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_fstype(char *page, char **start, off_t off, int count,
 160                         int *eof, void *data)
 161 {
 162         struct super_block *sb = (struct super_block*)data;
 163
 164         LASSERT(sb != NULL);
 165         *eof = 1;
 166         return snprintf(page, count, "%s\n", sb->s_type->name);
 167 }
 168
 169 static int ll_rd_sb_uuid(char *page, char **start, off_t off, int count,
 170                          int *eof, void *data)
 171 {
 172         struct super_block *sb = (struct super_block *)data;
 173
 174         LASSERT(sb != NULL);
 175         *eof = 1;
 176         return snprintf(page, count, "%s\n", ll_s2sbi(sb)->ll_sb_uuid.uuid);
 177 }
 178
 179 static int ll_rd_max_readahead_mb(char *page, char **start, off_t off,
 180                                    int count, int *eof, void *data)
 181 {
 182         struct super_block *sb = data;
 183         struct ll_sb_info *sbi = ll_s2sbi(sb);
 184         long pages_number;
 185         int mult;
 186
 187         spin_lock(&sbi->ll_lock);
 188         pages_number = sbi->ll_ra_info.ra_max_pages;
 189         spin_unlock(&sbi->ll_lock);
 190
 191         mult = 1 << (20 - CFS_PAGE_SHIFT);
 192         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 193 }
 194
 195 static int ll_wr_max_readahead_mb(struct file *file, const char *buffer,
 196                                    unsigned long count, void *data)
 197 {
 198         struct super_block *sb = data;
 199         struct ll_sb_info *sbi = ll_s2sbi(sb);
 200         int mult, rc, pages_number;
 201
 202         mult = 1 << (20 - CFS_PAGE_SHIFT);
 203         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 204         if (rc)
 205                 return rc;
 206
 207         if (pages_number < 0 || pages_number > num_physpages / 2) {
 208                 CERROR("can't set file readahead more than %lu MB\n",
 209                         num_physpages >> (20 - CFS_PAGE_SHIFT + 1)); /*1/2 of RAM*/
 210                 return -ERANGE;
 211         }
 212
 213         spin_lock(&sbi->ll_lock);
 214         sbi->ll_ra_info.ra_max_pages = pages_number;
 215         spin_unlock(&sbi->ll_lock);
 216
 217         return count;
 218 }
 219
 220 static int ll_rd_max_read_ahead_whole_mb(char *page, char **start, off_t off,
 221                                        int count, int *eof, void *data)
 222 {
 223         struct super_block *sb = data;
 224         struct ll_sb_info *sbi = ll_s2sbi(sb);
 225         long pages_number;
 226         int mult;
 227
 228         spin_lock(&sbi->ll_lock);
 229         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
 230         spin_unlock(&sbi->ll_lock);
 231
 232         mult = 1 << (20 - CFS_PAGE_SHIFT);
 233         return lprocfs_read_frac_helper(page, count, pages_number, mult);
 234 }
 235
 236 static int ll_wr_max_read_ahead_whole_mb(struct file *file, const char *buffer,
 237                                        unsigned long count, void *data)
 238 {
 239         struct super_block *sb = data;
 240         struct ll_sb_info *sbi = ll_s2sbi(sb);
 241         int mult, rc, pages_number;
 242
 243         mult = 1 << (20 - CFS_PAGE_SHIFT);
 244         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 245         if (rc)
 246                 return rc;
 247
 248         /* Cap this at the current max readahead window size, the readahead
 249          * algorithm does this anyway so it's pointless to set it larger. */
 250         if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
 251                 CERROR("can't set max_read_ahead_whole_mb more than "
 252                        "max_read_ahead_mb: %lu\n",
 253                        sbi->ll_ra_info.ra_max_pages >> (20 - CFS_PAGE_SHIFT));
 254                 return -ERANGE;
 255         }
 256
 257         spin_lock(&sbi->ll_lock);
 258         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 259         spin_unlock(&sbi->ll_lock);
 260
 261         return count;
 262 }
 263
 264 static int ll_rd_max_cached_mb(char *page, char **start, off_t off,
 265                                int count, int *eof, void *data)
 266 {
 267         struct super_block *sb = data;
 268         struct ll_sb_info *sbi = ll_s2sbi(sb);
 269         long pages_number;
 270         int mult;
 271
 272         spin_lock(&sbi->ll_lock);
 273         pages_number = sbi->ll_async_page_max;
 274         spin_unlock(&sbi->ll_lock);
 275
 276         mult = 1 << (20 - CFS_PAGE_SHIFT);
 277         return lprocfs_read_frac_helper(page, count, pages_number, mult);;
 278 }
 279
 280 static int ll_wr_max_cached_mb(struct file *file, const char *buffer,
 281                                   unsigned long count, void *data)
 282 {
 283         struct super_block *sb = data;
 284         struct ll_sb_info *sbi = ll_s2sbi(sb);
 285         int mult, rc, pages_number;
 286
 287         mult = 1 << (20 - CFS_PAGE_SHIFT);
 288         rc = lprocfs_write_frac_helper(buffer, count, &pages_number, mult);
 289         if (rc)
 290                 return rc;
 291
 292         if (pages_number < 0 || pages_number > num_physpages) {
 293                 CERROR("can't set max cache more than %lu MB\n",
 294                         num_physpages >> (20 - CFS_PAGE_SHIFT));
 295                 return -ERANGE;
 296         }
 297
 298         spin_lock(&sbi->ll_lock);
 299         sbi->ll_async_page_max = pages_number ;
 300         spin_unlock(&sbi->ll_lock);
 301
 302         if (!sbi->ll_osc_exp)
 303                 /* Not set up yet, don't call llap_shrink_cache */
 304                 return count;
 305
