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