drivers/staging/lustre/lustre/llite/lproc_llite.c

   1 /*
   2  * GPL HEADER START
   3  *
   4  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  13  * General Public License version 2 for more details (a copy is included
  14  * in the LICENSE file that accompanied this code).
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * version 2 along with this program; If not, see
  18  * http://www.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2015, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_LLITE
  33
  34 #include <lprocfs_status.h>
  35 #include <linux/seq_file.h>
  36 #include <obd_support.h>
  37
  38 #include "llite_internal.h"
  39 #include "vvp_internal.h"
  40
  41 /* debugfs llite mount point registration */
  42 static const struct file_operations ll_rw_extents_stats_fops;
  43 static const struct file_operations ll_rw_extents_stats_pp_fops;
  44 static const struct file_operations ll_rw_offset_stats_fops;
  45
  46 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
  47                               char *buf)
  48 {
  49         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
  50                                               ll_kobj);
  51         struct obd_statfs osfs;
  52         int rc;
  53
  54         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
  55                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
  56                                 OBD_STATFS_NODELAY);
  57         if (!rc)
  58                 return sprintf(buf, "%u\n", osfs.os_bsize);
  59
  60         return rc;
  61 }
  62 LUSTRE_RO_ATTR(blocksize);
  63
  64 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
  65                                 char *buf)
  66 {
  67         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
  68                                               ll_kobj);
  69         struct obd_statfs osfs;
  70         int rc;
  71
  72         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
  73                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
  74                                 OBD_STATFS_NODELAY);
  75         if (!rc) {
  76                 __u32 blk_size = osfs.os_bsize >> 10;
  77                 __u64 result = osfs.os_blocks;
  78
  79                 while (blk_size >>= 1)
  80                         result <<= 1;
  81
  82                 rc = sprintf(buf, "%llu\n", result);
  83         }
  84
  85         return rc;
  86 }
  87 LUSTRE_RO_ATTR(kbytestotal);
  88
  89 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
  90                                char *buf)
  91 {
  92         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
  93                                               ll_kobj);
  94         struct obd_statfs osfs;
  95         int rc;
  96
  97         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
  98                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
  99                                 OBD_STATFS_NODELAY);
 100         if (!rc) {
 101                 __u32 blk_size = osfs.os_bsize >> 10;
 102                 __u64 result = osfs.os_bfree;
 103
 104                 while (blk_size >>= 1)
 105                         result <<= 1;
 106
 107                 rc = sprintf(buf, "%llu\n", result);
 108         }
 109
 110         return rc;
 111 }
 112 LUSTRE_RO_ATTR(kbytesfree);
 113
 114 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 115                                 char *buf)
 116 {
 117         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 118                                               ll_kobj);
 119         struct obd_statfs osfs;
 120         int rc;
 121
 122         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
 123                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 124                                 OBD_STATFS_NODELAY);
 125         if (!rc) {
 126                 __u32 blk_size = osfs.os_bsize >> 10;
 127                 __u64 result = osfs.os_bavail;
 128
 129                 while (blk_size >>= 1)
 130                         result <<= 1;
 131
 132                 rc = sprintf(buf, "%llu\n", result);
 133         }
 134
 135         return rc;
 136 }
 137 LUSTRE_RO_ATTR(kbytesavail);
 138
 139 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 140                                char *buf)
 141 {
 142         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 143                                               ll_kobj);
 144         struct obd_statfs osfs;
 145         int rc;
 146
 147         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
 148                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 149                                 OBD_STATFS_NODELAY);
 150         if (!rc)
 151                 return sprintf(buf, "%llu\n", osfs.os_files);
 152
 153         return rc;
 154 }
 155 LUSTRE_RO_ATTR(filestotal);
 156
 157 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 158                               char *buf)
 159 {
 160         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 161                                               ll_kobj);
 162         struct obd_statfs osfs;
 163         int rc;
 164
 165         rc = ll_statfs_internal(sbi->ll_sb, &osfs,
 166                                 cfs_time_shift_64(-OBD_STATFS_CACHE_SECONDS),
 167                                 OBD_STATFS_NODELAY);
 168         if (!rc)
 169                 return sprintf(buf, "%llu\n", osfs.os_ffree);
 170
 171         return rc;
 172 }
 173 LUSTRE_RO_ATTR(filesfree);
 174
 175 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
 176                                 char *buf)
 177 {
 178         return sprintf(buf, "local client\n");
 179 }
 180 LUSTRE_RO_ATTR(client_type);
 181
 182 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 183                            char *buf)
 184 {
 185         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 186                                               ll_kobj);
 187
 188         return sprintf(buf, "%s\n", sbi->ll_sb->s_type->name);
 189 }
 190 LUSTRE_RO_ATTR(fstype);
 191
 192 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 193                          char *buf)
 194 {
 195         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 196                                               ll_kobj);
 197
 198         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 199 }
 200 LUSTRE_RO_ATTR(uuid);
 201
 202 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 203 {
 204         struct super_block *sb = m->private;
 205
 206         /*
 207          * See description of statistical counters in struct cl_site, and
 208          * struct lu_site.
