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