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_contention_time(char *page, char **start, off_t off,
 464                                  int count, int *eof, void *data)
 465 {
 466         struct super_block *sb = data;
 467
 468         *eof = 1;
 469         return snprintf(page, count, "%u\n", ll_s2sbi(sb)->ll_contention_time);
 470
 471 }
 472
 473 static int ll_wr_contention_time(struct file *file, const char *buffer,
 474                                  unsigned long count, void *data)
 475 {
 476         struct super_block *sb = data;
 477         struct ll_sb_info *sbi = ll_s2sbi(sb);
 478
 479         return lprocfs_write_helper(buffer, count,&sbi->ll_contention_time) ?:
 480                 count;
 481 }
 482
 483 static int ll_rd_lockless_truncate(char *page, char **start, off_t off,
 484                                    int count, int *eof, void *data)
 485 {
 486         struct super_block *sb = data;
 487
 488         *eof = 1;
 489         return snprintf(page, count, "%u\n",
 490                         ll_s2sbi(sb)->ll_lockless_truncate_enable);
 491 }
 492
 493 static int ll_wr_lockless_truncate(struct file *file, const char *buffer,
 494                                    unsigned long count, void *data)
 495 {
 496         struct super_block *sb = data;
 497         struct ll_sb_info *sbi = ll_s2sbi(sb);
 498
 499         return lprocfs_write_helper(buffer, count,
 500                                     &sbi->ll_lockless_truncate_enable)
 501                                     ?: count;
 502 }
 503
 504 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 505         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 506         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 507         { "fstype",       ll_rd_fstype,           0, 0 },
 508         { "blocksize",    ll_rd_blksize,          0, 0 },
 509         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 510         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 511         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 512         { "filestotal",   ll_rd_filestotal,       0, 0 },
 513         { "filesfree",    ll_rd_filesfree,        0, 0 },
 514         { "client_type",  ll_rd_client_type,      0, 0 },
 515         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 516         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 517                                ll_wr_max_readahead_mb, 0 },
 518         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 519                                      ll_wr_max_read_ahead_whole_mb, 0 },
 520         { "max_cached_mb",    ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 521         { "checksum_pages",   ll_rd_checksum, ll_wr_checksum, 0 },
 522         { "max_rw_chunk",     ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 523         { "stats_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 524         { "stats_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 525         { "stats_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 526         { "contention_seconds", ll_rd_contention_time, ll_wr_contention_time, 0},
 527         { "lockless_truncate", ll_rd_lockless_truncate,
 528                                ll_wr_lockless_truncate, 0},
 529         { 0 }
 530 };
 531
 532 #define MAX_STRING_SIZE 128
 533
 534 struct llite_file_opcode {
 535         __u32       opcode;
 536         __u32       type;
 537         const char *opname;
 538 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 539         /* file operation */
 540         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 541         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 542         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 543                                    "writeback_from_writepage" },
 544         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 545                                    "writeback_from_pressure" },
 546         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 547                                    "writeback_ok_pages" },
 548         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 549                                    "writeback_failed_pages" },
 550         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 551                                    "read_bytes" },
 552         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 553                                    "write_bytes" },
 554         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 555                                    "brw_read" },
 556         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 557                                    "brw_write" },
 558
 559         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 560         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 561         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 562         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 563         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 564         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 565         /* inode operation */
 566         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 567         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
 568         { LPROC_LL_LOCKLESS_TRUNC, LPROCFS_TYPE_REGS, "lockless_truncate"},
 569         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
 570         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 571         /* special inode operation */
 572         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 573         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 574         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 575         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 576         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