 209          */
 210         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 211 }
 212
 213 LPROC_SEQ_FOPS_RO(ll_site_stats);
 214
 215 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
 216                                       struct attribute *attr, char *buf)
 217 {
 218         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 219                                               ll_kobj);
 220         long pages_number;
 221         int mult;
 222
 223         spin_lock(&sbi->ll_lock);
 224         pages_number = sbi->ll_ra_info.ra_max_pages;
 225         spin_unlock(&sbi->ll_lock);
 226
 227         mult = 1 << (20 - PAGE_SHIFT);
 228         return lprocfs_read_frac_helper(buf, PAGE_SIZE, pages_number, mult);
 229 }
 230
 231 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
 232                                        struct attribute *attr,
 233                                        const char *buffer,
 234                                        size_t count)
 235 {
 236         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 237                                               ll_kobj);
 238         int rc;
 239         unsigned long pages_number;
 240
 241         rc = kstrtoul(buffer, 10, &pages_number);
 242         if (rc)
 243                 return rc;
 244
 245         pages_number *= 1 << (20 - PAGE_SHIFT); /* MB -> pages */
 246
 247         if (pages_number > totalram_pages / 2) {
 248                 CERROR("can't set file readahead more than %lu MB\n",
 249                        totalram_pages >> (20 - PAGE_SHIFT + 1)); /*1/2 of RAM*/
 250                 return -ERANGE;
 251         }
 252
 253         spin_lock(&sbi->ll_lock);
 254         sbi->ll_ra_info.ra_max_pages = pages_number;
 255         spin_unlock(&sbi->ll_lock);
 256
 257         return count;
 258 }
 259 LUSTRE_RW_ATTR(max_read_ahead_mb);
 260
 261 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
 262                                                struct attribute *attr,
 263                                                char *buf)
 264 {
 265         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 266                                               ll_kobj);
 267         long pages_number;
 268         int mult;
 269
 270         spin_lock(&sbi->ll_lock);
 271         pages_number = sbi->ll_ra_info.ra_max_pages_per_file;
 272         spin_unlock(&sbi->ll_lock);
 273
 274         mult = 1 << (20 - PAGE_SHIFT);
 275         return lprocfs_read_frac_helper(buf, PAGE_SIZE, pages_number, mult);
 276 }
 277
 278 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
 279                                                 struct attribute *attr,
 280                                                 const char *buffer,
 281                                                 size_t count)
 282 {
 283         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 284                                               ll_kobj);
 285         int rc;
 286         unsigned long pages_number;
 287
 288         rc = kstrtoul(buffer, 10, &pages_number);
 289         if (rc)
 290                 return rc;
 291
 292         if (pages_number > sbi->ll_ra_info.ra_max_pages) {
 293                 CERROR("can't set file readahead more than max_read_ahead_mb %lu MB\n",
 294                        sbi->ll_ra_info.ra_max_pages);
 295                 return -ERANGE;
 296         }
 297
 298         spin_lock(&sbi->ll_lock);
 299         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 300         spin_unlock(&sbi->ll_lock);
 301
 302         return count;
 303 }
 304 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
 305
 306 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
 307                                             struct attribute *attr,
 308                                             char *buf)
 309 {
 310         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 311                                               ll_kobj);
 312         long pages_number;
 313         int mult;
 314
 315         spin_lock(&sbi->ll_lock);
 316         pages_number = sbi->ll_ra_info.ra_max_read_ahead_whole_pages;
 317         spin_unlock(&sbi->ll_lock);
 318
 319         mult = 1 << (20 - PAGE_SHIFT);
 320         return lprocfs_read_frac_helper(buf, PAGE_SIZE, pages_number, mult);
 321 }
 322
 323 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
 324                                              struct attribute *attr,
 325                                              const char  *buffer,
 326                                              size_t count)
 327 {
 328         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 329                                               ll_kobj);
 330         int rc;
 331         unsigned long pages_number;
 332
 333         rc = kstrtoul(buffer, 10, &pages_number);
 334         if (rc)
 335                 return rc;
 336
 337         /* Cap this at the current max readahead window size, the readahead
 338          * algorithm does this anyway so it's pointless to set it larger.