 306         if (sbi->ll_async_page_count >= sbi->ll_async_page_max)
 307                 llap_shrink_cache(sbi, 0);
 308
 309         return count;
 310 }
 311
 312 static int ll_rd_checksum(char *page, char **start, off_t off,
 313                           int count, int *eof, void *data)
 314 {
 315         struct super_block *sb = data;
 316         struct ll_sb_info *sbi = ll_s2sbi(sb);
 317
 318         return snprintf(page, count, "%u\n",
 319                         (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 320 }
 321
 322 static int ll_wr_checksum(struct file *file, const char *buffer,
 323                           unsigned long count, void *data)
 324 {
 325         struct super_block *sb = data;
 326         struct ll_sb_info *sbi = ll_s2sbi(sb);
 327         int val, rc;
 328
 329         if (!sbi->ll_osc_exp)
 330                 /* Not set up yet */
 331                 return -EAGAIN;
 332
 333         rc = lprocfs_write_helper(buffer, count, &val);
 334         if (rc)
 335                 return rc;
 336         if (val)
 337                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 338         else
 339                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 340
 341         rc = obd_set_info_async(sbi->ll_osc_exp, strlen("checksum"), "checksum",
 342                                 sizeof(val), &val, NULL);
 343         if (rc)
 344                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 345
 346         return count;
 347 }
 348
 349 static int ll_rd_max_rw_chunk(char *page, char **start, off_t off,
 350                           int count, int *eof, void *data)
 351 {
 352         struct super_block *sb = data;
 353
 354         return snprintf(page, count, "%lu\n", ll_s2sbi(sb)->ll_max_rw_chunk);
 355 }
 356
 357 static int ll_wr_max_rw_chunk(struct file *file, const char *buffer,
 358                           unsigned long count, void *data)
 359 {
 360         struct super_block *sb = data;
 361         int rc, val;
 362
 363         rc = lprocfs_write_helper(buffer, count, &val);
 364         if (rc)
 365                 return rc;
 366         ll_s2sbi(sb)->ll_max_rw_chunk = val;
 367         return count;
 368 }
 369
 370 static int ll_rd_track_id(char *page, int count, void *data,
 371                           enum vfs_track_type type)
 372 {
 373         struct super_block *sb = data;
 374
 375         if (ll_s2sbi(sb)->ll_vfs_track_type == type) {
 376                 return snprintf(page, count, "%d\n",
 377                                 ll_s2sbi(sb)->ll_vfs_track_id);
 378
 379         } else if (ll_s2sbi(sb)->ll_vfs_track_type == VFS_TRACK_ALL) {
 380                 return snprintf(page, count, "0 (all)\n");
 381         } else {
 382                 return snprintf(page, count, "untracked\n");
 383         }
 384 }
 385
 386 static int ll_wr_track_id(const char *buffer, unsigned long count, void *data,
 387                           enum vfs_track_type type)
 388 {
 389         struct super_block *sb = data;
 390         int rc, pid;
 391
 392         rc = lprocfs_write_helper(buffer, count, &pid);
 393         if (rc)
 394                 return rc;
 395         ll_s2sbi(sb)->ll_vfs_track_id = pid;
 396         if (pid == 0)
 397                 ll_s2sbi(sb)->ll_vfs_track_type = VFS_TRACK_ALL;
 398         else
 399                 ll_s2sbi(sb)->ll_vfs_track_type = type;
 400         lprocfs_clear_stats(ll_s2sbi(sb)->ll_vfs_ops_stats);
 401         return count;
 402 }
 403
 404 static int ll_rd_track_pid(char *page, char **start, off_t off,
 405                           int count, int *eof, void *data)
 406 {
 407         return (ll_rd_track_id(page, count, data, VFS_TRACK_PID));
 408 }
 409
 410 static int ll_wr_track_pid(struct file *file, const char *buffer,
 411                           unsigned long count, void *data)
 412 {
 413         return (ll_wr_track_id(buffer, count, data, VFS_TRACK_PID));
 414 }
 415
 416 static int ll_rd_track_ppid(char *page, char **start, off_t off,
 417                           int count, int *eof, void *data)
 418 {
 419         return (ll_rd_track_id(page, count, data, VFS_TRACK_PPID));
 420 }
 421
 422 static int ll_wr_track_ppid(struct file *file, const char *buffer,
 423                           unsigned long count, void *data)
 424 {
 425         return (ll_wr_track_id(buffer, count, data, VFS_TRACK_PPID));
 426 }
 427
 428 static int ll_rd_track_gid(char *page, char **start, off_t off,
 429                           int count, int *eof, void *data)
 430 {
 431         return (ll_rd_track_id(page, count, data, VFS_TRACK_GID));
 432 }
 433
 434 static int ll_wr_track_gid(struct file *file, const char *buffer,
 435                           unsigned long count, void *data)
 436 {
 437         return (ll_wr_track_id(buffer, count, data, VFS_TRACK_GID));
 438 }
 439
 440 static struct lprocfs_vars lprocfs_obd_vars[] = {
 441         { "uuid",         ll_rd_sb_uuid,          0, 0 },
 442         //{ "mntpt_path",   ll_rd_path,             0, 0 },
 443         { "fstype",       ll_rd_fstype,           0, 0 },
 444         { "blocksize",    ll_rd_blksize,          0, 0 },
 445         { "kbytestotal",  ll_rd_kbytestotal,      0, 0 },
 446         { "kbytesfree",   ll_rd_kbytesfree,       0, 0 },
 447         { "kbytesavail",  ll_rd_kbytesavail,      0, 0 },
 448         { "filestotal",   ll_rd_filestotal,       0, 0 },
 449         { "filesfree",    ll_rd_filesfree,        0, 0 },
 450         //{ "filegroups",   lprocfs_rd_filegroups,  0, 0 },
 451         { "max_read_ahead_mb", ll_rd_max_readahead_mb,
