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 847         if (IS_ERR(sbi->ll_proc_root)) {
 848                 err = PTR_ERR(sbi->ll_proc_root);
 849                 sbi->ll_proc_root = NULL;
 850                 RETURN(err);
 851         }
 852
 853         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 854         if (entry == NULL)
 855                 GOTO(out, err = -ENOMEM);
 856         entry->proc_fops = &llite_dump_pgcache_fops;
 857         entry->data = sbi;
 858
 859         sbi->ll_ra_stats = lprocfs_alloc_stats(LL_RA_STAT,
 860                                                LPROCFS_STATS_FLAG_NONE);
 861         for (id = 0; id < LL_RA_STAT; id++)
 862                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
 863                         ra_stats_string[id], "pages");
 864         lprocfs_register_stats(sbi->ll_proc_root, "read_ahead_stats",
 865                 sbi->ll_ra_stats);
 866
 867         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 868         if (entry == NULL)
 869                  GOTO(out, err = -ENOMEM);
 870         entry->proc_fops = &ll_rw_extents_stats_fops;
 871         entry->data = sbi;
 872
 873         entry = create_proc_entry("extents_stats_per_process", 0644,
 874                                   sbi->ll_proc_root);
 875         if (entry == NULL)
 876                  GOTO(out, err = -ENOMEM);
 877         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 878         entry->data = sbi;
 879
 880         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 881         if (entry == NULL)
 882                 GOTO(out, err = -ENOMEM);
 883         entry->proc_fops = &ll_rw_offset_stats_fops;
 884         entry->data = sbi;
 885
 886         /* File operations stats */
 887         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 888                                             LPROCFS_STATS_FLAG_NONE);
 889         if (sbi->ll_stats == NULL)
 890                 GOTO(out, err = -ENOMEM);
 891         /* do counter init */
 892         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 893                 __u32 type = llite_opcode_table[id].type;
 894                 void *ptr = NULL;
 895                 if (type & LPROCFS_TYPE_REGS)
 896                         ptr = "regs";
 897                 else if (type & LPROCFS_TYPE_BYTES)
 898                         ptr = "bytes";
 899                 else if (type & LPROCFS_TYPE_PAGES)
 900                         ptr = "pages";
 901                 lprocfs_counter_init(sbi->ll_stats,
 902                                      llite_opcode_table[id].opcode,
 903                                      (type & LPROCFS_CNTR_AVGMINMAX),
 904                                      llite_opcode_table[id].opname, ptr);
 905         }
 906         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 907         if (err)
 908                 GOTO(out, err);
 909
 910         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 911         if (err)
 912                 GOTO(out, err);
 913
 914         /* MDC info */
 915         obd = class_name2obd(mdc);
 916
 917         LASSERT(obd != NULL);
 918         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 919         LASSERT(obd->obd_type->typ_name != NULL);
 920
 921         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 922                  obd->obd_type->typ_name);
 923         lvars[0].read_fptr = lprocfs_rd_name;
 924         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 925         if (err)
 926                 GOTO(out, err);
 927
 928         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 929         lvars[0].read_fptr = lprocfs_rd_uuid;
 930         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 931         if (err)
 932                 GOTO(out, err);
 933
 934         /* OSC */
 935         obd = class_name2obd(osc);
 936
 937         LASSERT(obd != NULL);
 938         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 939         LASSERT(obd->obd_type->typ_name != NULL);
 940
 941         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 942                  obd->obd_type->typ_name);
 943         lvars[0].read_fptr = lprocfs_rd_name;
 944         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 945         if (err)
 946                 GOTO(out, err);
 947
 948         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 949         lvars[0].read_fptr = lprocfs_rd_uuid;
 950         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 951 out:
 952         if (err) {
 953                 lprocfs_remove(&sbi->ll_proc_root);
 954                 lprocfs_free_stats(&sbi->ll_stats);
 955         }