 577         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
 578         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
 579         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 580                                    "direct_read" },
 581         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 582                                    "direct_write" },
 583         { LPROC_LL_LOCKLESS_READ,  LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 584                                    "lockless_read_bytes" },
 585         { LPROC_LL_LOCKLESS_WRITE, LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 586                                    "lockless_write_bytes" },
 587
 588 };
 589
 590 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
 591 {
 592         if (!sbi->ll_stats)
 593                 return;
 594         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 595                 lprocfs_counter_add(sbi->ll_stats, op, count);
 596         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
 597                  sbi->ll_stats_track_id == current->pid)
 598                 lprocfs_counter_add(sbi->ll_stats, op, count);
 599         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
 600                  sbi->ll_stats_track_id == current->p_pptr->pid)
 601                 lprocfs_counter_add(sbi->ll_stats, op, count);
 602         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
 603                  sbi->ll_stats_track_id == current->gid)
 604                 lprocfs_counter_add(sbi->ll_stats, op, count);
 605 }
 606 EXPORT_SYMBOL(ll_stats_ops_tally);
 607
 608 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 609                                 struct super_block *sb, char *osc, char *mdc)
 610 {
 611         struct lprocfs_vars lvars[2];
 612         struct lustre_sb_info *lsi = s2lsi(sb);
 613         struct ll_sb_info *sbi = ll_s2sbi(sb);
 614         struct obd_device *obd;
 615         char name[MAX_STRING_SIZE + 1], *ptr;
 616         int err, id, len;
 617         struct proc_dir_entry *entry;
 618         ENTRY;
 619
 620         memset(lvars, 0, sizeof(lvars));
 621
 622         name[MAX_STRING_SIZE] = '\0';
 623         lvars[0].name = name;
 624
 625         LASSERT(sbi != NULL);
 626         LASSERT(mdc != NULL);
 627         LASSERT(osc != NULL);
 628
 629         /* Get fsname */
 630         len = strlen(lsi->lsi_lmd->lmd_profile);
 631         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 632         if (ptr && (strcmp(ptr, "-client") == 0))
 633                 len -= 7;
 634
 635         /* Mount info */
 636         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 637                  lsi->lsi_lmd->lmd_profile, sb);
 638
 639         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 640         if (IS_ERR(sbi->ll_proc_root)) {
 641                 err = PTR_ERR(sbi->ll_proc_root);
 642                 sbi->ll_proc_root = NULL;
 643                 RETURN(err);
 644         }
 645
 646         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 647         if (entry == NULL)
 648                 GOTO(out, err = -ENOMEM);
 649         entry->proc_fops = &llite_dump_pgcache_fops;
 650         entry->data = sbi;
 651
 652         entry = create_proc_entry("read_ahead_stats", 0644, sbi->ll_proc_root);
 653         if (entry == NULL)
 654                 GOTO(out, err = -ENOMEM);
 655         entry->proc_fops = &ll_ra_stats_fops;
 656         entry->data = sbi;
 657
 658         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 659         if (entry == NULL)
 660                  GOTO(out, err = -ENOMEM);
 661         entry->proc_fops = &ll_rw_extents_stats_fops;
 662         entry->data = sbi;
 663
 664         entry = create_proc_entry("extents_stats_per_process", 0644,
 665                                   sbi->ll_proc_root);
 666         if (entry == NULL)
 667                  GOTO(out, err = -ENOMEM);
 668         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 669         entry->data = sbi;
 670
 671         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 672         if (entry == NULL)
 673                 GOTO(out, err = -ENOMEM);
 674         entry->proc_fops = &ll_rw_offset_stats_fops;
 675         entry->data = sbi;
 676
 677         /* File operations stats */
 678         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
 679                                             LPROCFS_STATS_FLAG_PERCPU);
 680         if (sbi->ll_stats == NULL)
 681                 GOTO(out, err = -ENOMEM);
 682         /* do counter init */
 683         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 684                 __u32 type = llite_opcode_table[id].type;
 685                 void *ptr = NULL;
 686                 if (type & LPROCFS_TYPE_REGS)
 687                         ptr = "regs";
 688                 else if (type & LPROCFS_TYPE_BYTES)
 689                         ptr = "bytes";
 690                 else if (type & LPROCFS_TYPE_PAGES)
 691                         ptr = "pages";
 692                 lprocfs_counter_init(sbi->ll_stats,
 693                                      llite_opcode_table[id].opcode,
 694                                      (type & LPROCFS_CNTR_AVGMINMAX),
 695                                      llite_opcode_table[id].opname, ptr);
 696         }
 697         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 698         if (err)
 699                 GOTO(out, err);
 700
 701         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_llite_obd_vars, sb);
 702         if (err)
 703                 GOTO(out, err);
 704
 705         /* MDC info */
 706         obd = class_name2obd(mdc);
 707
 708         LASSERT(obd != NULL);