 452                                ll_wr_max_readahead_mb, 0 },
 453         { "max_read_ahead_whole_mb", ll_rd_max_read_ahead_whole_mb,
 454                                      ll_wr_max_read_ahead_whole_mb, 0 },
 455         { "max_cached_mb",  ll_rd_max_cached_mb, ll_wr_max_cached_mb, 0 },
 456         { "checksum_pages", ll_rd_checksum, ll_wr_checksum, 0 },
 457         { "max_rw_chunk",   ll_rd_max_rw_chunk, ll_wr_max_rw_chunk, 0 },
 458         { "vfs_track_pid",  ll_rd_track_pid, ll_wr_track_pid, 0 },
 459         { "vfs_track_ppid", ll_rd_track_ppid, ll_wr_track_ppid, 0 },
 460         { "vfs_track_gid",  ll_rd_track_gid, ll_wr_track_gid, 0 },
 461         { 0 }
 462 };
 463
 464 #define MAX_STRING_SIZE 128
 465
 466 struct llite_file_opcode {
 467         __u32       opcode;
 468         __u32       type;
 469         const char *opname;
 470 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
 471         /* file operation */
 472         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
 473         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
 474         { LPROC_LL_WB_WRITEPAGE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 475                                    "writeback_from_writepage" },
 476         { LPROC_LL_WB_PRESSURE,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 477                                    "writeback_from_pressure" },
 478         { LPROC_LL_WB_OK,          LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 479                                    "writeback_ok_pages" },
 480         { LPROC_LL_WB_FAIL,        LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 481                                    "writeback_failed_pages" },
 482         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 483                                    "read_bytes" },
 484         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
 485                                    "write_bytes" },
 486         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 487                                    "brw_read" },
 488         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 489                                    "brw_write" },
 490
 491         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
 492         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
 493         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
 494         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
 495         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
 496         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
 497         /* inode operation */
 498         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
 499         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "punch" },
 500 #if (LINUX_VERSION_CODE > KERNEL_VERSION(2,5,0))
 501         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
 502 #else
 503         { LPROC_LL_REVALIDATE,     LPROCFS_TYPE_REGS, "getattr" },
 504 #endif
 505         /* special inode operation */
 506         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
 507         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
 508         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
 509         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
 510         { LPROC_LL_DIRECT_READ,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 511                                    "direct_read" },
 512         { LPROC_LL_DIRECT_WRITE,   LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
 513                                    "direct_write" },
 514
 515 };
 516
 517 void ll_vfs_ops_tally(struct ll_sb_info *sbi, int op)
 518 {
 519         if (sbi->ll_vfs_ops_stats && sbi->ll_vfs_track_type == VFS_TRACK_ALL)
 520                 lprocfs_counter_incr(sbi->ll_vfs_ops_stats, op);
 521         else if (sbi->ll_vfs_track_type == VFS_TRACK_PID &&
 522                  sbi->ll_vfs_track_id == current->pid)
 523                 lprocfs_counter_incr(sbi->ll_vfs_ops_stats, op);
 524         else if (sbi->ll_vfs_track_type == VFS_TRACK_PPID &&
 525                  sbi->ll_vfs_track_id == current->p_pptr->pid)
 526                 lprocfs_counter_incr(sbi->ll_vfs_ops_stats, op);
 527         else if (sbi->ll_vfs_track_type == VFS_TRACK_GID &&
 528                  sbi->ll_vfs_track_id == current->gid)
 529                 lprocfs_counter_incr(sbi->ll_vfs_ops_stats, op);
 530 }
 531
 532 int lprocfs_register_mountpoint(struct proc_dir_entry *parent,
 533                                 struct super_block *sb, char *osc, char *mdc)
 534 {
 535         struct lprocfs_vars lvars[2];
 536         struct lustre_sb_info *lsi = s2lsi(sb);
 537         struct ll_sb_info *sbi = ll_s2sbi(sb);
 538         struct obd_device *obd;
 539         char name[MAX_STRING_SIZE + 1], *ptr;
 540         int err, id, len;
 541         struct lprocfs_stats *vfs_ops_stats = NULL;
 542         struct proc_dir_entry *entry;
 543         ENTRY;
 544
 545         memset(lvars, 0, sizeof(lvars));
 546
 547         name[MAX_STRING_SIZE] = '\0';
 548         lvars[0].name = name;
 549
 550         LASSERT(sbi != NULL);
 551         LASSERT(mdc != NULL);
 552         LASSERT(osc != NULL);
 553
 554         /* Get fsname */
 555         len = strlen(lsi->lsi_lmd->lmd_profile);
 556         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
 557         if (ptr && (strcmp(ptr, "-client") == 0))
 558                 len -= 7;
 559
 560         /* Mount info */
 561         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
 562                  lsi->lsi_lmd->lmd_profile, sb);
 563
 564         sbi->ll_proc_root = lprocfs_register(name, parent, NULL, NULL);