 956         RETURN(err);
 957 }
 958
 959 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 960 {
 961         if (sbi->ll_proc_root) {
 962                 lprocfs_remove(&sbi->ll_proc_root);
 963                 lprocfs_free_stats(&sbi->ll_ra_stats);
 964                 lprocfs_free_stats(&sbi->ll_stats);
 965         }
 966 }
 967 #undef MAX_STRING_SIZE
 968
 969 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 970                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 971                         if (!has_flags)                                 \
 972                                 has_flags = 1;                          \
 973                         else                                            \
 974                                 seq_putc(seq, '|');                     \
 975                         seq_puts(seq, #flag);                           \
 976                 }                                                       \
 977         } while(0)
 978
 979 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 980 {
 981         struct ll_async_page *dummy_llap = seq->private;
 982
 983         if (dummy_llap->llap_magic == 2)
 984                 return NULL;
 985
 986         return (void *)1;
 987 }
 988
 989 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 990 {
 991         struct ll_async_page *llap, *dummy_llap = seq->private;
 992         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 993         struct ll_pglist_data *pd;
 994         int cpu = dummy_llap->llap_pglist_cpu;
 995
 996         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 997          * it in our own state */
 998         if (dummy_llap->llap_magic == 0) {
 999                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
1000                                 "inode index count [ page flags ]\n");
1001                 return 0;
1002         }
1003
1004         pd = ll_pglist_cpu_lock(sbi, cpu);
1005         llap = llite_pglist_next_llap(&pd->llpd_list,
1006                                       &dummy_llap->llap_pglist_item);
1007         if (llap != NULL)  {
1008                 int has_flags = 0, i;
1009                 struct page *page = llap->llap_page;
1010                 unsigned long gen = 0UL;
1011
1012                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
1013                          llap->llap_origin);
1014
1015                 for_each_online_cpu(i)
1016                          gen += LL_PGLIST_DATA_CPU(sbi, i)->llpd_gen;
1017
1018                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
1019                            "%lu %u [",
1020                            gen,
1021                            llap, llap->llap_cookie,
1022                            llap_origins[llap->llap_origin],
1023                            llap->llap_write_queued ? "wq" : "- ",
1024                            llap->llap_defer_uptodate ? "du" : "- ",
1025                            PageWriteback(page) ? "wb" : "-",
1026                            page, page->mapping->host->i_ino,
1027                            page->mapping->host->i_generation,
1028                            page->mapping->host, page->index,
1029                            page_count(page));
1030                 seq_page_flag(seq, page, locked, has_flags);
1031                 seq_page_flag(seq, page, error, has_flags);
1032                 seq_page_flag(seq, page, referenced, has_flags);
1033                 seq_page_flag(seq, page, uptodate, has_flags);
1034                 seq_page_flag(seq, page, dirty, has_flags);
1035 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
1036                 seq_page_flag(seq, page, highmem, has_flags);
1037 #endif
1038                 seq_page_flag(seq, page, writeback, has_flags);
1039                 if (!has_flags)
1040                         seq_puts(seq, "-]\n");
1041                 else
1042                         seq_puts(seq, "]\n");
1043         }
1044         ll_pglist_cpu_unlock(sbi, cpu);
1045
1046         return 0;
1047 }
1048
1049 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
1050                                          loff_t *pos)
1051 {
1052         struct ll_async_page *llap, *dummy_llap = seq->private;
1053         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1054         struct ll_pglist_data *pd, *next;
1055         int cpu = dummy_llap->llap_pglist_cpu;
1056
1057         /* bail if we just displayed the banner */
1058         if (dummy_llap->llap_magic == 0) {
1059                 dummy_llap->llap_magic = 1;
1060                 return dummy_llap;
1061         }
1062
1063         /* we've just displayed the llap that is after us in the list.
