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  *  Copyright (C) 2002 Cluster File Systems, Inc.
   5  *
   6  *   This file is part of Lustre, http://www.lustre.org.
   7  *
   8  *   Lustre is free software; you can redistribute it and/or
   9  *   modify it under the terms of version 2 of the GNU General Public
  10  *   License as published by the Free Software Foundation.
  11  *
  12  *   Lustre is distributed in the hope that it will be useful,
  13  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  *   GNU General Public License for more details.
  16  *
  17  *   You should have received a copy of the GNU General Public License
  18  *   along with Lustre; if not, write to the Free Software
  19  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  20  *
  21  */
  22 #define DEBUG_SUBSYSTEM S_LLITE
  23
  24 #include <linux/version.h>
  25 #include <lustre_lite.h>
  26 #include <lprocfs_status.h>
  27 #include <linux/seq_file.h>
  28 #include <obd_support.h>
  29
  30 #include "llite_internal.h"
  31
  32 struct proc_dir_entry *proc_lustre_fs_root;
  33
  34 #ifdef LPROCFS
  35 /* /proc/lustre/llite mount point registration */
  36 struct file_operations llite_dump_pgcache_fops;
  37 struct file_operations ll_ra_stats_fops;
  38 struct file_operations ll_rw_extents_stats_fops;
  39 struct file_operations ll_rw_extents_stats_pp_fops;
  40 struct file_operations ll_rw_offset_stats_fops;
  41
  42 static int ll_rd_blksize(char *page, char **start, off_t off, int count,
  43                          int *eof, void *data)
  44 {
  45         struct super_block *sb = (struct super_block *)data;
  46         struct obd_statfs osfs;
  47         int rc;
  48
  49         LASSERT(sb != NULL);
  50         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  51                                 OBD_STATFS_NODELAY);
  52         if (!rc) {
  53               *eof = 1;
  54               rc = snprintf(page, count, "%u\n", osfs.os_bsize);
  55         }
  56
  57         return rc;
  58 }
  59
  60 static int ll_rd_kbytestotal(char *page, char **start, off_t off, int count,
  61                              int *eof, void *data)
  62 {
  63         struct super_block *sb = (struct super_block *)data;
  64         struct obd_statfs osfs;
  65         int rc;
  66
  67         LASSERT(sb != NULL);
  68         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  69                                 OBD_STATFS_NODELAY);
  70         if (!rc) {
  71                 __u32 blk_size = osfs.os_bsize >> 10;
  72                 __u64 result = osfs.os_blocks;
  73
  74                 while (blk_size >>= 1)
  75                         result <<= 1;
  76
  77                 *eof = 1;
  78                 rc = snprintf(page, count, LPU64"\n", result);
  79         }
  80         return rc;
  81
  82 }
  83
  84 static int ll_rd_kbytesfree(char *page, char **start, off_t off, int count,
  85                             int *eof, void *data)
  86 {
  87         struct super_block *sb = (struct super_block *)data;
  88         struct obd_statfs osfs;
  89         int rc;
  90
  91         LASSERT(sb != NULL);
  92         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
  93                                 OBD_STATFS_NODELAY);
  94         if (!rc) {
  95                 __u32 blk_size = osfs.os_bsize >> 10;
  96                 __u64 result = osfs.os_bfree;
  97
  98                 while (blk_size >>= 1)
  99                         result <<= 1;
 100
 101                 *eof = 1;
 102                 rc = snprintf(page, count, LPU64"\n", result);
 103         }
 104         return rc;
 105 }
 106
 107 static int ll_rd_kbytesavail(char *page, char **start, off_t off, int count,
 108                              int *eof, void *data)
 109 {
 110         struct super_block *sb = (struct super_block *)data;
 111         struct obd_statfs osfs;
 112         int rc;
 113
 114         LASSERT(sb != NULL);
 115         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 116                                 OBD_STATFS_NODELAY);
 117         if (!rc) {
 118                 __u32 blk_size = osfs.os_bsize >> 10;
 119                 __u64 result = osfs.os_bavail;
 120
 121                 while (blk_size >>= 1)
 122                         result <<= 1;
 123
 124                 *eof = 1;
 125                 rc = snprintf(page, count, LPU64"\n", result);
 126         }
 127         return rc;
 128 }
 129
 130 static int ll_rd_filestotal(char *page, char **start, off_t off, int count,
 131                             int *eof, void *data)
 132 {
 133         struct super_block *sb = (struct super_block *)data;
 134         struct obd_statfs osfs;
 135         int rc;
 136
 137         LASSERT(sb != NULL);
 138         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 139                                 OBD_STATFS_NODELAY);
 140         if (!rc) {
 141                  *eof = 1;
 142                  rc = snprintf(page, count, LPU64"\n", osfs.os_files);
 143         }
 144         return rc;
 145 }
 146
 147 static int ll_rd_filesfree(char *page, char **start, off_t off, int count,
 148                            int *eof, void *data)
 149 {
 150         struct super_block *sb = (struct super_block *)data;
 151         struct obd_statfs osfs;
 152         int rc;
 153
 154         LASSERT(sb != NULL);
 155         rc = ll_statfs_internal(sb, &osfs, cfs_time_current_64() - HZ,
 156                                 OBD_STATFS_NODELAY);
 157         if (!rc) {
 158                  *eof = 1;
 159                  rc = snprintf(page, count, LPU64"\n", osfs.os_ffree);
 160         }
 161         return rc;
 162
 163 }
 164
 165 static int ll_rd_client_type(char *page, char **start, off_t off, int count,
 166                             int *eof, void *data)
 167 {
 168         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)data);
 169         int rc;
 170
 171         LASSERT(sbi != NULL);
 172
 173         *eof = 1;
 174         if (sbi->ll_flags & LL_SBI_RMT_CLIENT)
 175                 rc = snprintf(page, count, "remote client\n");
 176         else
 177                 rc = snprintf(page, count, "local client\n");
 178
 179         return rc;
 180 }
 181
 182 static int ll_rd_fstype(char *page, char **start, off_t off, int count,
 183                         int *eof, void *data)
 184 {
 185         struct super_block *sb = (struct super_block*)data;
 186
 187         LASSERT(sb != NULL);
 188         *eof = 1;
 189         return snprintf(page, count, "%s\n", sb->s_type->name);
 190 }
 191
 192 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
 193                          int *eof, void *data)
 194 {
 195         struct super_block *sb = (struct super_block *)data;
 196
 197         LASSERT(sb != NULL);
 198         *eof = 1;
 199         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
 200 }
 201
 202 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
 203                                    int count, int *eof, void *data)
 204 {
 205         struct super_block *sb = data;
 206         struct ll_sb_info *sbi = ll_s2sbi(sb);