 339          */
 340         if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 341                 CERROR("can't set max_read_ahead_whole_mb more than max_read_ahead_per_file_mb: %lu\n",
 342                        sbi->ll_ra_info.ra_max_pages_per_file >> (20 - PAGE_SHIFT));
 343                 return -ERANGE;
 344         }
 345
 346         spin_lock(&sbi->ll_lock);
 347         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 348         spin_unlock(&sbi->ll_lock);
 349
 350         return count;
 351 }
 352 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
 353
 354 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 355 {
 356         struct super_block     *sb    = m->private;
 357         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 358         struct cl_client_cache *cache = sbi->ll_cache;
 359         int shift = 20 - PAGE_SHIFT;
 360         long max_cached_mb;
 361         long unused_mb;
 362
 363         max_cached_mb = cache->ccc_lru_max >> shift;
 364         unused_mb = atomic_long_read(&cache->ccc_lru_left) >> shift;
 365         seq_printf(m,
 366                    "users: %d\n"
 367                    "max_cached_mb: %ld\n"
 368                    "used_mb: %ld\n"
 369                    "unused_mb: %ld\n"
 370                    "reclaim_count: %u\n",
 371                    atomic_read(&cache->ccc_users),
 372                    max_cached_mb,
 373                    max_cached_mb - unused_mb,
 374                    unused_mb,
 375                    cache->ccc_lru_shrinkers);
 376         return 0;
 377 }
 378
 379 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
 380                                           const char __user *buffer,
 381                                           size_t count, loff_t *off)
 382 {
 383         struct super_block *sb = ((struct seq_file *)file->private_data)->private;
 384         struct ll_sb_info *sbi = ll_s2sbi(sb);
 385         struct cl_client_cache *cache = sbi->ll_cache;
 386         struct lu_env *env;
 387         long diff = 0;
 388         long nrpages = 0;
 389         u16 refcheck;
 390         long pages_number;
 391         int mult;
 392         long rc;
 393         u64 val;
 394         char kernbuf[128];
 395
 396         if (count >= sizeof(kernbuf))
 397                 return -EINVAL;
 398
 399         if (copy_from_user(kernbuf, buffer, count))
 400                 return -EFAULT;
 401         kernbuf[count] = 0;
 402
 403         mult = 1 << (20 - PAGE_SHIFT);
 404         buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
 405                   kernbuf;
 406         rc = lprocfs_write_frac_u64_helper(buffer, count, &val, mult);
 407         if (rc)
 408                 return rc;
 409
 410         if (val > LONG_MAX)
 411                 return -ERANGE;
 412         pages_number = (long)val;
 413
 414         if (pages_number < 0 || pages_number > totalram_pages) {
 415                 CERROR("%s: can't set max cache more than %lu MB\n",
 416                        ll_get_fsname(sb, NULL, 0),
 417                        totalram_pages >> (20 - PAGE_SHIFT));
 418                 return -ERANGE;
 419         }
 420
 421         spin_lock(&sbi->ll_lock);
 422         diff = pages_number - cache->ccc_lru_max;
 423         spin_unlock(&sbi->ll_lock);
 424
 425         /* easy - add more LRU slots. */
 426         if (diff >= 0) {
 427                 atomic_long_add(diff, &cache->ccc_lru_left);
 428                 rc = 0;
 429                 goto out;
 430         }
 431
 432         env = cl_env_get(&refcheck);
 433         if (IS_ERR(env))
 434                 return 0;
 435
 436         diff = -diff;
 437         while (diff > 0) {
 438                 long tmp;
 439
 440                 /* reduce LRU budget from free slots. */
 441                 do {
 442                         long ov, nv;
 443
 444                         ov = atomic_long_read(&cache->ccc_lru_left);
 445                         if (ov == 0)
 446                                 break;
 447
 448                         nv = ov > diff ? ov - diff : 0;
 449                         rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
 450                         if (likely(ov == rc)) {
 451                                 diff -= ov - nv;
 452                                 nrpages += ov - nv;
 453                                 break;
 454                         }
 455                 } while (1);
 456
 457                 if (diff <= 0)
 458                         break;
 459
 460                 if (!sbi->ll_dt_exp) { /* being initialized */
 461                         rc = 0;
 462                         goto out;
 463                 }
 464
 465                 /* difficult - have to ask OSCs to drop LRU slots. */
 466                 tmp = diff << 1;
 467                 rc = obd_set_info_async(env, sbi->ll_dt_exp,
 468                                         sizeof(KEY_CACHE_LRU_SHRINK),
 469                                         KEY_CACHE_LRU_SHRINK,
 470                                         sizeof(tmp), &tmp, NULL);
 471                 if (rc < 0)
 472                         break;
 473         }
 474         cl_env_put(env, &refcheck);
 475
 476 out:
 477         if (rc >= 0) {
 478                 spin_lock(&sbi->ll_lock);
 479                 cache->ccc_lru_max = pages_number;
 480                 spin_unlock(&sbi->ll_lock);
 481                 rc = count;
 482         } else {
 483                 atomic_long_add(nrpages, &cache->ccc_lru_left);
 484         }
 485         return rc;
 486 }
 487
 488 LPROC_SEQ_FOPS(ll_max_cached_mb);
 489
 490 static ssize_t checksum_pages_show(struct kobject *kobj, struct attribute *attr,
 491                                    char *buf)
 492 {
 493         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 494                                               ll_kobj);
 495
 496         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 497 }
 498
 499 static ssize_t checksum_pages_store(struct kobject *kobj,
 500                                     struct attribute *attr,
 501                                     const char *buffer,
 502                                     size_t count)
 503 {
 504         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 505                                               ll_kobj);
 506         int rc;
 507         unsigned long val;
 508
 509         if (!sbi->ll_dt_exp)
 510                 /* Not set up yet */
 511                 return -EAGAIN;
 512
 513         rc = kstrtoul(buffer, 10, &val);
 514         if (rc)
 515                 return rc;
 516         if (val)
 517                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 518         else
 519                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 520
 521         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 522                                 KEY_CHECKSUM, sizeof(val), &val, NULL);
 523         if (rc)
 524                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 525
 526         return count;
 527 }
 528 LUSTRE_RW_ATTR(checksum_pages);
 529
 530 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 531                               enum stats_track_type type)
 532 {
 533         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 534                                               ll_kobj);
 535
 536         if (sbi->ll_stats_track_type == type)
 537                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 538         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 539                 return sprintf(buf, "0 (all)\n");