 565         if (IS_ERR(sbi->ll_proc_root)) {
 566                 err = PTR_ERR(sbi->ll_proc_root);
 567                 sbi->ll_proc_root = NULL;
 568                 RETURN(err);
 569         }
 570
 571         entry = create_proc_entry("dump_page_cache", 0444, sbi->ll_proc_root);
 572         if (entry == NULL)
 573                 GOTO(out, err = -ENOMEM);
 574         entry->proc_fops = &llite_dump_pgcache_fops;
 575         entry->data = sbi;
 576
 577         entry = create_proc_entry("read_ahead_stats", 0644, sbi->ll_proc_root);
 578         if (entry == NULL)
 579                 GOTO(out, err = -ENOMEM);
 580         entry->proc_fops = &ll_ra_stats_fops;
 581         entry->data = sbi;
 582
 583         entry = create_proc_entry("extents_stats", 0644, sbi->ll_proc_root);
 584         if (entry == NULL)
 585                  GOTO(out, err = -ENOMEM);
 586         entry->proc_fops = &ll_rw_extents_stats_fops;
 587         entry->data = sbi;
 588
 589         entry = create_proc_entry("extents_stats_per_process", 0644,
 590                                   sbi->ll_proc_root);
 591         if (entry == NULL)
 592                  GOTO(out, err = -ENOMEM);
 593         entry->proc_fops = &ll_rw_extents_stats_pp_fops;
 594         entry->data = sbi;
 595
 596         entry = create_proc_entry("offset_stats", 0644, sbi->ll_proc_root);
 597         if (entry == NULL)
 598                 GOTO(out, err = -ENOMEM);
 599         entry->proc_fops = &ll_rw_offset_stats_fops;
 600         entry->data = sbi;
 601
 602         /* File operations stats */
 603         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES);
 604         if (sbi->ll_stats == NULL)
 605                 GOTO(out, err = -ENOMEM);
 606         /* do counter init */
 607         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
 608                 __u32 type = llite_opcode_table[id].type;
 609                 void *ptr = NULL;
 610                 if (type & LPROCFS_TYPE_REGS)
 611                         ptr = "regs";
 612                 else {
 613                         if (type & LPROCFS_TYPE_BYTES)
 614                                 ptr = "bytes";
 615                         else {
 616                                 if (type & LPROCFS_TYPE_PAGES)
 617                                         ptr = "pages";
 618                         }
 619                 }
 620                 lprocfs_counter_init(sbi->ll_stats,
 621                                      llite_opcode_table[id].opcode,
 622                                      (type & LPROCFS_CNTR_AVGMINMAX),
 623                                      llite_opcode_table[id].opname, ptr);
 624         }
 625         err = lprocfs_register_stats(sbi->ll_proc_root, "stats", sbi->ll_stats);
 626         if (err)
 627                 GOTO(out, err);
 628
 629         /* VFS operations stats */
 630         vfs_ops_stats = sbi->ll_vfs_ops_stats =
 631                 lprocfs_alloc_stats(VFS_OPS_LAST);
 632         if (vfs_ops_stats == NULL)
 633                 GOTO(out, err = -ENOMEM);
 634         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_READ, 0, "read", "reqs");
 635         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_WRITE, 0, "write", "reqs");
 636         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_IOCTL, 0, "ioctl", "reqs");
 637         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_OPEN, 0, "open", "reqs");
 638         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_RELEASE, 0, "release",
 639                              "reqs");
 640         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_MMAP, 0, "mmap", "reqs");
 641         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_SEEK, 0, "seek", "reqs");
 642         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_FSYNC, 0, "fsync", "reqs");
 643         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_FLOCK, 0, "flock", "reqs");
 644         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_SETATTR, 0, "setattr",
 645                              "reqs");
 646         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_GETATTR, 0, "getattr",
 647                              "reqs");
 648         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_SETXATTR, 0, "setxattr",
 649                              "reqs");
 650         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_GETXATTR, 0, "getxattr",
 651                              "reqs");
 652         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_LISTXATTR, 0, "listxattr",
 653                              "reqs");
 654         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_REMOVEXATTR, 0,
 655                              "removexattr", "reqs");
 656         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_TRUNCATE, 0, "truncate",
 657                              "reqs");
 658         lprocfs_counter_init(vfs_ops_stats, VFS_OPS_INODE_PERMISSION, 0,
 659                              "inode_permission", "reqs");
 660
 661         err = lprocfs_register_stats(sbi->ll_proc_root, "vfs_ops_stats",
 662                                      vfs_ops_stats);
 663         if (err)
 664                 GOTO(out, err);
 665
 666         /* Static configuration info */
 667         err = lprocfs_add_vars(sbi->ll_proc_root, lprocfs_obd_vars, sb);
 668         if (err)
 669                 GOTO(out, err);
 670
 671         /* MDC info */
 672         obd = class_name2obd(mdc);
 673
 674         LASSERT(obd != NULL);
 675         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 676         LASSERT(obd->obd_type->typ_name != NULL);