 207         long pages_number;
 208         int mult;
 209
 210         spin_lock(&sbi->ll_lock);
 211         pages_number = sbi->ll_ra_info.ra_max_pages;
 212         spin_unlock(&sbi->ll_lock);
 213
 214         mult = 1 << (20 - PAGE_CACHE_SHIFT);
 215         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 216 }
 217
 218 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
 219                                    unsigned long count, void *data)
 220 {
 221         struct super_block *sb = data;
 222         struct ll_sb_info *sbi = ll_s2sbi(sb);
 223         int mult, rc, pages_number;
 224
 225         mult = 1 << (20 - CFS_PAGE_SHIFT);
 226         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 227         if (rc)
 228                 return rc;
 229
 230         if (pages_number < 0 || pages_number > num_physpages / 2) {
 231                 CERROR("can't set file readahead more than %lu MB\n",
 232                         num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
 233                 return -ERANGE;
 234         }
 235
 236         spin_lock(&sbi->ll_lock);
 237         sbi->ll_ra_info.ra_max_pages = pages_number;
 238         spin_unlock(&sbi->ll_lock);
 239
 240         return count;
 241 }
 242
 243 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
 244                                        int count, int *eof, void *data)
 245 {
 246         struct super_block *sb = data;
 247         struct ll_sb_info *sbi = ll_s2sbi(sb);
 248         long pages_number;
 249         int mult;
 250
 251         spin_lock(&sbi->ll_lock);
 252         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
 253         spin_unlock(&sbi->ll_lock);
 254
 255         mult = 1 << (20 - CFS_PAGE_SHIFT);
 256         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 257 }
 258
 259 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
 260                                        unsigned long count, void *data)
 261 {
 262         struct super_block *sb = data;
 263         struct ll_sb_info *sbi = ll_s2sbi(sb);
 264         int mult, rc, pages_number;
 265
 266         mult = 1 << (20 - CFS_PAGE_SHIFT);
 267         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 268         if (rc)
 269                 return rc;
 270
 271         /* Cap this at the current max readahead window size, the readahead
 272          * algorithm does this anyway so it's pointless to set it larger. */
 273         if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
 274                 CERROR("can't set max_read_ahead_whole_mb more than "
 275                        "max_read_ahead_mb: %lu\n",
 276                        sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
 277                 return -ERANGE;
 278         }
 279
 280         spin_lock(&sbi->ll_lock);
 281         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 282         spin_unlock(&sbi->ll_lock);
 283
 284         return count;
 285 }
 286
 287 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
 288                                int count, int *eof, void *data)
 289 {
 290         struct super_block *sb = data;
 291         struct ll_sb_info *sbi = ll_s2sbi(sb);
 292         long pages_number;
 293         int mult;
 294
 295         spin_lock(&sbi->ll_lock);
 296         pages_number = sbi->ll_async_page_max;
 297         spin_unlock(&sbi->ll_lock);
 298
 299         mult = 1 << (20 - CFS_PAGE_SHIFT);
 300         return lprocfs_read_frac_helper(page, count, pages_number, mult);;
 301 }
 302
 303 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
 304                                   unsigned long count, void *data)
 305 {
 306         struct super_block *sb = data;
 307         struct ll_sb_info *sbi = ll_s2sbi(sb);
 308         int mult, rc, pages_number;
 309
 310         mult = 1 << (20 - CFS_PAGE_SHIFT);
 311         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 312         if (rc)
 313                 return rc;
 314
 315         if (pages_number < 0 || pages_number > num_physpages) {
 316                 CERROR("can't set max cache more than %lu MB\n",
 317                         num_physpages >> (20 - CFS_PAGE_SHIFT));
 318                 return -ERANGE;
 319         }
 320
 321         spin_lock(&sbi->ll_lock);
 322         sbi->ll_async_page_max = pages_number ;
 323         spin_unlock(&sbi->ll_lock);
 324
 325         if (!sbi->ll_dt_exp)
 326                 /* Not set up yet, don't call llap_shrink_cache */
 327                 return count;
 328
 329         if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
 330                 llap_shrink_cache(sbi, 0);
 331
 332         return count;
 333 }
 334
 335 static int ll_rd_checksum(char *page, char **start, off_t off,
 336                           int count, int *eof, void *data)
 337 {
 338         struct super_block *sb = data;
 339         struct ll_sb_info *sbi = ll_s2sbi(sb);
 340
 341         return snprintf(page, count, "%u\n",
 342                         (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 343 }
 344
 345 static int ll_wr_checksum(struct file *file, const char *buffer,
 346                           unsigned long count, void *data)
 347 {
 348         struct super_block *sb = data;
 349         struct ll_sb_info *sbi = ll_s2sbi(sb);
 350         int val, rc;
 351
 352         if (!sbi->ll_dt_exp)
 353                 /* Not set up yet */
 354                 return -EAGAIN;
 355
 356         rc = lprocfs_write_helper(buffer, count, &val);
 357         if (rc)
 358                 return rc;
 359         if (val)
 360                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 361         else
 362                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 363
 364         rc = obd_set_info_async(sbi->ll_dt_exp, strlen("checksum"), "checksum",
 365                                 sizeof(val), &val, NULL);
 366         if (rc)
 367                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 368
 369         return count;
 370 }
 371
 372 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
 373                           int count, int *eof, void *data)
 374 {
 375         struct super_block *sb = data;
 376
 377         return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
 378 }
 379
 380 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
 381                           unsigned long count, void *data)
 382 {
 383         struct super_block *sb = data;
 384         int rc, val;
 385
 386         rc = lprocfs_write_helper(buffer, count, &val);
 387         if (rc)
 388                 return rc;
 389         ll_s2sbi(sb)->ll_max_rw_chunk = val;
 390         return count;
 391 }
 392
 393 static int ll_rd_track_id(char *page, int count, void *data,
 394                           enum stats_track_type type)
 395 {
 396         struct super_block *sb = data;
 397
 398         if (ll_s2sbi(sb)->ll_stats_track_type == type) {
 399                 return snprintf(page, count, "%d\n",
 400                                 ll_s2sbi(sb)->ll_stats_track_id);
 401