1064          * we advance to a position beyond it, returning null if there
1065          * isn't another llap in the list beyond that new position. */
1066         pd = ll_pglist_cpu_lock(sbi, cpu);
1067         llap = llite_pglist_next_llap(&pd->llpd_list,
1068                         &dummy_llap->llap_pglist_item);
1069         list_del_init(&dummy_llap->llap_pglist_item);
1070         if (llap) {
1071                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
1072                 llap = llite_pglist_next_llap(&pd->llpd_list,
1073                                 &dummy_llap->llap_pglist_item);
1074         }
1075         if (llap == NULL) {
1076                 int i = cpu + 1;
1077                 for (next = NULL; i < num_possible_cpus(); i++, next = NULL) {
1078                         next = ll_pglist_cpu_lock(sbi, i);
1079                         if (!list_empty(&next->llpd_list))
1080                                 break;
1081                         ll_pglist_cpu_unlock(sbi, i);
1082                 }
1083                 if (next != NULL) {
1084                         list_move(&dummy_llap->llap_pglist_item,
1085                                   &next->llpd_list);
1086                         dummy_llap->llap_pglist_cpu = i;
1087                         ll_pglist_cpu_unlock(sbi, cpu);
1088                         llap = llite_pglist_next_llap(&next->llpd_list,
1089                                         &dummy_llap->llap_pglist_item);
1090                         LASSERT(llap);
1091                         cpu = i;
1092                 }
1093         }
1094         ll_pglist_cpu_unlock(sbi, cpu);
1095
1096         ++*pos;
1097         if (llap == NULL) {
1098                 dummy_llap->llap_magic = 2;
1099                 return NULL;
1100         }
1101         return dummy_llap;
1102 }
1103
1104 static void null_stop(struct seq_file *seq, void *v)
1105 {
1106 }
1107
1108 struct seq_operations llite_dump_pgcache_seq_sops = {
1109         .start = llite_dump_pgcache_seq_start,
1110         .stop = null_stop,
1111         .next = llite_dump_pgcache_seq_next,
1112         .show = llite_dump_pgcache_seq_show,
1113 };
1114
1115 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
1116  * while we walk the entire list, and we don't want to have to seek into
1117  * the right position in the list as an app advances with many syscalls.  we
1118  * allocate a dummy llap and hang it off file->private.  its position in
1119  * the list records where the app is currently displaying.  this way our
1120  * seq .start and .stop don't actually do anything.  .next returns null
1121  * when the dummy hits the end of the list which eventually leads to .release
1122  * where we tear down.  this kind of displaying is super-racey, so we put
1123  * a generation counter on the list so the output shows when the list
1124  * changes between reads.
1125  */
1126 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
1127 {
1128         struct proc_dir_entry *dp = PDE(inode);
1129         struct ll_async_page *dummy_llap;
1130         struct seq_file *seq;
1131         struct ll_sb_info *sbi = dp->data;
1132         struct ll_pglist_data *pd;
1133         int rc = -ENOMEM;
1134
1135         LPROCFS_ENTRY_AND_CHECK(dp);
1136
1137         OBD_ALLOC_PTR_WAIT(dummy_llap);
1138         if (dummy_llap == NULL)
1139                 GOTO(out, rc);
1140
1141         dummy_llap->llap_page = NULL;
1142         dummy_llap->llap_cookie = sbi;
1143         dummy_llap->llap_magic = 0;
1144         dummy_llap->llap_pglist_cpu = 0;
1145
1146         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
1147         if (rc) {
1148                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1149                 GOTO(out, rc);
1150         }
1151         seq = file->private_data;
1152         seq->private = dummy_llap;
1153
1154         pd = ll_pglist_cpu_lock(sbi, 0);
1155         list_add(&dummy_llap->llap_pglist_item, &pd->llpd_list);
1156         ll_pglist_cpu_unlock(sbi, 0);
1157
1158 out:
1159         if (rc)
1160                 LPROCFS_EXIT();
1161         return rc;
1162 }
1163
1164 static int llite_dump_pgcache_seq_release(struct inode *inode,
1165                                           struct file *file)
1166 {
1167         struct seq_file *seq = file->private_data;
1168         struct ll_async_page *dummy_llap = seq->private;
1169         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
1170         int cpu = dummy_llap->llap_pglist_cpu;
1171
1172         ll_pglist_cpu_lock(sbi, cpu);
1173         if (!list_empty(&dummy_llap->llap_pglist_item))
1174                 list_del_init(&dummy_llap->llap_pglist_item);
1175         ll_pglist_cpu_unlock(sbi, cpu);
1176         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
1177
1178         return lprocfs_seq_release(inode, file);
1179 }
1180
1181 struct file_operations llite_dump_pgcache_fops = {
1182         .owner   = THIS_MODULE,
1183         .open    = llite_dump_pgcache_seq_open,
1184         .read    = seq_read,
1185         .release = llite_dump_pgcache_seq_release,
1186 };
1187
1188 #define pct(a,b) (b ? a * 100 / b : 0)
1189
1190 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1191                                    struct seq_file *seq, int which)
1192 {
1193         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1194         unsigned long start, end, r, w;
1195         char *unitp = "KMGTPEZY";