 709         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 710         LASSERT(obd->obd_type->typ_name != NULL);
 711
 712         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 713                  obd->obd_type->typ_name);
 714         lvars[0].read_fptr = lprocfs_rd_name;
 715         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 716         if (err)
 717                 GOTO(out, err);
 718
 719         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 720         lvars[0].read_fptr = lprocfs_rd_uuid;
 721         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 722         if (err)
 723                 GOTO(out, err);
 724
 725         /* OSC */
 726         obd = class_name2obd(osc);
 727
 728         LASSERT(obd != NULL);
 729         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 730         LASSERT(obd->obd_type->typ_name != NULL);
 731
 732         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 733                  obd->obd_type->typ_name);
 734         lvars[0].read_fptr = lprocfs_rd_name;
 735         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 736         if (err)
 737                 GOTO(out, err);
 738
 739         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 740         lvars[0].read_fptr = lprocfs_rd_uuid;
 741         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 742 out:
 743         if (err) {
 744                 lprocfs_remove(&sbi->ll_proc_root);
 745                 lprocfs_free_stats(&sbi->ll_stats);
 746         }
 747         RETURN(err);
 748 }
 749
 750 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 751 {
 752         if (sbi->ll_proc_root) {
 753                 lprocfs_remove(&sbi->ll_proc_root);
 754                 lprocfs_free_stats(&sbi->ll_stats);
 755         }
 756 }
 757 #undef MAX_STRING_SIZE
 758
 759 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 760                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 761                         if (!has_flags)                                 \
 762                                 has_flags = 1;                          \
 763                         else                                            \
 764                                 seq_putc(seq, '|');                     \
 765                         seq_puts(seq, #flag);                           \
 766                 }                                                       \
 767         } while(0);
 768
 769 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 770 {
 771         struct ll_async_page *dummy_llap = seq->private;
 772
 773         if (dummy_llap->llap_magic == 2)
 774                 return NULL;
 775
 776         return (void *)1;
 777 }
 778
 779 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 780 {
 781         struct ll_async_page *llap, *dummy_llap = seq->private;
 782         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 783
 784         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 785          * it in our own state */
 786         if (dummy_llap->llap_magic == 0) {
 787                 seq_printf(seq, "gener |  llap  cookie  origin wq du wb | page "
 788                                 "inode index count [ page flags ]\n");
 789                 return 0;
 790         }
 791
 792         spin_lock(&sbi->ll_lock);
 793
 794         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 795         if (llap != NULL)  {
 796                 int has_flags = 0;
 797                 struct page *page = llap->llap_page;
 798
 799                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 800                          llap->llap_origin);
 801
 802                 seq_printf(seq," %5lu | %p %p %s %s %s %s | %p %lu/%u(%p) "
 803                            "%lu %u [",
 804                            sbi->ll_pglist_gen,
 805                            llap, llap->llap_cookie,
 806                            llap_origins[llap->llap_origin],
 807                            llap->llap_write_queued ? "wq" : "- ",
 808                            llap->llap_defer_uptodate ? "du" : "- ",
 809                            PageWriteback(page) ? "wb" : "-",
 810                            page, page->mapping->host->i_ino,
 811                            page->mapping->host->i_generation,
 812                            page->mapping->host, page->index,
 813                            page_count(page));
 814                 seq_page_flag(seq, page, locked, has_flags);
 815                 seq_page_flag(seq, page, error, has_flags);
 816                 seq_page_flag(seq, page, referenced, has_flags);
 817                 seq_page_flag(seq, page, uptodate, has_flags);
 818                 seq_page_flag(seq, page, dirty, has_flags);
 819 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 820                 seq_page_flag(seq, page, highmem, has_flags);
 821 #endif
 822                 seq_page_flag(seq, page, writeback, has_flags);
 823                 if (!has_flags)
 824                         seq_puts(seq, "-]\n");
 825                 else
 826                         seq_puts(seq, "]\n");
 827         }
 828
 829         spin_unlock(&sbi->ll_lock);
 830
 831         return 0;
 832 }
 833
 834 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
 835                                          loff_t *pos)
 836 {
 837         struct ll_async_page *llap, *dummy_llap = seq->private;
 838         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 839
 840         /* bail if we just displayed the banner */
 841         if (dummy_llap->llap_magic == 0) {
 842                 dummy_llap->llap_magic = 1;
 843                 return dummy_llap;
 844         }
 845
 846         /* we've just displayed the llap that is after us in the list.