 402         } else if (ll_s2sbi(sb)->ll_stats_track_type == STATS_TRACK_ALL) {
 403                 return snprintf(page, count, "0 (all)\n");
 404         } else {
 405                 return snprintf(page, count, "untracked\n");
 406         }
 407 }
 408
 409 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
 410                           enum stats_track_type type)
 411 {
 412         struct super_block *sb = data;
 413         int rc, pid;
 414
 415         rc = lprocfs_write_helper(buffer, count, &pid);
 416         if (rc)
 417                 return rc;
 418         ll_s2sbi(sb)->ll_stats_track_id = pid;
 419         if (pid == 0)
 420                 ll_s2sbi(sb)->ll_stats_track_type = STATS_TRACK_ALL;
 421         else
 422                 ll_s2sbi(sb)->ll_stats_track_type = type;
 423         lprocfs_clear_stats(ll_s2sbi(sb)->ll_stats);
 424         return count;
 425 }
 426
 427 static int ll_rd_track_pid(char *page, char **start, off_t off,
 428                           int count, int *eof, void *data)
 429 {
 430         return (ll_rd_track_id(page, count, data, STATS_TRACK_PID));
 431 }
 432
 433 static int ll_wr_track_pid(struct file *file, const char *buffer,
 434                           unsigned long count, void *data)
 435 {
 436         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PID));
 437 }
 438
 439 static int ll_rd_track_ppid(char *page, char **start, off_t off,
 440                           int count, int *eof, void *data)
 441 {
 442         return (ll_rd_track_id(page, count, data, STATS_TRACK_PPID));
 443 }
 444
 445 static int ll_wr_track_ppid(struct file *file, const char *buffer,
 446                           unsigned long count, void *data)
 447 {
 448         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_PPID));
 449 }
 450
 451 static int ll_rd_track_gid(char *page, char **start, off_t off,
 452                           int count, int *eof, void *data)
 453 {
 454         return (ll_rd_track_id(page, count, data, STATS_TRACK_GID));
 455 }
 456
 457 static int ll_wr_track_gid(struct file *file, const char *buffer,
 458                           unsigned long count, void *data)
 459 {
 460         return (ll_wr_track_id(buffer, count, data, STATS_TRACK_GID));
 461 }
 462
 463 static int ll_rd_statahead_max(char *page, char **start, off_t off,
 464                                int count, int *eof, void *data)
 465 {
 466         struct super_block *sb = data;
 467         struct ll_sb_info *sbi = ll_s2sbi(sb);
 468
 469         return snprintf(page, count, "%u\n", sbi->ll_sa_max);
 470 }
 471
 472 static int ll_wr_statahead_max(struct file *file, const char *buffer,
 473                                unsigned long count, void *data)
 474 {
 475         struct super_block *sb = data;
 476         struct ll_sb_info *sbi = ll_s2sbi(sb);
 477         int val, rc;
 478
 479         rc = lprocfs_write_helper(buffer, count, &val);
 480         if (rc)
 481                 return rc;
 482
 483         if (val >= 0 && val <= LL_SA_RPC_MAX)
 484                 sbi->ll_sa_max = val;
 485         else
 486                 CERROR("Bad statahead_max value %d. Valid values are in the "
 487                        "range [0, %d]\n", val, LL_SA_RPC_MAX);
 488
 489         return count;
 490 }
 491
 492 static int ll_rd_statahead_stats(char *page, char **start, off_t off,
 493                                  int count, int *eof, void *data)
 494 {
 495         struct super_block *sb = data;
 496         struct ll_sb_info *sbi = ll_s2sbi(sb);
 497
 498         return snprintf(page, count,
 499                         "statahead wrong: %u\n"
 500                         "statahead total: %u\n"
 501                         "ls blocked:      %llu\n"
 502                         "ls cached:       %llu\n"
 503                         "hit count:       %llu\n"
 504                         "miss count:      %llu\n",
 505                         sbi->ll_sa_wrong,
 506                         sbi->ll_sa_total,
 507                         sbi->ll_sa_blocked,
 508                         sbi->ll_sa_cached,
 509                         sbi->ll_sa_hit,
 510                         sbi->ll_sa_miss);
 511 }
 512
 513 static int ll_rd_contention_time(char *page, char **start, off_t off,
 514                                  int count, int *eof, void *data)
 515 {
 516         struct super_block *sb = data;
 517
 518         *eof = 1;
 519         return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
 520
 521 }
 522
 523 static int ll_wr_contention_time(struct file *file, const char *buffer,
 524                                  unsigned long count, void *data)
 525 {
 526         struct super_block *sb = data;
 527         struct ll_sb_info *sbi = ll_s2sbi(sb);
 528
 529         return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
 530                 count;
 531 }
 532
 533 static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
 534                                    int count, int *eof, void *data)
 535 {
 536         struct super_block *sb = data;
 537
 538         *eof = 1;
 539         return snprintf(page, count, "%u\n",
 540                         ll_s2sbi(sb)->ll_lockless_truncate_enable);
 541 }
 542
 543 static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
 544                                    unsigned long count, void *data)
 545 {
 546         struct super_block *sb = data;
 547         struct ll_sb_info *sbi = ll_s2sbi(sb);
 548
 549         return lprocfs_write_helper(buffer, count,
 550                                     &sbi->ll_lockless_truncate_enable)
 551                                     ?: count;
 552 }
 553
 554 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 555         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 556         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 557         { "fstype",       ll_rd_fstype,           0, 0 },
 558         { "blocksize",    ll_rd_blksize,          0, 0 },
 559         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 560         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 561         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 562         { "filestotal",   ll_rd_filestotal,       0, 0 },
 563         { "filesfree",    ll_rd_filesfree,        0, 0 },
 564         { "client_type",  ll_rd_client_type,      0, 0 },
 565         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 566         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 567                                ll_wr_max_readahead_mb, 0 },
 568         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 569                                      ll_wr_max_read_ahead_whole_mb, 0 },
 570         { "max_cached_mb",    ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 571         { "checksum_pages",   ll_rd_checksum, ll_wr_checksum, 0 },
 572         { "max_rw_chunk",     ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 573         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 574         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 575         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 576         { "statahead_max",    ll_rd_statahead_max, ll_wr_statahead_max, 0 },