 677
 678         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 679                  obd->obd_type->typ_name);
 680         lvars[0].read_fptr = lprocfs_rd_name;
 681         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 682         if (err)
 683                 GOTO(out, err);
 684
 685         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 686         lvars[0].read_fptr = lprocfs_rd_uuid;
 687         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 688         if (err)
 689                 GOTO(out, err);
 690
 691         /* OSC */
 692         obd = class_name2obd(osc);
 693
 694         LASSERT(obd != NULL);
 695         LASSERT(obd->obd_magic == OBD_DEVICE_MAGIC);
 696         LASSERT(obd->obd_type->typ_name != NULL);
 697
 698         snprintf(name, MAX_STRING_SIZE, "%s/common_name",
 699                  obd->obd_type->typ_name);
 700         lvars[0].read_fptr = lprocfs_rd_name;
 701         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 702         if (err)
 703                 GOTO(out, err);
 704
 705         snprintf(name, MAX_STRING_SIZE, "%s/uuid", obd->obd_type->typ_name);
 706         lvars[0].read_fptr = lprocfs_rd_uuid;
 707         err = lprocfs_add_vars(sbi->ll_proc_root, lvars, obd);
 708 out:
 709         if (err) {
 710                 lprocfs_remove(&sbi->ll_proc_root);
 711                 lprocfs_free_stats(&sbi->ll_stats);
 712                 lprocfs_free_stats(&sbi->ll_vfs_ops_stats);
 713         }
 714         RETURN(err);
 715 }
 716
 717 void lprocfs_unregister_mountpoint(struct ll_sb_info *sbi)
 718 {
 719         if (sbi->ll_proc_root) {
 720                 lprocfs_remove(&sbi->ll_proc_root);
 721                 lprocfs_free_stats(&sbi->ll_stats);
 722                 lprocfs_free_stats(&sbi->ll_vfs_ops_stats);
 723         }
 724 }
 725 #undef MAX_STRING_SIZE
 726
 727 #define seq_page_flag(seq, page, flag, has_flags) do {                  \
 728                 if (test_bit(PG_##flag, &(page)->flags)) {              \
 729                         if (!has_flags)                                 \
 730                                 has_flags = 1;                          \
 731                         else                                            \
 732                                 seq_putc(seq, '|');                     \
 733                         seq_puts(seq, #flag);                           \
 734                 }                                                       \
 735         } while(0);
 736
 737 static void *llite_dump_pgcache_seq_start(struct seq_file *seq, loff_t *pos)
 738 {
 739         struct ll_async_page *dummy_llap = seq->private;
 740
 741         if (dummy_llap->llap_magic == 2)
 742                 return NULL;
 743
 744         return (void *)1;
 745 }
 746
 747 static int llite_dump_pgcache_seq_show(struct seq_file *seq, void *v)
 748 {
 749         struct ll_async_page *llap, *dummy_llap = seq->private;
 750         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 751
 752         /* 2.4 doesn't seem to have SEQ_START_TOKEN, so we implement
 753          * it in our own state */
 754         if (dummy_llap->llap_magic == 0) {
 755                 seq_printf(seq, "gener |  llap  cookie  origin wq du | page "
 756                                 "inode index count [ page flags ]\n");
 757                 return 0;
 758         }
 759
 760         spin_lock(&sbi->ll_lock);
 761
 762         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 763         if (llap != NULL)  {
 764                 int has_flags = 0;
 765                 struct page *page = llap->llap_page;
 766
 767                 LASSERTF(llap->llap_origin < LLAP__ORIGIN_MAX, "%u\n",
 768                          llap->llap_origin);
 769
 770                 seq_printf(seq, "%5lu | %p %p %s %s %s | %p %p %lu %u [",
 771                            sbi->ll_pglist_gen,
 772                            llap, llap->llap_cookie,
 773                            llap_origins[llap->llap_origin],
 774                            llap->llap_write_queued ? "wq" : "- ",
 775                            llap->llap_defer_uptodate ? "du" : "- ",
 776                            page, page->mapping->host, page->index,
 777                            page_count(page));
 778                 seq_page_flag(seq, page, locked, has_flags);
 779                 seq_page_flag(seq, page, error, has_flags);
 780                 seq_page_flag(seq, page, referenced, has_flags);
 781                 seq_page_flag(seq, page, uptodate, has_flags);
 782                 seq_page_flag(seq, page, dirty, has_flags);
 783 #if (LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,12))
 784                 seq_page_flag(seq, page, highmem, has_flags);
 785 #endif
 786                 if (!has_flags)
 787                         seq_puts(seq, "-]\n");
 788                 else
 789                         seq_puts(seq, "]\n");
 790         }
 791
 792         spin_unlock(&sbi->ll_lock);
 793
 794         return 0;
 795 }
 796
 797 static void *llite_dump_pgcache_seq_next(struct seq_file *seq, void *v,
 798                                          loff_t *pos)
 799 {
 800         struct ll_async_page *llap, *dummy_llap = seq->private;
 801         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 802
 803         /* bail if we just displayed the banner */
 804         if (dummy_llap->llap_magic == 0) {
 805                 dummy_llap->llap_magic = 1;
 806                 return dummy_llap;
 807         }
 808
 809         /* we've just displayed the llap that is after us in the list.