 847          * we advance to a position beyond it, returning null if there
 848          * isn't another llap in the list beyond that new position. */
 849         spin_lock(&sbi->ll_lock);
 850         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 851         list_del_init(&dummy_llap->llap_pglist_item);
 852         if (llap) {
 853                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
 854                 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
 855         }
 856         spin_unlock(&sbi->ll_lock);
 857
 858         ++*pos;
 859         if (llap == NULL) {
 860                 dummy_llap->llap_magic = 2;
 861                 return NULL;
 862         }
 863         return dummy_llap;
 864 }
 865
 866 static void null_stop(struct seq_file *seq, void *v)
 867 {
 868 }
 869
 870 struct seq_operations llite_dump_pgcache_seq_sops = {
 871         .start = llite_dump_pgcache_seq_start,
 872         .stop = null_stop,
 873         .next = llite_dump_pgcache_seq_next,
 874         .show = llite_dump_pgcache_seq_show,
 875 };
 876
 877 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
 878  * while we walk the entire list, and we don't want to have to seek into
 879  * the right position in the list as an app advances with many syscalls.  we
 880  * allocate a dummy llap and hang it off file->private.  its position in
 881  * the list records where the app is currently displaying.  this way our
 882  * seq .start and .stop don't actually do anything.  .next returns null
 883  * when the dummy hits the end of the list which eventually leads to .release
 884  * where we tear down.  this kind of displaying is super-racey, so we put
 885  * a generation counter on the list so the output shows when the list
 886  * changes between reads.
 887  */
 888 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
 889 {
 890         struct proc_dir_entry *dp = PDE(inode);
 891         struct ll_async_page *dummy_llap;
 892         struct seq_file *seq;
 893         struct ll_sb_info *sbi = dp->data;
 894         int rc = -ENOMEM;
 895
 896         LPROCFS_ENTRY_AND_CHECK(dp);
 897
 898         OBD_ALLOC_PTR_WAIT(dummy_llap);
 899         if (dummy_llap == NULL)
 900                 GOTO(out, rc);
 901         dummy_llap->llap_page = NULL;
 902         dummy_llap->llap_cookie = sbi;
 903         dummy_llap->llap_magic = 0;
 904
 905         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
 906         if (rc) {
 907                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 908                 GOTO(out, rc);
 909         }
 910         seq = file->private_data;
 911         seq->private = dummy_llap;
 912
 913         spin_lock(&sbi->ll_lock);
 914         list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
 915         spin_unlock(&sbi->ll_lock);
 916
 917 out:
 918         if (rc)
 919                 LPROCFS_EXIT();
 920         return rc;
 921 }
 922
 923 static int llite_dump_pgcache_seq_release(struct inode *inode,
 924                                           struct file *file)
 925 {
 926         struct seq_file *seq = file->private_data;
 927         struct ll_async_page *dummy_llap = seq->private;
 928         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 929
 930         spin_lock(&sbi->ll_lock);
 931         if (!list_empty(&dummy_llap->llap_pglist_item))
 932                 list_del_init(&dummy_llap->llap_pglist_item);
 933         spin_unlock(&sbi->ll_lock);
 934         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 935
 936         return lprocfs_seq_release(inode, file);
 937 }
 938
 939 struct file_operations llite_dump_pgcache_fops = {
 940         .owner   = THIS_MODULE,
 941         .open    = llite_dump_pgcache_seq_open,
 942         .read    = seq_read,
 943         .release = llite_dump_pgcache_seq_release,
 944 };
 945
 946 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
 947 {
 948         struct timeval now;
 949         struct ll_sb_info *sbi = seq->private;
 950         struct ll_ra_info *ra = &sbi->ll_ra_info;
 951         int i;
 952         static char *ra_stat_strings[] = {
 953                 [RA_STAT_HIT] = "hits",
 954                 [RA_STAT_MISS] = "misses",
 955                 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
 956                 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
 957                 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
 958                 [RA_STAT_FAILED_MATCH] = "failed lock match",
 959                 [RA_STAT_DISCARDED] = "read but discarded",
 960                 [RA_STAT_ZERO_LEN] = "zero length file",
 961                 [RA_STAT_ZERO_WINDOW] = "zero size window",
 962                 [RA_STAT_EOF] = "read-ahead to EOF",
 963                 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
 964                 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
 965         };
 966
 967         do_gettimeofday(&now);
 968
 969         spin_lock(&sbi->ll_lock);
 970
 971         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
 972                    now.tv_sec, now.tv_usec);
 973         seq_printf(seq, "pending issued pages:           %lu\n",
 974                    ra->ra_cur_pages);
 975
 976         for(i = 0; i < _NR_RA_STAT; i++)
 977                 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