 577         { "statahead_stats",  ll_rd_statahead_stats, 0, 0 },
 578         { "contention_seconds", ll_rd_contention_time, ll_wr_contention_time, 0},
 579         { "lockless_truncate", ll_rd_lockless_truncate,
 580                                ll_wr_lockless_truncate, 0},
 581         { 0 }
 582 };
 583
 584 #define MAX_STRING_SIZE 128
 585
 586 struct llite_file_opcode {
 587         __u32       opcode;
 588         __u32       type;
 589         const char *opname;
 590 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 591         /* file operation */
 592         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 593         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 594         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 595                                    "writeback_from_writepage" },
 596         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 597                                    "writeback_from_pressure" },
 598         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 599                                    "writeback_ok_pages" },
 600         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 601                                    "writeback_failed_pages" },
 602         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 603                                    "read_bytes" },
 604         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 605                                    "write_bytes" },
 606         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 607                                    "brw_read" },
 608         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 609                                    "brw_write" },
 610
 611         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 612         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 613         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 614         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 615         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 616         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 617         /* inode operation */
 618         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 619         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 620         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
 621         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 622         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 623         /* special inode operation */
 624         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 625         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 626         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 627         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 628         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 629         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 630         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 631         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 632                                    "direct_read" },
 633         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 634                                    "direct_write" },
 635         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 636                                    "lockless_read_bytes" },
 637         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 638                                    "lockless_write_bytes" },
 639
 640 };
 641
 642 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 643 {
 644         if (!sbi->ll_stats)
 645                 return;
 646         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 647                 lprocfs_counter_add(sbi->ll_stats, op, count);
 648         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 649                  sbi->ll_stats_track_id == current->pid)
 650                 lprocfs_counter_add(sbi->ll_stats, op, count);
 651         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 652                  sbi->ll_stats_track_id == current->p_pptr->pid)
 653                 lprocfs_counter_add(sbi->ll_stats, op, count);
 654         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 655                  sbi->ll_stats_track_id == current->gid)
 656                 lprocfs_counter_add(sbi->ll_stats, op, count);
 657 }
 658 EXPORT_SYMBOL(ll_stats_ops_tally);
 659
 660 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 661                                 struct super_block *sb, char *osc, char *mdc)
 662 {
 663         struct lprocfs_vars lvars[2];
 664         struct lustre_sb_info *lsi = s2lsi(sb);
 665         struct ll_sb_info *sbi = ll_s2sbi(sb);
 666         struct obd_device *obd;
 667         char name[MAX_STRING_SIZE + 1], *ptr;
 668         int err, id, len, rc;
 669         ENTRY;
 670
 671         memset(lvars, 0, sizeof(lvars));
 672
 673         name[MAX_STRING_SIZE] = '\0';
 674         lvars[0].name = name;
 675
 676         LASSERT(sbi != NULL);
 677         LASSERT(mdc != NULL);
 678         LASSERT(osc != NULL);
 679
 680         /* Get fsname */
 681         len = strlen(lsi->lsi_lmd->lmd_profile);
 682         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 683         if (ptr && (strcmp(ptr, "-client") == 0))
 684                 len -= 7;
 685
 686         /* Mount info */
 687         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 688                  lsi->lsi_lmd->lmd_profile, sb);
 689
 690         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 691         if (IS_ERR(sbi->ll_proc_root)) {
 692                 err = PTR_ERR(sbi->ll_proc_root);
 693                 sbi->ll_proc_root = NULL;
 694                 RETURN(err);
 695         }
 696
 697
 698         rc = lprocfs_seq_create(sbi->ll_proc_root, "dump_page_cache", 0444,
 699                                 &llite_dump_pgcache_fops, sbi);
 700         if (rc)
 701                 CWARN("Error adding the dump_page_cache file\n");
 702
 703         rc = lprocfs_seq_create(sbi->ll_proc_root, "read_ahead_stats", 0644,
 704                                 &ll_ra_stats_fops, sbi);
 705         if (rc)
 706                 CWARN("Error adding the read_ahead_stats file\n");
 707
 708         rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats", 0644,
 709                                 &ll_rw_extents_stats_fops, sbi);
 710         if (rc)
 711                 CWARN("Error adding the extent_stats file\n");
 712
 713         rc = lprocfs_seq_create(sbi->ll_proc_root, "extents_stats_per_process",
 714                                 0644, &ll_rw_extents_stats_pp_fops, sbi);
 715         if (rc)
 716                 CWARN("Error adding the extents_stats_per_process file\n");
 717
 718         rc = lprocfs_seq_create(sbi->ll_proc_root, "offset_stats", 0644,
 719                                 &ll_rw_offset_stats_fops, sbi);
 720         if (rc)
 721                 CWARN("Error adding the offset_stats file\n");
 722
 723         /* File operations stats */