 810          * we advance to a position beyond it, returning null if there
 811          * isn't another llap in the list beyond that new position. */
 812         spin_lock(&sbi->ll_lock);
 813         llap = llite_pglist_next_llap(sbi, &dummy_llap->llap_pglist_item);
 814         list_del_init(&dummy_llap->llap_pglist_item);
 815         if (llap) {
 816                 list_add(&dummy_llap->llap_pglist_item,&llap->llap_pglist_item);
 817                 llap =llite_pglist_next_llap(sbi,&dummy_llap->llap_pglist_item);
 818         }
 819         spin_unlock(&sbi->ll_lock);
 820
 821         ++*pos;
 822         if (llap == NULL) {
 823                 dummy_llap->llap_magic = 2;
 824                 return NULL;
 825         }
 826         return dummy_llap;
 827 }
 828
 829 static void null_stop(struct seq_file *seq, void *v)
 830 {
 831 }
 832
 833 struct seq_operations llite_dump_pgcache_seq_sops = {
 834         .start = llite_dump_pgcache_seq_start,
 835         .stop = null_stop,
 836         .next = llite_dump_pgcache_seq_next,
 837         .show = llite_dump_pgcache_seq_show,
 838 };
 839
 840 /* we're displaying llaps in a list_head list.  we don't want to hold a lock
 841  * while we walk the entire list, and we don't want to have to seek into
 842  * the right position in the list as an app advances with many syscalls.  we
 843  * allocate a dummy llap and hang it off file->private.  its position in
 844  * the list records where the app is currently displaying.  this way our
 845  * seq .start and .stop don't actually do anything.  .next returns null
 846  * when the dummy hits the end of the list which eventually leads to .release
 847  * where we tear down.  this kind of displaying is super-racey, so we put
 848  * a generation counter on the list so the output shows when the list
 849  * changes between reads.
 850  */
 851 static int llite_dump_pgcache_seq_open(struct inode *inode, struct file *file)
 852 {
 853         struct proc_dir_entry *dp = PDE(inode);
 854         struct ll_async_page *dummy_llap;
 855         struct seq_file *seq;
 856         struct ll_sb_info *sbi = dp->data;
 857         int rc = -ENOMEM;
 858
 859         LPROCFS_ENTRY_AND_CHECK(dp);
 860
 861         OBD_ALLOC_GFP(dummy_llap, sizeof(*dummy_llap), GFP_KERNEL);
 862         if (dummy_llap == NULL)
 863                 GOTO(out, rc);
 864
 865         dummy_llap->llap_page = NULL;
 866         dummy_llap->llap_cookie = sbi;
 867         dummy_llap->llap_magic = 0;
 868
 869         rc = seq_open(file, &llite_dump_pgcache_seq_sops);
 870         if (rc) {
 871                 OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 872                 GOTO(out, rc);
 873         }
 874         seq = file->private_data;
 875         seq->private = dummy_llap;
 876
 877         spin_lock(&sbi->ll_lock);
 878         list_add(&dummy_llap->llap_pglist_item, &sbi->ll_pglist);
 879         spin_unlock(&sbi->ll_lock);
 880
 881 out:
 882         if (rc)
 883                 LPROCFS_EXIT();
 884         return rc;
 885 }
 886
 887 static int llite_dump_pgcache_seq_release(struct inode *inode,
 888                                           struct file *file)
 889 {
 890         struct seq_file *seq = file->private_data;
 891         struct ll_async_page *dummy_llap = seq->private;
 892         struct ll_sb_info *sbi = dummy_llap->llap_cookie;
 893
 894         spin_lock(&sbi->ll_lock);
 895         if (!list_empty(&dummy_llap->llap_pglist_item))
 896                 list_del_init(&dummy_llap->llap_pglist_item);
 897         spin_unlock(&sbi->ll_lock);
 898         OBD_FREE(dummy_llap, sizeof(*dummy_llap));
 899
 900         return lprocfs_seq_release(inode, file);
 901 }
 902
 903 struct file_operations llite_dump_pgcache_fops = {
 904         .owner   = THIS_MODULE,
 905         .open    = llite_dump_pgcache_seq_open,
 906         .read    = seq_read,
 907         .release = llite_dump_pgcache_seq_release,
 908 };
 909
 910 static int ll_ra_stats_seq_show(struct seq_file *seq, void *v)
 911 {
 912         struct timeval now;
 913         struct ll_sb_info *sbi = seq->private;
 914         struct ll_ra_info *ra = &sbi->ll_ra_info;
 915         int i;
 916         static char *ra_stat_strings[] = {
 917                 [RA_STAT_HIT] = "hits",
 918                 [RA_STAT_MISS] = "misses",
 919                 [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
 920                 [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
 921                 [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
 922                 [RA_STAT_FAILED_MATCH] = "failed lock match",
 923                 [RA_STAT_DISCARDED] = "read but discarded",
 924                 [RA_STAT_ZERO_LEN] = "zero length file",
 925                 [RA_STAT_ZERO_WINDOW] = "zero size window",
 926                 [RA_STAT_EOF] = "read-ahead to EOF",
 927                 [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
 928                 [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
 929         };
 930
 931         do_gettimeofday(&now);
 932
 933         spin_lock(&sbi->ll_lock);
 934
 935         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
 936                    now.tv_sec, now.tv_usec);
 937         seq_printf(seq, "pending issued pages:           %lu\n",
 938                    ra->ra_cur_pages);
 939
 940         for(i = 0; i < _NR_RA_STAT; i++)
 941                 seq_printf(seq, "%-25s %lu\n", ra_stat_strings[i],
 942                            ra->ra_stats[i]);
 943
 944         spin_unlock(&sbi->ll_lock);
 945
 946         return 0;
 947 }
 948
 949 static ssize_t ll_ra_stats_seq_write(struct file *file, const char *buf,
 950                                        size_t len, loff_t *off)
 951 {
 952         struct seq_file *seq = file->private_data;
 953         struct ll_sb_info *sbi = seq->private;
 954         struct ll_ra_info *ra = &sbi->ll_ra_info;
 955
 956         spin_lock(&sbi->ll_lock);
 957         memset(ra->ra_stats, 0, sizeof(ra->ra_stats));
 958         spin_unlock(&sbi->ll_lock);
 959
 960         return len;
 961 }
 962
 963 LPROC_SEQ_FOPS(ll_ra_stats);
 964
 965 #define pct(a,b) (b ? a * 100 / b : 0)
 966
 967 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
 968                                    struct seq_file *seq, int which)
 969 {
 970         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
 971         unsigned long start, end, r, w;
 972         char *unitp = "KMGTPEZY";
 973         int i, units = 10;
 974         struct per_process_info *pp_info = &io_extents->pp_extents[which];
 975
 976         read_cum = 0;
 977         write_cum = 0;
 978         start = 0;
 979
 980         for(i = 0; i < LL_HIST_MAX; i++) {