 540         else
 541                 return sprintf(buf, "untracked\n");
 542 }
 543
 544 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 545                               size_t count,
 546                               enum stats_track_type type)
 547 {
 548         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 549                                               ll_kobj);
 550         int rc;
 551         unsigned long pid;
 552
 553         rc = kstrtoul(buffer, 10, &pid);
 554         if (rc)
 555                 return rc;
 556         sbi->ll_stats_track_id = pid;
 557         if (pid == 0)
 558                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 559         else
 560                 sbi->ll_stats_track_type = type;
 561         lprocfs_clear_stats(sbi->ll_stats);
 562         return count;
 563 }
 564
 565 static ssize_t stats_track_pid_show(struct kobject *kobj,
 566                                     struct attribute *attr,
 567                                     char *buf)
 568 {
 569         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 570 }
 571
 572 static ssize_t stats_track_pid_store(struct kobject *kobj,
 573                                      struct attribute *attr,
 574                                      const char *buffer,
 575                                      size_t count)
 576 {
 577         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 578 }
 579 LUSTRE_RW_ATTR(stats_track_pid);
 580
 581 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 582                                      struct attribute *attr,
 583                                      char *buf)
 584 {
 585         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 586 }
 587
 588 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 589                                       struct attribute *attr,
 590                                       const char *buffer,
 591                                       size_t count)
 592 {
 593         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 594 }
 595 LUSTRE_RW_ATTR(stats_track_ppid);
 596
 597 static ssize_t stats_track_gid_show(struct kobject *kobj,
 598                                     struct attribute *attr,
 599                                     char *buf)
 600 {
 601         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 602 }
 603
 604 static ssize_t stats_track_gid_store(struct kobject *kobj,
 605                                      struct attribute *attr,
 606                                      const char *buffer,
 607                                      size_t count)
 608 {
 609         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 610 }
 611 LUSTRE_RW_ATTR(stats_track_gid);
 612
 613 static ssize_t statahead_max_show(struct kobject *kobj,
 614                                   struct attribute *attr,
 615                                   char *buf)
 616 {
 617         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 618                                               ll_kobj);
 619
 620         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 621 }
 622
 623 static ssize_t statahead_max_store(struct kobject *kobj,
 624                                    struct attribute *attr,
 625                                    const char *buffer,
 626                                    size_t count)
 627 {
 628         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 629                                               ll_kobj);
 630         int rc;
 631         unsigned long val;
 632
 633         rc = kstrtoul(buffer, 10, &val);
 634         if (rc)
 635                 return rc;
 636
 637         if (val <= LL_SA_RPC_MAX)
 638                 sbi->ll_sa_max = val;
 639         else
 640                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 641                        val, LL_SA_RPC_MAX);
 642
 643         return count;
 644 }
 645 LUSTRE_RW_ATTR(statahead_max);
 646
 647 static ssize_t statahead_agl_show(struct kobject *kobj,
 648                                   struct attribute *attr,
 649                                   char *buf)
 650 {
 651         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 652                                               ll_kobj);
 653
 654         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
 655 }
 656
 657 static ssize_t statahead_agl_store(struct kobject *kobj,
 658                                    struct attribute *attr,
 659                                    const char *buffer,
 660                                    size_t count)
 661 {
 662         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 663                                               ll_kobj);
 664         int rc;
 665         unsigned long val;
 666
 667         rc = kstrtoul(buffer, 10, &val);
 668         if (rc)
 669                 return rc;
 670
 671         if (val)
 672                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 673         else
 674                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
 675
 676         return count;
 677 }
 678 LUSTRE_RW_ATTR(statahead_agl);
 679
 680 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 681 {
 682         struct super_block *sb = m->private;
 683         struct ll_sb_info *sbi = ll_s2sbi(sb);
 684
 685         seq_printf(m,
 686                    "statahead total: %u\n"
 687                    "statahead wrong: %u\n"
 688                    "agl total: %u\n",
 689                    atomic_read(&sbi->ll_sa_total),
 690                    atomic_read(&sbi->ll_sa_wrong),
 691                    atomic_read(&sbi->ll_agl_total));
 692         return 0;
 693 }
 694
 695 LPROC_SEQ_FOPS_RO(ll_statahead_stats);
 696
 697 static ssize_t lazystatfs_show(struct kobject *kobj,
 698                                struct attribute *attr,
 699                                char *buf)
 700 {
 701         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 702                                               ll_kobj);
 703
 704         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_LAZYSTATFS ? 1 : 0);
 705 }
 706
 707 static ssize_t lazystatfs_store(struct kobject *kobj,
 708                                 struct attribute *attr,
 709                                 const char *buffer,
 710                                 size_t count)
 711 {
 712         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 713                                               ll_kobj);
 714         int rc;
 715         unsigned long val;
 716
 717         rc = kstrtoul(buffer, 10, &val);
 718         if (rc)
 719                 return rc;
 720
 721         if (val)
 722                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 723         else
 724                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 725
 726         return count;
 727 }
 728 LUSTRE_RW_ATTR(lazystatfs);
 729
 730 static ssize_t max_easize_show(struct kobject *kobj,
 731                                struct attribute *attr,
 732                                char *buf)
 733 {
 734         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 735                                               ll_kobj);
 736         unsigned int ealen;
 737         int rc;
 738
 739         rc = ll_get_max_mdsize(sbi, &ealen);
 740         if (rc)
 741                 return rc;
 742
 743         return sprintf(buf, "%u\n", ealen);
 744 }
 745 LUSTRE_RO_ATTR(max_easize);
 746
 747 /**
 748  * Get default_easize.