 724         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 725                                             LPROCFS_STATS_FLAG_PERCPU);
 726         if (sbi->ll_stats == NULL)
 727                 GOTO(out, err = -ENOMEM);
 728         /* do counter init */
 729         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 730                 __u32 type = llite_opcode_table[id].type;
 731                 void *ptr = NULL;
 732                 if (type & LPROCFS_TYPE_REGS)
 733                         ptr = "regs";
 734                 else if (type & LPROCFS_TYPE_BYTES)
 735                         ptr = "bytes";
 736                 else if (type & LPROCFS_TYPE_PAGES)
 737                         ptr = "pages";
 738                 lprocfs_counter_init(sbi->ll_stats,
 739                                      llite_opcode_table[id].opcode,
 740                                      (type & LPROCFS_CNTR_AVGMINMAX),
 741                                      llite_opcode_table[id].opname, ptr);
 742         }
 743         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 744         if (err)
 745                 GOTO(out, err);
 746
 747         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 748         if (err)
 749                 GOTO(out, err);
 750
 751         /* MDC info */
 752         obd = class_name2obd(mdc);
 753
 754         LASSERT(obd != NULL);
 755         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 756         LASSERT(obd->obd_type->typ_name != NULL);
 757
 758         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 759                  obd->obd_type->typ_name);
 760         lvars[0].read_fptr = lprocfs_rd_name;
 761         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 762         if (err)
 763                 GOTO(out, err);
 764
 765         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 766         lvars[0].read_fptr = lprocfs_rd_uuid;
 767         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 768         if (err)
 769                 GOTO(out, err);
 770
 771         /* OSC */
 772         obd = class_name2obd(osc);
 773
 774         LASSERT(obd != NULL);
 775         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 776         LASSERT(obd->obd_type->typ_name != NULL);
 777
 778         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 779                  obd->obd_type->typ_name);
 780         lvars[0].read_fptr = lprocfs_rd_name;
 781         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 782         if (err)
 783                 GOTO(out, err);
 784
 785         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 786         lvars[0].read_fptr = lprocfs_rd_uuid;
 787         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 788 out:
 789         if (err) {
 790                 lprocfs_remove(&sbi->ll_proc_root);
 791                 lprocfs_free_stats(&sbi->ll_stats);
 792         }
 793         RETURN(err);
 794 }
 795
 796 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 797 {
 798         if (sbi->ll_proc_root) {
 799                 lprocfs_remove(&sbi->ll_proc_root);
 800                 lprocfs_free_stats(&sbi->ll_stats);
 801         }
 802 }
 803 #undef MAX_STRING_SIZE
 804
 805 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 806                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 807                         if (!has_flags)                                 \
 808                                 has_flags = 1;                          \
 809                         else                                            \
 810                                 seq_putc(seq, '|');                     \
 811                         seq_puts(seq, #flag);                           \
 812                 }                                                       \
 813         } while(0);
 814
 815 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 816 {
 817         struct ll_async_page *dummy_llap = seq->private;
 818
 819         if (dummy_llap->llap_magic == 2)
 820                 return NULL;
 821
 822         return (void *)1;
 823 }
 824
 825 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 826 {
 827         struct ll_async_page *llap, *dummy_llap = seq->private;
 828         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 829
 830         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 831          * it in our own state */
 832         if (dummy_llap->llap_magic == 0) {
 833                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
 834                                 "inode index count [ page flags ]\n");
 835                 return 0;
 836         }
 837
 838         spin_lock(&sbi->ll_lock);
 839
 840         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 841         if (llap != NULL)  {
 842                 int has_flags = 0;
 843                 struct page *page = llap->llap_page;
 844
 845                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 846                          llap->llap_origin);
 847
 848                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
 849                            "%lu %u [",
 850                            sbi->ll_pglist_gen,
 851                            llap, llap->llap_cookie,
 852                            llap_origins[llap->llap_origin],
 853                            llap->llap_write_queued ? "wq" : "- ",
 854                            llap->llap_defer_uptodate ? "du" : "- ",
 855                            PageWriteback(page) ? "wb" : "-",
 856                            page, page->mapping->host->i_ino,
 857                            page->mapping->host->i_generation,
 858                            page->mapping->host, page->index,
 859                            page_count(page));
 860                 seq_page_flag(seq, page, locked, has_flags);
 861                 seq_page_flag(seq, page, error, has_flags);
 862                 seq_page_flag(seq, page, referenced, has_flags);
 863                 seq_page_flag(seq, page, uptodate, has_flags);
 864                 seq_page_flag(seq, page, dirty, has_flags);
 865 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 866                 seq_page_flag(seq, page, highmem, has_flags);
 867 #endif
 868                 seq_page_flag(seq, page, writeback, has_flags);
 869                 if (!has_flags)
 870                         seq_puts(seq, "-]\n");
 871                 else
 872                         seq_puts(seq, "]\n");
 873         }
 874
 875         spin_unlock(&sbi->ll_lock);
 876
 877         return 0;
 878 }
 879
 880 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
 881                                          loff_t *pos)
 882 {
 883         struct ll_async_page *llap, *dummy_llap = seq->private;
 884         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 885
 886         /* bail if we just displayed the banner */
 887         if (dummy_llap->llap_magic == 0) {
 888                 dummy_llap->llap_magic = 1;
 889                 return dummy_llap;
 890         }
 891
 892         /* we've just displayed the llap that is after us in the list.