 981                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
 982                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
 983         }
 984
 985         for(i = 0; i < LL_HIST_MAX; i++) {
 986                 r = pp_info->pp_r_hist.oh_buckets[i];
 987                 w = pp_info->pp_w_hist.oh_buckets[i];
 988                 read_cum += r;
 989                 write_cum += w;
 990                 end = 1 << (i + LL_HIST_START - units);
 991                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | "
 992                            "%14lu %4lu %4lu\n", start, *unitp, end, *unitp,
 993                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
 994                            r, pct(r, read_tot), pct(read_cum, read_tot),
 995                            w, pct(w, write_tot), pct(write_cum, write_tot));
 996                 start = end;
 997                 if (start == 1<<10) {
 998                         start = 1;
 999                         units += 10;
1000                         unitp++;
1001                 }
1002                 if (read_cum == read_tot && write_cum == write_tot)
1003                         break;
1004         }
1005 }
1006
1007 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1008 {
1009         struct timeval now;
1010         struct ll_sb_info *sbi = seq->private;
1011         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1012         int k;
1013
1014         do_gettimeofday(&now);
1015
1016         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1017                    now.tv_sec, now.tv_usec);
1018         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1019         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1020                    "extents", "calls", "%", "cum%",
1021                    "calls", "%", "cum%");
1022
1023         spin_lock(&sbi->ll_lock);
1024         for(k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1025                 if(io_extents->pp_extents[k].pid != 0) {
1026                         seq_printf(seq, "\nPID: %d\n",
1027                                    io_extents->pp_extents[k].pid);
1028                         ll_display_extents_info(io_extents, seq, k);
1029                 }
1030         }
1031         spin_unlock(&sbi->ll_lock);
1032
1033         return 0;
1034 }
1035
1036 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1037                                                 const char *buf, size_t len,
1038                                                 loff_t *off)
1039 {
1040         struct seq_file *seq = file->private_data;
1041         struct ll_sb_info *sbi = seq->private;
1042         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1043         int i;
1044
1045         spin_lock(&sbi->ll_lock);
1046         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1047                 io_extents->pp_extents[i].pid = 0;
1048                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1049                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1050         }
1051         spin_unlock(&sbi->ll_lock);
1052         return len;
1053 }
1054
1055 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1056
1057 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1058 {
1059         struct timeval now;
1060         struct ll_sb_info *sbi = seq->private;
1061         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1062
1063         do_gettimeofday(&now);
1064
1065         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1066                    now.tv_sec, now.tv_usec);
1067
1068         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1069         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1070                    "extents", "calls", "%", "cum%",
1071                    "calls", "%", "cum%");
1072
1073         spin_lock(&sbi->ll_lock);
1074         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1075         spin_unlock(&sbi->ll_lock);
1076
1077         return 0;
1078 }
1079
1080 static ssize_t ll_rw_extents_stats_seq_write(struct file *file, const char *buf,
1081                                         size_t len, loff_t *off)
1082 {
1083         struct seq_file *seq = file->private_data;
1084         struct ll_sb_info *sbi = seq->private;
1085         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1086
1087         lprocfs_oh_clear(&io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist);
1088         lprocfs_oh_clear(&io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist);
1089
1090         return len;
1091 }
1092
1093 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1094
1095 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid, struct file
1096                                *file, size_t count, int rw)
1097 {
1098         int i, cur = -1;
1099         struct ll_rw_process_info *process;
1100         struct ll_rw_process_info *offset;
1101         int *off_count = &sbi->ll_rw_offset_entry_count;
1102         int *process_count = &sbi->ll_offset_process_count;
1103         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1104
1105         process = sbi->ll_rw_process_info;
1106         offset = sbi->ll_rw_offset_info;
1107
1108         spin_lock(&sbi->ll_lock);
1109         /* Extent statistics */
1110         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1111                 if(io_extents->pp_extents[i].pid == pid) {
1112                         cur = i;
1113                         break;
1114                 }
1115         }
1116
1117         if (cur == -1) {
1118                 /* new process */
1119                 sbi->ll_extent_process_count =
1120                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1121                 cur = sbi->ll_extent_process_count;
1122                 io_extents->pp_extents[cur].pid = pid;
1123                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1124                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1125         }
1126
1127         for(i = 0; (count >= (1 << LL_HIST_START << i)) &&
1128              (i < (LL_HIST_MAX - 1)); i++);
1129         if (rw == 0) {
1130                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1131                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1132         } else {
1133                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1134                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1135         }
1136
1137         /* Offset statistics */
1138         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1139                 if (process[i].rw_pid == pid) {
1140                         if (process[i].rw_last_file != file) {