1196         int i, units = 10;
1197         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1198
1199         read_cum = 0;
1200         write_cum = 0;
1201         start = 0;
1202
1203         for(i = 0; i < LL_HIST_MAX; i++) {
1204                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1205                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1206         }
1207
1208         for(i = 0; i < LL_HIST_MAX; i++) {
1209                 r = pp_info->pp_r_hist.oh_buckets[i];
1210                 w = pp_info->pp_w_hist.oh_buckets[i];
1211                 read_cum += r;
1212                 write_cum += w;
1213                 end = 1 << (i + LL_HIST_START - units);
1214                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1215                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1216                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1217                            r, pct(r, read_tot), pct(read_cum, read_tot),
1218                            w, pct(w, write_tot), pct(write_cum, write_tot));
1219                 start = end;
1220                 if (start == 1<<10) {
1221                         start = 1;
1222                         units += 10;
1223                         unitp++;
1224                 }
1225                 if (read_cum == read_tot && write_cum == write_tot)
1226                         break;
1227         }
1228 }
1229
1230 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1231 {
1232         struct timeval now;
1233         struct ll_sb_info *sbi = seq->private;
1234         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1235         int k;
1236
1237         do_gettimeofday(&now);
1238
1239         if (!sbi->ll_rw_stats_on) {
1240                 seq_printf(seq, "disabled\n"
1241                                 "write anything in this file to activate, "
1242                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1243                 return 0;
1244         }
1245         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1246                    now.tv_sec, now.tv_usec);
1247         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1248         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1249                    "extents", "calls", "%", "cum%",
1250                    "calls", "%", "cum%");
1251         spin_lock(&sbi->ll_pp_extent_lock);
1252         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1253                 if(io_extents->pp_extents[k].pid != 0) {
1254                         seq_printf(seq, "\nPID: %d\n",
1255                                    io_extents->pp_extents[k].pid);
1256                         ll_display_extents_info(io_extents, seq, k);
1257                 }
1258         }
1259         spin_unlock(&sbi->ll_pp_extent_lock);
1260         return 0;
1261 }
1262
1263 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1264                                                 const char *buf, size_t len,
1265                                                 loff_t *off)
1266 {
1267         struct seq_file *seq = file->private_data;
1268         struct ll_sb_info *sbi = seq->private;
1269         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1270         int i;
1271         int value = 1, rc = 0;
1272
1273         rc = lprocfs_write_helper(buf, len, &value);
1274         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1275                        strcmp(buf, "Disabled") == 0))
1276                 value = 0;
1277
1278         if (value == 0)
1279                 sbi->ll_rw_stats_on = 0;
1280         else
1281                 sbi->ll_rw_stats_on = 1;
1282
1283         spin_lock(&sbi->ll_pp_extent_lock);
1284         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1285                 io_extents->pp_extents[i].pid = 0;
1286                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1287                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1288         }
1289         spin_unlock(&sbi->ll_pp_extent_lock);
1290         return len;
1291 }
1292
1293 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1294
1295 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1296 {
1297         struct timeval now;
1298         struct ll_sb_info *sbi = seq->private;
1299         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1300
1301         do_gettimeofday(&now);
1302
1303         if (!sbi->ll_rw_stats_on) {
1304                 seq_printf(seq, "disabled\n"
1305                                 "write anything in this file to activate, "
1306                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1307                 return 0;
1308         }
1309         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1310                    now.tv_sec, now.tv_usec);
1311
1312         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1313         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1314                    "extents", "calls", "%", "cum%",
1315                    "calls", "%", "cum%");
1316         spin_lock(&sbi->ll_lock);
1317         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1318         spin_unlock(&sbi->ll_lock);
1319
1320         return 0;
1321 }
1322
1323 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1324                                         size_t len, loff_t *off)
1325 {
1326         struct seq_file *seq = file->private_data;