 893          * we advance to a position beyond it, returning null if there
 894          * isn't another llap in the list beyond that new position. */
 895         spin_lock(&sbi->ll_lock);
 896         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 897         list_del_init(&dummy_llap->llap_pglist_item);
 898         if (llap) {
 899                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
 900                 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
 901         }
 902         spin_unlock(&sbi->ll_lock);
 903
 904         ++*pos;
 905         if (llap == NULL) {
 906                 dummy_llap->llap_magic = 2;
 907                 return NULL;
 908         }
 909         return dummy_llap;
 910 }
 911
 912 static void null_stop(struct seq_file *seq, void *v)
 913 {
 914 }
 915
 916 struct seq_operations llite_dump_pgcache_seq_sops = {
 917         .start = llite_dump_pgcache_seq_start,
 918         .stop = null_stop,
 919         .next = llite_dump_pgcache_seq_next,
 920         .show = llite_dump_pgcache_seq_show,
 921 };
 922
 923 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
 924  * while we walk the entire list, and we don't want to have to seek into
 925  * the right position in the list as an app advances with many syscalls.  we
 926  * allocate a dummy llap and hang it off file->private.  its position in
 927  * the list records where the app is currently displaying.  this way our
 928  * seq .start and .stop don't actually do anything.  .next returns null
 929  * when the dummy hits the end of the list which eventually leads to .release
 930  * where we tear down.  this kind of displaying is super-racey, so we put
 931  * a generation counter on the list so the output shows when the list
 932  * changes between reads.
 933  */
 934 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
 935 {
 936         struct proc_dir_entry *dp = PDE(inode);
 937         struct ll_async_page *dummy_llap;
 938         struct seq_file *seq;
 939         struct ll_sb_info *sbi = dp->data;
 940         int rc = -ENOMEM;
 941
 942         LPROCFS_ENTRY_AND_CHECK(dp);
 943
 944         OBD_ALLOC_PTR_WAIT(dummy_llap);
 945         if (dummy_llap == NULL)
 946                 GOTO(out, rc);
 947         dummy_llap->llap_page = NULL;
 948         dummy_llap->llap_cookie = sbi;
 949         dummy_llap->llap_magic = 0;
 950
 951         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
 952         if (rc) {
 953                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 954                 GOTO(out, rc);
 955         }
 956         seq = file->private_data;
 957         seq->private = dummy_llap;
 958
 959         spin_lock(&sbi->ll_lock);
 960         list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
 961         spin_unlock(&sbi->ll_lock);
 962
 963 out:
 964         if (rc)
 965                 LPROCFS_EXIT();
 966         return rc;
 967 }
 968
 969 static int llite_dump_pgcache_seq_release(struct inode *inode,
 970                                           struct file *file)
 971 {
 972         struct seq_file *seq = file->private_data;
 973         struct ll_async_page *dummy_llap = seq->private;
 974         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 975
 976         spin_lock(&sbi->ll_lock);
 977         if (!list_empty(&dummy_llap->llap_pglist_item))
 978                 list_del_init(&dummy_llap->llap_pglist_item);
 979         spin_unlock(&sbi->ll_lock);
 980         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 981
 982         return lprocfs_seq_release(inode, file);
 983 }
 984
 985 struct file_operations llite_dump_pgcache_fops = {
 986         .owner   = THIS_MODULE,
 987         .open    = llite_dump_pgcache_seq_open,
 988         .read    = seq_read,
 989         .release = llite_dump_pgcache_seq_release,
 990 };
 991
 992 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
 993 {
 994         struct timeval now;
 995         struct ll_sb_info *sbi = seq->private;
 996         struct ll_ra_info *ra = &sbi->ll_ra_info;
 997         int i;
 998         static char *ra_stat_strings[] = {
 999                 [RA_STAT_HIT] = "hits",
1000                 [RA_STAT_MISS] = "misses",
1001                 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1002                 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1003                 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1004                 [RA_STAT_FAILED_MATCH] = "failed lock match",
1005                 [RA_STAT_DISCARDED] = "read but discarded",
1006                 [RA_STAT_ZERO_LEN] = "zero length file",
1007                 [RA_STAT_ZERO_WINDOW] = "zero size window",
1008                 [RA_STAT_EOF] = "read-ahead to EOF",
1009                 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1010                 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1011         };
1012
1013         do_gettimeofday(&now);
1014
1015         spin_lock(&sbi->ll_lock);
1016
1017         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1018                    now.tv_sec, now.tv_usec);
1019         seq_printf(seq, "pending issued pages:           %lu\n",
1020                    ra->ra_cur_pages);
1021
1022         for(i = 0; i < _NR_RA_STAT; i++)
1023                 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
1024                            ra->ra_stats[i]);
1025
1026         spin_unlock(&sbi->ll_lock);
1027
1028         return 0;
1029 }
1030
1031 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
1032                                        size_t len, loff_t *off)
1033 {
1034         struct seq_file *seq = file->private_data;
1035         struct ll_sb_info *sbi = seq->private;
1036         struct ll_ra_info *ra = &sbi->ll_ra_info;
1037
1038         spin_lock(&sbi->ll_lock);
1039         memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
1040         spin_unlock(&sbi->ll_lock);
1041
1042         return len;
1043 }
1044
1045 LPROC_SEQ_FOPS(ll_ra_stats);
1046
1047 #define pct(a,b) (b ? a * 100 / b : 0)
1048
1049 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1050                                    struct seq_file *seq, int which)
1051 {
1052         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1053         unsigned long start, end, r, w;
1054         char *unitp = "KMGTPEZY";
1055         int i, units = 10;
1056         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1057
1058         read_cum = 0;
1059         write_cum = 0;
1060         start = 0;
1061
1062         for(i = 0; i < LL_HIST_MAX; i++) {
1063                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1064                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1065         }
1066
1067         for(i = 0; i < LL_HIST_MAX; i++) {
1068                 r = pp_info->pp_r_hist.oh_buckets[i];
1069                 w = pp_info->pp_w_hist.oh_buckets[i];
1070                 read_cum += r;
1071                 write_cum += w;
1072                 end = 1 << (i + LL_HIST_START - units);