 749  *
 750  * \see client_obd::cl_default_mds_easize
 751  *
 752  * \param[in] kobj      kernel object for sysfs tree
 753  * \param[in] attr      attribute of this kernel object
 754  * \param[in] buf       buffer to write data into
 755  *
 756  * \retval positive     \a count on success
 757  * \retval negative     negated errno on failure
 758  */
 759 static ssize_t default_easize_show(struct kobject *kobj,
 760                                    struct attribute *attr,
 761                                    char *buf)
 762 {
 763         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 764                                               ll_kobj);
 765         unsigned int ealen;
 766         int rc;
 767
 768         rc = ll_get_default_mdsize(sbi, &ealen);
 769         if (rc)
 770                 return rc;
 771
 772         return sprintf(buf, "%u\n", ealen);
 773 }
 774
 775 /**
 776  * Set default_easize.
 777  *
 778  * Range checking on the passed value is handled by
 779  * ll_set_default_mdsize().
 780  *
 781  * \see client_obd::cl_default_mds_easize
 782  *
 783  * \param[in] kobj      kernel object for sysfs tree
 784  * \param[in] attr      attribute of this kernel object
 785  * \param[in] buffer    string passed from user space
 786  * \param[in] count     \a buffer length
 787  *
 788  * \retval positive     \a count on success
 789  * \retval negative     negated errno on failure
 790  */
 791 static ssize_t default_easize_store(struct kobject *kobj,
 792                                     struct attribute *attr,
 793                                     const char *buffer,
 794                                     size_t count)
 795 {
 796         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 797                                               ll_kobj);
 798         unsigned long val;
 799         int rc;
 800
 801         rc = kstrtoul(buffer, 10, &val);
 802         if (rc)
 803                 return rc;
 804
 805         rc = ll_set_default_mdsize(sbi, val);
 806         if (rc)
 807                 return rc;
 808
 809         return count;
 810 }
 811 LUSTRE_RW_ATTR(default_easize);
 812
 813 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 814 {
 815         const char *str[] = LL_SBI_FLAGS;
 816         struct super_block *sb = m->private;
 817         int flags = ll_s2sbi(sb)->ll_flags;
 818         int i = 0;
 819
 820         while (flags != 0) {
 821                 if (ARRAY_SIZE(str) <= i) {
 822                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi flags please.\n",
 823                                ll_get_fsname(sb, NULL, 0));
 824                         return -EINVAL;
 825                 }
 826
 827                 if (flags & 0x1)
 828                         seq_printf(m, "%s ", str[i]);
 829                 flags >>= 1;
 830                 ++i;
 831         }
 832         seq_puts(m, "\b\n");
 833         return 0;
 834 }
 835
 836 LPROC_SEQ_FOPS_RO(ll_sbi_flags);
 837
 838 static ssize_t xattr_cache_show(struct kobject *kobj,
 839                                 struct attribute *attr,
 840                                 char *buf)
 841 {
 842         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 843                                               ll_kobj);
 844
 845         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
 846 }
 847
 848 static ssize_t xattr_cache_store(struct kobject *kobj,
 849                                  struct attribute *attr,
 850                                  const char *buffer,
 851                                  size_t count)
 852 {
 853         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 854                                               ll_kobj);
 855         int rc;
 856         unsigned long val;
 857
 858         rc = kstrtoul(buffer, 10, &val);
 859         if (rc)
 860                 return rc;
 861
 862         if (val != 0 && val != 1)
 863                 return -ERANGE;
 864
 865         if (val == 1 && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
 866                 return -ENOTSUPP;
 867
 868         sbi->ll_xattr_cache_enabled = val;
 869
 870         return count;
 871 }
 872 LUSTRE_RW_ATTR(xattr_cache);
 873
 874 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
 875 {
 876         struct super_block     *sb    = m->private;
 877         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 878         struct cl_client_cache *cache = sbi->ll_cache;
 879         long pages;
 880         int mb;
 881
 882         pages = atomic_long_read(&cache->ccc_unstable_nr);
 883         mb = (pages * PAGE_SIZE) >> 20;
 884
 885         seq_printf(m,
 886                    "unstable_check:     %8d\n"
 887                    "unstable_pages: %12ld\n"
 888                    "unstable_mb:        %8d\n",
 889                    cache->ccc_unstable_check, pages, mb);
 890
 891         return 0;
 892 }
 893
 894 static ssize_t ll_unstable_stats_seq_write(struct file *file,
 895                                            const char __user *buffer,