 978                            ra->ra_stats[i]);
 979
 980         spin_unlock(&sbi->ll_lock);
 981
 982         return 0;
 983 }
 984
 985 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
 986                                        size_t len, loff_t *off)
 987 {
 988         struct seq_file *seq = file->private_data;
 989         struct ll_sb_info *sbi = seq->private;
 990         struct ll_ra_info *ra = &sbi->ll_ra_info;
 991
 992         spin_lock(&sbi->ll_lock);
 993         memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
 994         spin_unlock(&sbi->ll_lock);
 995
 996         return len;
 997 }
 998
 999 LPROC_SEQ_FOPS(ll_ra_stats);
1000
1001 #define pct(a,b) (b ? a * 100 / b : 0)
1002
1003 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1004                                    struct seq_file *seq, int which)
1005 {
1006         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1007         unsigned long start, end, r, w;
1008         char *unitp = "KMGTPEZY";
1009         int i, units = 10;
1010         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1011
1012         read_cum = 0;
1013         write_cum = 0;
1014         start = 0;
1015
1016         for(i = 0; i < LL_HIST_MAX; i++) {
1017                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1018                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1019         }
1020
1021         for(i = 0; i < LL_HIST_MAX; i++) {
1022                 r = pp_info->pp_r_hist.oh_buckets[i];
1023                 w = pp_info->pp_w_hist.oh_buckets[i];
1024                 read_cum += r;
1025                 write_cum += w;
1026                 end = 1 << (i + LL_HIST_START - units);
1027                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
1028                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
1029                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1030                            r, pct(r, read_tot), pct(read_cum, read_tot),
1031                            w, pct(w, write_tot), pct(write_cum, write_tot));
1032                 start = end;
1033                 if (start == 1<<10) {
1034                         start = 1;
1035                         units += 10;
1036                         unitp++;
1037                 }
1038                 if (read_cum == read_tot && write_cum == write_tot)
1039                         break;
1040         }
1041 }
1042
1043 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1044 {
1045         struct timeval now;
1046         struct ll_sb_info *sbi = seq->private;
1047         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1048         int k;
1049
1050         do_gettimeofday(&now);
1051
1052         if (!sbi->ll_rw_stats_on) {
1053                 seq_printf(seq, "Disabled\n"
1054                                 "Write anything in this file to activate\n");
1055                 return 0;
1056         }
1057         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1058                    now.tv_sec, now.tv_usec);
1059         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1060         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1061                    "extents", "calls", "%", "cum%",
1062                    "calls", "%", "cum%");
1063         spin_lock(&sbi->ll_pp_extent_lock);
1064         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1065                 if(io_extents->pp_extents[k].pid != 0) {
1066                         seq_printf(seq, "\nPID: %d\n",
1067                                    io_extents->pp_extents[k].pid);
1068                         ll_display_extents_info(io_extents, seq, k);
1069                 }
1070         }
1071         spin_unlock(&sbi->ll_pp_extent_lock);
1072         return 0;
1073 }
1074
1075 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1076                                                 const char *buf, size_t len,
1077                                                 loff_t *off)
1078 {
1079         struct seq_file *seq = file->private_data;
1080         struct ll_sb_info *sbi = seq->private;
1081         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1082         int i;
1083
1084         sbi->ll_rw_stats_on = 1;
1085         spin_lock(&sbi->ll_pp_extent_lock);
1086         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1087                 io_extents->pp_extents[i].pid = 0;
1088                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1089                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1090         }
1091         spin_unlock(&sbi->ll_pp_extent_lock);
1092         return len;
1093 }
1094
1095 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1096
1097 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1098 {
1099         struct timeval now;
1100         struct ll_sb_info *sbi = seq->private;
1101         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1102
1103         do_gettimeofday(&now);
1104
1105         if (!sbi->ll_rw_stats_on) {
1106                 seq_printf(seq, "Disabled\n"
1107                                 "Write anything in this file to activate\n");
1108                 return 0;
1109         }
1110         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1111                    now.tv_sec, now.tv_usec);