1073                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1074                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1075                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1076                            r, pct(r, read_tot), pct(read_cum, read_tot),
1077                            w, pct(w, write_tot), pct(write_cum, write_tot));
1078                 start = end;
1079                 if (start == 1<<10) {
1080                         start = 1;
1081                         units += 10;
1082                         unitp++;
1083                 }
1084                 if (read_cum == read_tot && write_cum == write_tot)
1085                         break;
1086         }
1087 }
1088
1089 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1090 {
1091         struct timeval now;
1092         struct ll_sb_info *sbi = seq->private;
1093         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1094         int k;
1095
1096         do_gettimeofday(&now);
1097
1098         if (!sbi->ll_rw_stats_on) {
1099                 seq_printf(seq, "Disabled\n"
1100                                 "Write anything in this file to activate\n");
1101                 return 0;
1102         }
1103         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1104                    now.tv_sec, now.tv_usec);
1105         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1106         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1107                    "extents", "calls", "%", "cum%",
1108                    "calls", "%", "cum%");
1109         spin_lock(&sbi->ll_pp_extent_lock);
1110         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1111                 if(io_extents->pp_extents[k].pid != 0) {
1112                         seq_printf(seq, "\nPID: %d\n",
1113                                    io_extents->pp_extents[k].pid);
1114                         ll_display_extents_info(io_extents, seq, k);
1115                 }
1116         }
1117         spin_unlock(&sbi->ll_pp_extent_lock);
1118         return 0;
1119 }
1120
1121 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1122                                                 const char *buf, size_t len,
1123                                                 loff_t *off)
1124 {
1125         struct seq_file *seq = file->private_data;
1126         struct ll_sb_info *sbi = seq->private;
1127         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1128         int i;
1129
1130         sbi->ll_rw_stats_on = 1;
1131         spin_lock(&sbi->ll_pp_extent_lock);
1132         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1133                 io_extents->pp_extents[i].pid = 0;
1134                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1135                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1136         }
1137         spin_unlock(&sbi->ll_pp_extent_lock);
1138         return len;
1139 }
1140
1141 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1142
1143 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1144 {
1145         struct timeval now;
1146         struct ll_sb_info *sbi = seq->private;
1147         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1148
1149         do_gettimeofday(&now);
1150
1151         if (!sbi->ll_rw_stats_on) {
1152                 seq_printf(seq, "Disabled\n"
1153                                 "Write anything in this file to activate\n");
1154                 return 0;
1155         }
1156         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1157                    now.tv_sec, now.tv_usec);
1158
1159         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1160         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1161                    "extents", "calls", "%", "cum%",
1162                    "calls", "%", "cum%");
1163         spin_lock(&sbi->ll_lock);
1164         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1165         spin_unlock(&sbi->ll_lock);
1166
1167         return 0;
1168 }
1169
1170 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1171                                         size_t len, loff_t *off)
1172 {
1173         struct seq_file *seq = file->private_data;
1174         struct ll_sb_info *sbi = seq->private;
1175         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1176         int i;
1177
1178         sbi->ll_rw_stats_on = 1;
1179         spin_lock(&sbi->ll_pp_extent_lock);
1180         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1181         {
1182                 io_extents->pp_extents[i].pid = 0;
1183                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1184                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1185         }
1186         spin_unlock(&sbi->ll_pp_extent_lock);
1187
1188         return len;
1189 }
1190
1191 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1192
1193 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1194                                *file, size_t count, int rw)
1195 {
1196         int i, cur = -1;
1197         struct ll_rw_process_info *process;
1198         struct ll_rw_process_info *offset;
1199         int *off_count = &sbi->ll_rw_offset_entry_count;
1200         int *process_count = &sbi->ll_offset_process_count;
1201         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1202
1203         if(!sbi->ll_rw_stats_on)
1204                 return;
1205         process = sbi->ll_rw_process_info;
1206         offset = sbi->ll_rw_offset_info;
1207
1208         spin_lock(&sbi->ll_pp_extent_lock);
1209         /* Extent statistics */
1210         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1211                 if(io_extents->pp_extents[i].pid == pid) {
1212                         cur = i;
1213                         break;
1214                 }
1215         }
1216
1217         if (cur == -1) {
1218                 /* new process */
1219                 sbi->ll_extent_process_count =
1220                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1221                 cur = sbi->ll_extent_process_count;
1222                 io_extents->pp_extents[cur].pid = pid;
1223                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1224                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1225         }
1226
1227         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1228              (i < (LL_HIST_MAX - 1)); i++);
1229         if (rw == 0) {
1230                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1231                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1232         } else {
1233                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1234                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1235         }
1236         spin_unlock(&sbi->ll_pp_extent_lock);
1237
1238         spin_lock(&sbi->ll_process_lock);
1239         /* Offset statistics */
1240         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1241                 if (process[i].rw_pid == pid) {
1242                         if (process[i].rw_last_file != file) {