 896                                            size_t count, loff_t *off)
 897 {
 898         struct super_block *sb = ((struct seq_file *)file->private_data)->private;
 899         struct ll_sb_info *sbi = ll_s2sbi(sb);
 900         char kernbuf[128];
 901         int val, rc;
 902
 903         if (!count)
 904                 return 0;
 905         if (count >= sizeof(kernbuf))
 906                 return -EINVAL;
 907
 908         if (copy_from_user(kernbuf, buffer, count))
 909                 return -EFAULT;
 910         kernbuf[count] = 0;
 911
 912         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
 913                   kernbuf;
 914         rc = lprocfs_write_helper(buffer, count, &val);
 915         if (rc < 0)
 916                 return rc;
 917
 918         /* borrow lru lock to set the value */
 919         spin_lock(&sbi->ll_cache->ccc_lru_lock);
 920         sbi->ll_cache->ccc_unstable_check = !!val;
 921         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
 922
 923         return count;
 924 }
 925 LPROC_SEQ_FOPS(ll_unstable_stats);
 926
 927 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
 928 {
 929         struct super_block *sb = m->private;
 930         struct ll_sb_info *sbi = ll_s2sbi(sb);
 931         struct root_squash_info *squash = &sbi->ll_squash;
 932
 933         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
 934         return 0;
 935 }
 936
 937 static ssize_t ll_root_squash_seq_write(struct file *file,
 938                                         const char __user *buffer,
 939                                         size_t count, loff_t *off)
 940 {
 941         struct seq_file *m = file->private_data;
 942         struct super_block *sb = m->private;
 943         struct ll_sb_info *sbi = ll_s2sbi(sb);
 944         struct root_squash_info *squash = &sbi->ll_squash;
 945
 946         return lprocfs_wr_root_squash(buffer, count, squash,
 947                                       ll_get_fsname(sb, NULL, 0));
 948 }
 949 LPROC_SEQ_FOPS(ll_root_squash);
 950
 951 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
 952 {
 953         struct super_block *sb = m->private;
 954         struct ll_sb_info *sbi = ll_s2sbi(sb);
 955         struct root_squash_info *squash = &sbi->ll_squash;
 956         int len;
 957
 958         down_read(&squash->rsi_sem);
 959         if (!list_empty(&squash->rsi_nosquash_nids)) {
 960                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
 961                                         &squash->rsi_nosquash_nids);
 962                 m->count += len;
 963                 seq_puts(m, "\n");
 964         } else {
 965                 seq_puts(m, "NONE\n");
 966         }
 967         up_read(&squash->rsi_sem);
 968
 969         return 0;
 970 }
 971
 972 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
 973                                           const char __user *buffer,
 974                                           size_t count, loff_t *off)
 975 {
 976         struct seq_file *m = file->private_data;
 977         struct super_block *sb = m->private;
 978         struct ll_sb_info *sbi = ll_s2sbi(sb);
 979         struct root_squash_info *squash = &sbi->ll_squash;
 980         int rc;
 981
 982         rc = lprocfs_wr_nosquash_nids(buffer, count, squash,
 983                                       ll_get_fsname(sb, NULL, 0));
 984         if (rc < 0)
 985                 return rc;
 986
 987         ll_compute_rootsquash_state(sbi);
 988
 989         return rc;
 990 }
 991
 992 LPROC_SEQ_FOPS(ll_nosquash_nids);
 993
 994 static struct lprocfs_vars lprocfs_llite_obd_vars[] = {
 995         /* { "mntpt_path",   ll_rd_path,             0, 0 }, */
 996         { "site",         &ll_site_stats_fops,    NULL, 0 },
 997         /* { "filegroups",   lprocfs_rd_filegroups,  0, 0 }, */
 998         { "max_cached_mb",    &ll_max_cached_mb_fops, NULL },
 999         { "statahead_stats",  &ll_statahead_stats_fops, NULL, 0 },
1000         { "unstable_stats",   &ll_unstable_stats_fops, NULL },
1001         { "sbi_flags",        &ll_sbi_flags_fops, NULL, 0 },
1002         { .name =       "root_squash",
1003           .fops =       &ll_root_squash_fops                    },
1004         { .name =               "nosquash_nids",
1005           .fops =               &ll_nosquash_nids_fops          },
1006         { NULL }
1007 };
1008
1009 #define MAX_STRING_SIZE 128
1010
1011 static struct attribute *llite_attrs[] = {
1012         &lustre_attr_blocksize.attr,
1013         &lustre_attr_kbytestotal.attr,
1014         &lustre_attr_kbytesfree.attr,
1015         &lustre_attr_kbytesavail.attr,
1016         &lustre_attr_filestotal.attr,
1017         &lustre_attr_filesfree.attr,
1018         &lustre_attr_client_type.attr,
1019         &lustre_attr_fstype.attr,
1020         &lustre_attr_uuid.attr,
1021         &lustre_attr_max_read_ahead_mb.attr,
1022         &lustre_attr_max_read_ahead_per_file_mb.attr,
1023         &lustre_attr_max_read_ahead_whole_mb.attr,
1024         &lustre_attr_checksum_pages.attr,