1112
1113         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1114         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1115                    "extents", "calls", "%", "cum%",
1116                    "calls", "%", "cum%");
1117         spin_lock(&sbi->ll_lock);
1118         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1119         spin_unlock(&sbi->ll_lock);
1120
1121         return 0;
1122 }
1123
1124 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1125                                         size_t len, loff_t *off)
1126 {
1127         struct seq_file *seq = file->private_data;
1128         struct ll_sb_info *sbi = seq->private;
1129         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1130         int i;
1131
1132         sbi->ll_rw_stats_on = 1;
1133         spin_lock(&sbi->ll_pp_extent_lock);
1134         for(i = 0; i <= LL_PROCESS_HIST_MAX; i++)
1135         {
1136                 io_extents->pp_extents[i].pid = 0;
1137                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1138                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1139         }
1140         spin_unlock(&sbi->ll_pp_extent_lock);
1141
1142         return len;
1143 }
1144
1145 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1146
1147 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1148                                *file, size_t count, int rw)
1149 {
1150         int i, cur = -1;
1151         struct ll_rw_process_info *process;
1152         struct ll_rw_process_info *offset;
1153         int *off_count = &sbi->ll_rw_offset_entry_count;
1154         int *process_count = &sbi->ll_offset_process_count;
1155         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1156
1157         if(!sbi->ll_rw_stats_on)
1158                 return;
1159         process = sbi->ll_rw_process_info;
1160         offset = sbi->ll_rw_offset_info;
1161
1162         spin_lock(&sbi->ll_pp_extent_lock);
1163         /* Extent statistics */
1164         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1165                 if(io_extents->pp_extents[i].pid == pid) {
1166                         cur = i;
1167                         break;
1168                 }
1169         }
1170
1171         if (cur == -1) {
1172                 /* new process */
1173                 sbi->ll_extent_process_count =
1174                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1175                 cur = sbi->ll_extent_process_count;
1176                 io_extents->pp_extents[cur].pid = pid;
1177                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1178                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1179         }
1180
1181         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1182              (i < (LL_HIST_MAX - 1)); i++);
1183         if (rw == 0) {
1184                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1185                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1186         } else {
1187                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1188                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1189         }
1190         spin_unlock(&sbi->ll_pp_extent_lock);
1191
1192         spin_lock(&sbi->ll_process_lock);
1193         /* Offset statistics */
1194         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1195                 if (process[i].rw_pid == pid) {
1196                         if (process[i].rw_last_file != file) {
1197                                 process[i].rw_range_start = file->f_pos;
1198                                 process[i].rw_last_file_pos =
1199                                                         file->f_pos + count;
1200                                 process[i].rw_smallest_extent = count;
1201                                 process[i].rw_largest_extent = count;
1202                                 process[i].rw_offset = 0;
1203                                 process[i].rw_last_file = file;
1204                                 spin_unlock(&sbi->ll_process_lock);
1205                                 return;
1206                         }
1207                         if (process[i].rw_last_file_pos != file->f_pos) {
1208                                 *off_count =
1209                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1210                                 offset[*off_count].rw_op = process[i].rw_op;
1211                                 offset[*off_count].rw_pid = pid;
1212                                 offset[*off_count].rw_range_start =
1213                                         process[i].rw_range_start;
1214                                 offset[*off_count].rw_range_end =
1215                                         process[i].rw_last_file_pos;
1216                                 offset[*off_count].rw_smallest_extent =
1217                                         process[i].rw_smallest_extent;
1218                                 offset[*off_count].rw_largest_extent =
1219                                         process[i].rw_largest_extent;
1220                                 offset[*off_count].rw_offset =
1221                                         process[i].rw_offset;
1222                                 process[i].rw_op = rw;