1243                                 process[i].rw_range_start = file->f_pos;
1244                                 process[i].rw_last_file_pos =
1245                                                         file->f_pos + count;
1246                                 process[i].rw_smallest_extent = count;
1247                                 process[i].rw_largest_extent = count;
1248                                 process[i].rw_offset = 0;
1249                                 process[i].rw_last_file = file;
1250                                 spin_unlock(&sbi->ll_process_lock);
1251                                 return;
1252                         }
1253                         if (process[i].rw_last_file_pos != file->f_pos) {
1254                                 *off_count =
1255                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1256                                 offset[*off_count].rw_op = process[i].rw_op;
1257                                 offset[*off_count].rw_pid = pid;
1258                                 offset[*off_count].rw_range_start =
1259                                         process[i].rw_range_start;
1260                                 offset[*off_count].rw_range_end =
1261                                         process[i].rw_last_file_pos;
1262                                 offset[*off_count].rw_smallest_extent =
1263                                         process[i].rw_smallest_extent;
1264                                 offset[*off_count].rw_largest_extent =
1265                                         process[i].rw_largest_extent;
1266                                 offset[*off_count].rw_offset =
1267                                         process[i].rw_offset;
1268                                 process[i].rw_op = rw;
1269                                 process[i].rw_range_start = file->f_pos;
1270                                 process[i].rw_smallest_extent = count;
1271                                 process[i].rw_largest_extent = count;
1272                                 process[i].rw_offset = file->f_pos -
1273                                         process[i].rw_last_file_pos;
1274                         }
1275                         if(process[i].rw_smallest_extent > count)
1276                                 process[i].rw_smallest_extent = count;
1277                         if(process[i].rw_largest_extent < count)
1278                                 process[i].rw_largest_extent = count;
1279                         process[i].rw_last_file_pos = file->f_pos + count;
1280                         spin_unlock(&sbi->ll_process_lock);
1281                         return;
1282                 }
1283         }
1284         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1285         process[*process_count].rw_pid = pid;
1286         process[*process_count].rw_op = rw;
1287         process[*process_count].rw_range_start = file->f_pos;
1288         process[*process_count].rw_last_file_pos = file->f_pos + count;
1289         process[*process_count].rw_smallest_extent = count;
1290         process[*process_count].rw_largest_extent = count;
1291         process[*process_count].rw_offset = 0;
1292         process[*process_count].rw_last_file = file;
1293         spin_unlock(&sbi->ll_process_lock);
1294 }
1295
1296 char lpszt[] = LPSZ;
1297
1298 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1299 {
1300         struct timeval now;
1301         struct ll_sb_info *sbi = seq->private;
1302         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1303         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1304         char format[50];
1305         int i;
1306
1307         do_gettimeofday(&now);
1308
1309         if (!sbi->ll_rw_stats_on) {
1310                 seq_printf(seq, "Disabled\n"
1311                                 "Write anything in this file to activate\n");
1312                 return 0;
1313         }
1314         spin_lock(&sbi->ll_process_lock);
1315
1316         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1317                    now.tv_sec, now.tv_usec);
1318         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1319                    "R/W", "PID", "RANGE START", "RANGE END",
1320                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1321         sprintf(format, "%s%s%s%s%s\n",
1322                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1323         /* We stored the discontiguous offsets here; print them first */
1324         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1325                 if (offset[i].rw_pid != 0)
1326                         /* Is there a way to snip the '%' off of LPSZ? */
1327                         seq_printf(seq, format,
1328                                    offset[i].rw_op ? 'W' : 'R',
1329                                    offset[i].rw_pid,
1330                                    offset[i].rw_range_start,
1331                                    offset[i].rw_range_end,
1332                                    offset[i].rw_smallest_extent,
1333                                    offset[i].rw_largest_extent,
1334                                    offset[i].rw_offset);
1335         }
1336         /* Then print the current offsets for each process */
1337         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1338                 if (process[i].rw_pid != 0)
1339                         seq_printf(seq, format,
1340                                    process[i].rw_op ? 'W' : 'R',
1341                                    process[i].rw_pid,
1342                                    process[i].rw_range_start,
1343                                    process[i].rw_last_file_pos,
1344                                    process[i].rw_smallest_extent,
1345                                    process[i].rw_largest_extent,
1346                                    process[i].rw_offset);
1347         }
1348         spin_unlock(&sbi->ll_process_lock);
1349
1350         return 0;
1351 }
1352
1353 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1354                                        size_t len, loff_t *off)
1355 {
1356         struct seq_file *seq = file->private_data;
1357         struct ll_sb_info *sbi = seq->private;
1358         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1359         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1360
1361         sbi->ll_rw_stats_on = 1;
1362
1363         spin_lock(&sbi->ll_process_lock);
1364         sbi->ll_offset_process_count = 0;
1365         sbi->ll_rw_offset_entry_count = 0;
1366         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1367                LL_PROCESS_HIST_MAX);
1368         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1369                LL_OFFSET_HIST_MAX);
1370         spin_unlock(&sbi->ll_process_lock);
1371
1372         return len;
1373 }
1374
1375 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1376
1377 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1378 {
1379     lvars->module_vars  = NULL;
1380     lvars->obd_vars     = lprocfs_llite_obd_vars;
1381 }
1382 #endif /* LPROCFS */