1025         &lustre_attr_stats_track_pid.attr,
1026         &lustre_attr_stats_track_ppid.attr,
1027         &lustre_attr_stats_track_gid.attr,
1028         &lustre_attr_statahead_max.attr,
1029         &lustre_attr_statahead_agl.attr,
1030         &lustre_attr_lazystatfs.attr,
1031         &lustre_attr_max_easize.attr,
1032         &lustre_attr_default_easize.attr,
1033         &lustre_attr_xattr_cache.attr,
1034         NULL,
1035 };
1036
1037 static void llite_sb_release(struct kobject *kobj)
1038 {
1039         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1040                                               ll_kobj);
1041         complete(&sbi->ll_kobj_unregister);
1042 }
1043
1044 static struct kobj_type llite_ktype = {
1045         .default_attrs  = llite_attrs,
1046         .sysfs_ops      = &lustre_sysfs_ops,
1047         .release        = llite_sb_release,
1048 };
1049
1050 static const struct llite_file_opcode {
1051         __u32       opcode;
1052         __u32       type;
1053         const char *opname;
1054 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1055         /* file operation */
1056         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1057         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1058         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1059                                    "read_bytes" },
1060         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1061                                    "write_bytes" },
1062         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_PAGES,
1063                                    "brw_read" },
1064         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_PAGES,
1065                                    "brw_write" },
1066         { LPROC_LL_IOCTL,         LPROCFS_TYPE_REGS, "ioctl" },
1067         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
1068         { LPROC_LL_RELEASE,     LPROCFS_TYPE_REGS, "close" },
1069         { LPROC_LL_MAP,     LPROCFS_TYPE_REGS, "mmap" },
1070         { LPROC_LL_LLSEEK,       LPROCFS_TYPE_REGS, "seek" },
1071         { LPROC_LL_FSYNC,         LPROCFS_TYPE_REGS, "fsync" },
1072         { LPROC_LL_READDIR,     LPROCFS_TYPE_REGS, "readdir" },
1073         /* inode operation */
1074         { LPROC_LL_SETATTR,     LPROCFS_TYPE_REGS, "setattr" },
1075         { LPROC_LL_TRUNC,         LPROCFS_TYPE_REGS, "truncate" },
1076         { LPROC_LL_FLOCK,         LPROCFS_TYPE_REGS, "flock" },
1077         { LPROC_LL_GETATTR,     LPROCFS_TYPE_REGS, "getattr" },
1078         /* dir inode operation */
1079         { LPROC_LL_CREATE,       LPROCFS_TYPE_REGS, "create" },
1080         { LPROC_LL_LINK,           LPROCFS_TYPE_REGS, "link" },
1081         { LPROC_LL_UNLINK,       LPROCFS_TYPE_REGS, "unlink" },
1082         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_REGS, "symlink" },
1083         { LPROC_LL_MKDIR,         LPROCFS_TYPE_REGS, "mkdir" },
1084         { LPROC_LL_RMDIR,         LPROCFS_TYPE_REGS, "rmdir" },
1085         { LPROC_LL_MKNOD,         LPROCFS_TYPE_REGS, "mknod" },
1086         { LPROC_LL_RENAME,       LPROCFS_TYPE_REGS, "rename" },
1087         /* special inode operation */
1088         { LPROC_LL_STAFS,         LPROCFS_TYPE_REGS, "statfs" },
1089         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
1090         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
1091         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
1092         { LPROC_LL_GETXATTR_HITS,  LPROCFS_TYPE_REGS, "getxattr_hits" },
1093         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
1094         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
1095         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
1096 };
1097
1098 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1099 {
1100         if (!sbi->ll_stats)
1101                 return;
1102         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1103                 lprocfs_counter_add(sbi->ll_stats, op, count);
1104         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1105                  sbi->ll_stats_track_id == current->pid)
1106                 lprocfs_counter_add(sbi->ll_stats, op, count);
1107         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1108                  sbi->ll_stats_track_id == current->real_parent->pid)
1109                 lprocfs_counter_add(sbi->ll_stats, op, count);
1110         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1111                  sbi->ll_stats_track_id ==
1112                         from_kgid(&init_user_ns, current_gid()))
1113                 lprocfs_counter_add(sbi->ll_stats, op, count);
1114 }
1115 EXPORT_SYMBOL(ll_stats_ops_tally);
1116
1117 static const char *ra_stat_string[] = {
1118         [RA_STAT_HIT] = "hits",
1119         [RA_STAT_MISS] = "misses",
1120         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1121         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1122         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1123         [RA_STAT_FAILED_MATCH] = "failed lock match",