1327         struct ll_sb_info *sbi = seq->private;
1328         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1329         int i;
1330         int value = 1, rc = 0;
1331
1332         rc = lprocfs_write_helper(buf, len, &value);
1333         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1334                        strcmp(buf, "Disabled") == 0))
1335                 value = 0;
1336
1337         if (value == 0)
1338                 sbi->ll_rw_stats_on = 0;
1339         else
1340                 sbi->ll_rw_stats_on = 1;
1341         spin_lock(&sbi->ll_pp_extent_lock);
1342         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1343         {
1344                 io_extents->pp_extents[i].pid = 0;
1345                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1346                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1347         }
1348         spin_unlock(&sbi->ll_pp_extent_lock);
1349
1350         return len;
1351 }
1352
1353 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1354
1355 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1356                                *file, size_t count, int rw)
1357 {
1358         int i, cur = -1;
1359         struct ll_rw_process_info *process;
1360         struct ll_rw_process_info *offset;
1361         int *off_count = &sbi->ll_rw_offset_entry_count;
1362         int *process_count = &sbi->ll_offset_process_count;
1363         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1364
1365         if(!sbi->ll_rw_stats_on)
1366                 return;
1367         process = sbi->ll_rw_process_info;
1368         offset = sbi->ll_rw_offset_info;
1369
1370         spin_lock(&sbi->ll_pp_extent_lock);
1371         /* Extent statistics */
1372         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1373                 if(io_extents->pp_extents[i].pid == pid) {
1374                         cur = i;
1375                         break;
1376                 }
1377         }
1378
1379         if (cur == -1) {
1380                 /* new process */
1381                 sbi->ll_extent_process_count =
1382                         (sbi->ll_extent_process_count+1) % LL_PROCESS_HIST_MAX;
1383                 cur = sbi->ll_extent_process_count;
1384                 io_extents->pp_extents[cur].pid = pid;
1385                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1386                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1387         }
1388
1389         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1390              (i < (LL_HIST_MAX - 1)); i++);
1391         if (rw == 0) {
1392                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1393                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1394         } else {
1395                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1396                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1397         }
1398         spin_unlock(&sbi->ll_pp_extent_lock);
1399
1400         spin_lock(&sbi->ll_process_lock);
1401         /* Offset statistics */
1402         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1403                 if (process[i].rw_pid == pid) {
1404                         if (process[i].rw_last_file != file) {
1405                                 process[i].rw_range_start = file->f_pos;
1406                                 process[i].rw_last_file_pos =
1407                                                         file->f_pos + count;
1408                                 process[i].rw_smallest_extent = count;
1409                                 process[i].rw_largest_extent = count;
1410                                 process[i].rw_offset = 0;
1411                                 process[i].rw_last_file = file;
1412                                 spin_unlock(&sbi->ll_process_lock);
1413                                 return;
1414                         }
1415                         if (process[i].rw_last_file_pos != file->f_pos) {
1416                                 *off_count =
1417                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1418                                 offset[*off_count].rw_op = process[i].rw_op;
1419                                 offset[*off_count].rw_pid = pid;
1420                                 offset[*off_count].rw_range_start =
1421                                         process[i].rw_range_start;
1422                                 offset[*off_count].rw_range_end =
1423                                         process[i].rw_last_file_pos;
1424                                 offset[*off_count].rw_smallest_extent =
1425                                         process[i].rw_smallest_extent;
1426                                 offset[*off_count].rw_largest_extent =
1427                                         process[i].rw_largest_extent;
1428                                 offset[*off_count].rw_offset =
1429                                         process[i].rw_offset;
1430                                 process[i].rw_op = rw;
1431                                 process[i].rw_range_start = file->f_pos;
1432                                 process[i].rw_smallest_extent = count;
1433                                 process[i].rw_largest_extent = count;
1434                                 process[i].rw_offset = file->f_pos -
1435                                         process[i].rw_last_file_pos;