1223                                 process[i].rw_range_start = file->f_pos;
1224                                 process[i].rw_smallest_extent = count;
1225                                 process[i].rw_largest_extent = count;
1226                                 process[i].rw_offset = file->f_pos -
1227                                         process[i].rw_last_file_pos;
1228                         }
1229                         if(process[i].rw_smallest_extent > count)
1230                                 process[i].rw_smallest_extent = count;
1231                         if(process[i].rw_largest_extent < count)
1232                                 process[i].rw_largest_extent = count;
1233                         process[i].rw_last_file_pos = file->f_pos + count;
1234                         spin_unlock(&sbi->ll_process_lock);
1235                         return;
1236                 }
1237         }
1238         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1239         process[*process_count].rw_pid = pid;
1240         process[*process_count].rw_op = rw;
1241         process[*process_count].rw_range_start = file->f_pos;
1242         process[*process_count].rw_last_file_pos = file->f_pos + count;
1243         process[*process_count].rw_smallest_extent = count;
1244         process[*process_count].rw_largest_extent = count;
1245         process[*process_count].rw_offset = 0;
1246         process[*process_count].rw_last_file = file;
1247         spin_unlock(&sbi->ll_process_lock);
1248 }
1249
1250 char lpszt[] = LPSZ;
1251
1252 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1253 {
1254         struct timeval now;
1255         struct ll_sb_info *sbi = seq->private;
1256         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1257         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1258         char format[50];
1259         int i;
1260
1261         do_gettimeofday(&now);
1262
1263         if (!sbi->ll_rw_stats_on) {
1264                 seq_printf(seq, "Disabled\n"
1265                                 "Write anything in this file to activate\n");
1266                 return 0;
1267         }
1268         spin_lock(&sbi->ll_process_lock);
1269
1270         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1271                    now.tv_sec, now.tv_usec);
1272         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1273                    "R/W", "PID", "RANGE START", "RANGE END",
1274                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1275         sprintf(format, "%s%s%s%s%s\n",
1276                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1277         /* We stored the discontiguous offsets here; print them first */
1278         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1279                 if (offset[i].rw_pid != 0)
1280                         /* Is there a way to snip the '%' off of LPSZ? */
1281                         seq_printf(seq, format,
1282                                    offset[i].rw_op ? 'W' : 'R',
1283                                    offset[i].rw_pid,
1284                                    offset[i].rw_range_start,
1285                                    offset[i].rw_range_end,
1286                                    offset[i].rw_smallest_extent,
1287                                    offset[i].rw_largest_extent,
1288                                    offset[i].rw_offset);
1289         }
1290         /* Then print the current offsets for each process */
1291         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1292                 if (process[i].rw_pid != 0)
1293                         seq_printf(seq, format,
1294                                    process[i].rw_op ? 'W' : 'R',
1295                                    process[i].rw_pid,
1296                                    process[i].rw_range_start,
1297                                    process[i].rw_last_file_pos,
1298                                    process[i].rw_smallest_extent,
1299                                    process[i].rw_largest_extent,
1300                                    process[i].rw_offset);
1301         }
1302         spin_unlock(&sbi->ll_process_lock);
1303
1304         return 0;
1305 }
1306
1307 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1308                                        size_t len, loff_t *off)
1309 {
1310         struct seq_file *seq = file->private_data;
1311         struct ll_sb_info *sbi = seq->private;
1312         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1313         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1314
1315         sbi->ll_rw_stats_on = 1;
1316
1317         spin_lock(&sbi->ll_process_lock);
1318         sbi->ll_offset_process_count = 0;
1319         sbi->ll_rw_offset_entry_count = 0;
1320         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1321                LL_PROCESS_HIST_MAX);
1322         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1323                LL_OFFSET_HIST_MAX);
1324         spin_unlock(&sbi->ll_process_lock);
1325
1326         return len;
1327 }
1328
1329 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1330
1331 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1332 {
1333     lvars->module_vars  = NULL;
1334     lvars->obd_vars     = lprocfs_llite_obd_vars;
1335 }
1336 #endif /* LPROCFS */