1141                                 process[i].rw_range_start = file->f_pos;
1142                                 process[i].rw_last_file_pos =
1143                                                         file->f_pos + count;
1144                                 process[i].rw_smallest_extent = count;
1145                                 process[i].rw_largest_extent = count;
1146                                 process[i].rw_offset = 0;
1147                                 process[i].rw_last_file = file;
1148                                 spin_unlock(&sbi->ll_lock);
1149                                 return;
1150                         }
1151                         if (process[i].rw_last_file_pos != file->f_pos) {
1152                                 *off_count =
1153                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1154                                 offset[*off_count].rw_op = process[i].rw_op;
1155                                 offset[*off_count].rw_pid = pid;
1156                                 offset[*off_count].rw_range_start =
1157                                         process[i].rw_range_start;
1158                                 offset[*off_count].rw_range_end =
1159                                         process[i].rw_last_file_pos;
1160                                 offset[*off_count].rw_smallest_extent =
1161                                         process[i].rw_smallest_extent;
1162                                 offset[*off_count].rw_largest_extent =
1163                                         process[i].rw_largest_extent;
1164                                 offset[*off_count].rw_offset =
1165                                         process[i].rw_offset;
1166                                 process[i].rw_op = rw;
1167                                 process[i].rw_range_start = file->f_pos;
1168                                 process[i].rw_smallest_extent = count;
1169                                 process[i].rw_largest_extent = count;
1170                                 process[i].rw_offset = file->f_pos -
1171                                         process[i].rw_last_file_pos;
1172                         }
1173                         if(process[i].rw_smallest_extent > count)
1174                                 process[i].rw_smallest_extent = count;
1175                         if(process[i].rw_largest_extent < count)
1176                                 process[i].rw_largest_extent = count;
1177                         process[i].rw_last_file_pos = file->f_pos + count;
1178                         spin_unlock(&sbi->ll_lock);
1179                         return;
1180                 }
1181         }
1182         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1183         process[*process_count].rw_pid = pid;
1184         process[*process_count].rw_op = rw;
1185         process[*process_count].rw_range_start = file->f_pos;
1186         process[*process_count].rw_last_file_pos = file->f_pos + count;
1187         process[*process_count].rw_smallest_extent = count;
1188         process[*process_count].rw_largest_extent = count;
1189         process[*process_count].rw_offset = 0;
1190         process[*process_count].rw_last_file = file;
1191         spin_unlock(&sbi->ll_lock);
1192 }
1193
1194 char lpszt[] = LPSZ;
1195
1196 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1197 {
1198         struct timeval now;
1199         struct ll_sb_info *sbi = seq->private;
1200         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1201         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1202         char format[50];
1203         int i;
1204
1205         do_gettimeofday(&now);
1206
1207         spin_lock(&sbi->ll_lock);
1208
1209         seq_printf(seq, "snapshot_time:         %lu.%lu (secs.usecs)\n",
1210                    now.tv_sec, now.tv_usec);
1211         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1212                    "R/W", "PID", "RANGE START", "RANGE END",
1213                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1214         sprintf(format, "%s%s%s%s%s\n",
1215                 "%3c %10d %14Lu %14Lu %17", lpszt+1, " %17", lpszt+1, " %14Ld");
1216         /* We stored the discontiguous offsets here; print them first */
1217         for(i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1218                 if (offset[i].rw_pid != 0)
1219                         /* Is there a way to snip the '%' off of LPSZ? */
1220                         seq_printf(seq, format,
1221                                    offset[i].rw_op ? 'W' : 'R',
1222                                    offset[i].rw_pid,
1223                                    offset[i].rw_range_start,
1224                                    offset[i].rw_range_end,
1225                                    offset[i].rw_smallest_extent,
1226                                    offset[i].rw_largest_extent,
1227                                    offset[i].rw_offset);
1228         }
1229         /* Then print the current offsets for each process */
1230         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1231                 if (process[i].rw_pid != 0)
1232                         seq_printf(seq, format,
1233                                    process[i].rw_op ? 'W' : 'R',
1234                                    process[i].rw_pid,
1235                                    process[i].rw_range_start,
1236                                    process[i].rw_last_file_pos,
1237                                    process[i].rw_smallest_extent,
1238                                    process[i].rw_largest_extent,
1239                                    process[i].rw_offset);
1240         }
1241         spin_unlock(&sbi->ll_lock);
1242
1243         return 0;
1244 }
1245
1246 static ssize_t ll_rw_offset_stats_seq_write(struct file *file, const char *buf,
1247                                        size_t len, loff_t *off)
1248 {
1249         struct seq_file *seq = file->private_data;
1250         struct ll_sb_info *sbi = seq->private;
1251         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1252         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1253
1254         spin_lock(&sbi->ll_lock);
1255         sbi->ll_offset_process_count = 0;
1256         sbi->ll_rw_offset_entry_count = 0;
1257         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1258                LL_PROCESS_HIST_MAX);
1259         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1260                LL_OFFSET_HIST_MAX);
1261         spin_unlock(&sbi->ll_lock);
1262
1263         return len;
1264 }
1265
1266 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1267
1268 LPROCFS_INIT_VARS(llite, NULL, lprocfs_obd_vars)
1269 #endif /* LPROCFS */