1436                         }
1437                         if(process[i].rw_smallest_extent > count)
1438                                 process[i].rw_smallest_extent = count;
1439                         if(process[i].rw_largest_extent < count)
1440                                 process[i].rw_largest_extent = count;
1441                         process[i].rw_last_file_pos = file->f_pos + count;
1442                         spin_unlock(&sbi->ll_process_lock);
1443                         return;
1444                 }
1445         }
1446         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1447         process[*process_count].rw_pid = pid;
1448         process[*process_count].rw_op = rw;
1449         process[*process_count].rw_range_start = file->f_pos;
1450         process[*process_count].rw_last_file_pos = file->f_pos + count;
1451         process[*process_count].rw_smallest_extent = count;
1452         process[*process_count].rw_largest_extent = count;
1453         process[*process_count].rw_offset = 0;
1454         process[*process_count].rw_last_file = file;
1455         spin_unlock(&sbi->ll_process_lock);
1456 }
1457
1458 char lpszt[] = LPSZ;
1459
1460 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1461 {
1462         struct timeval now;
1463         struct ll_sb_info *sbi = seq->private;
1464         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1465         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1466         char format[50];
1467         int i;
1468
1469         do_gettimeofday(&now);
1470
1471         if (!sbi->ll_rw_stats_on) {
1472                 seq_printf(seq, "disabled\n"
1473                                 "write anything in this file to activate, "
1474                                 "then 0 or \"[D/d]isabled\" to deactivate\n");
1475                 return 0;
1476         }
1477         spin_lock(&sbi->ll_process_lock);
1478
1479         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1480                    now.tv_sec, now.tv_usec);
1481         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1482                    "R/W", "PID", "RANGE START", "RANGE END",
1483                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1484         sprintf(format, "%s%s%s%s%s\n",
1485                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1486         /* We stored the discontiguous offsets here; print them first */
1487         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1488                 if (offset[i].rw_pid != 0)
1489                         /* Is there a way to snip the '%' off of LPSZ? */
1490                         seq_printf(seq, format,
1491                                    offset[i].rw_op ? 'W' : 'R',
1492                                    offset[i].rw_pid,
1493                                    offset[i].rw_range_start,
1494                                    offset[i].rw_range_end,
1495                                    offset[i].rw_smallest_extent,
1496                                    offset[i].rw_largest_extent,
1497                                    offset[i].rw_offset);
1498         }
1499         /* Then print the current offsets for each process */
1500         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1501                 if (process[i].rw_pid != 0)
1502                         seq_printf(seq, format,
1503                                    process[i].rw_op ? 'W' : 'R',
1504                                    process[i].rw_pid,
1505                                    process[i].rw_range_start,
1506                                    process[i].rw_last_file_pos,
1507                                    process[i].rw_smallest_extent,
1508                                    process[i].rw_largest_extent,
1509                                    process[i].rw_offset);
1510         }
1511         spin_unlock(&sbi->ll_process_lock);
1512
1513         return 0;
1514 }
1515
1516 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1517                                        size_t len, loff_t *off)
1518 {
1519         struct seq_file *seq = file->private_data;
1520         struct ll_sb_info *sbi = seq->private;
1521         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1522         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1523         int value = 1, rc = 0;
1524
1525         rc = lprocfs_write_helper(buf, len, &value);
1526
1527         if (rc < 0 && (strcmp(buf, "disabled") == 0 ||
1528                            strcmp(buf, "Disabled") == 0))
1529                 value = 0;
1530
1531         if (value == 0)
1532                 sbi->ll_rw_stats_on = 0;
1533         else
1534                 sbi->ll_rw_stats_on = 1;
1535
1536         spin_lock(&sbi->ll_process_lock);
1537         sbi->ll_offset_process_count = 0;
1538         sbi->ll_rw_offset_entry_count = 0;
1539         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1540                LL_PROCESS_HIST_MAX);
1541         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1542                LL_OFFSET_HIST_MAX);
1543         spin_unlock(&sbi->ll_process_lock);
1544
1545         return len;
1546 }
1547
1548 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1549
1550 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1551 {
1552     lvars->module_vars  = NULL;
1553     lvars->obd_vars     = lprocfs_llite_obd_vars;
1554 }
1555 #endif /* LPROCFS */