1124         [RA_STAT_DISCARDED] = "read but discarded",
1125         [RA_STAT_ZERO_LEN] = "zero length file",
1126         [RA_STAT_ZERO_WINDOW] = "zero size window",
1127         [RA_STAT_EOF] = "read-ahead to EOF",
1128         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1129         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1130         [RA_STAT_FAILED_REACH_END] = "failed to reach end"
1131 };
1132
1133 int ldebugfs_register_mountpoint(struct dentry *parent,
1134                                  struct super_block *sb, char *osc, char *mdc)
1135 {
1136         struct lustre_sb_info *lsi = s2lsi(sb);
1137         struct ll_sb_info *sbi = ll_s2sbi(sb);
1138         struct obd_device *obd;
1139         struct dentry *dir;
1140         char name[MAX_STRING_SIZE + 1], *ptr;
1141         int err, id, len, rc;
1142
1143         name[MAX_STRING_SIZE] = '\0';
1144
1145         LASSERT(sbi);
1146         LASSERT(mdc);
1147         LASSERT(osc);
1148
1149         /* Get fsname */
1150         len = strlen(lsi->lsi_lmd->lmd_profile);
1151         ptr = strrchr(lsi->lsi_lmd->lmd_profile, '-');
1152         if (ptr && (strcmp(ptr, "-client") == 0))
1153                 len -= 7;
1154
1155         /* Mount info */
1156         snprintf(name, MAX_STRING_SIZE, "%.*s-%p", len,
1157                  lsi->lsi_lmd->lmd_profile, sb);
1158
1159         dir = ldebugfs_register(name, parent, NULL, NULL);
1160         if (IS_ERR_OR_NULL(dir)) {
1161                 err = dir ? PTR_ERR(dir) : -ENOMEM;
1162                 sbi->ll_debugfs_entry = NULL;
1163                 return err;
1164         }
1165         sbi->ll_debugfs_entry = dir;
1166
1167         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache", 0444,
1168                                  &vvp_dump_pgcache_file_ops, sbi);
1169         if (rc)
1170                 CWARN("Error adding the dump_page_cache file\n");
1171
1172         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1173                                  &ll_rw_extents_stats_fops, sbi);
1174         if (rc)
1175                 CWARN("Error adding the extent_stats file\n");
1176
1177         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1178                                  "extents_stats_per_process",
1179                                  0644, &ll_rw_extents_stats_pp_fops, sbi);
1180         if (rc)
1181                 CWARN("Error adding the extents_stats_per_process file\n");
1182
1183         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1184                                  &ll_rw_offset_stats_fops, sbi);
1185         if (rc)
1186                 CWARN("Error adding the offset_stats file\n");
1187
1188         /* File operations stats */
1189         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1190                                             LPROCFS_STATS_FLAG_NONE);
1191         if (!sbi->ll_stats) {
1192                 err = -ENOMEM;
1193                 goto out;
1194         }
1195         /* do counter init */
1196         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1197                 __u32 type = llite_opcode_table[id].type;
1198                 void *ptr = NULL;
1199
1200                 if (type & LPROCFS_TYPE_REGS)
1201                         ptr = "regs";
1202                 else if (type & LPROCFS_TYPE_BYTES)
1203                         ptr = "bytes";
1204                 else if (type & LPROCFS_TYPE_PAGES)
1205                         ptr = "pages";
1206                 lprocfs_counter_init(sbi->ll_stats,
1207                                      llite_opcode_table[id].opcode,
1208                                      (type & LPROCFS_CNTR_AVGMINMAX),
1209                                      llite_opcode_table[id].opname, ptr);
1210         }
1211         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1212                                       sbi->ll_stats);
1213         if (err)
1214                 goto out;
1215
1216         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1217                                                LPROCFS_STATS_FLAG_NONE);
1218         if (!sbi->ll_ra_stats) {
1219                 err = -ENOMEM;
1220                 goto out;
1221         }
1222
1223         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1224                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1225                                      ra_stat_string[id], "pages");
1226
1227         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1228                                       sbi->ll_ra_stats);
1229         if (err)
1230                 goto out;
1231
1232         err = ldebugfs_add_vars(sbi->ll_debugfs_entry,
1233                                 lprocfs_llite_obd_vars, sb);
1234         if (err)
1235                 goto out;
1236
1237         sbi->ll_kobj.kset = llite_kset;
1238         init_completion(&sbi->ll_kobj_unregister);
1239         err = kobject_init_and_add(&sbi->ll_kobj, &llite_ktype, NULL,
1240                                    "%s", name);
1241         if (err)
1242                 goto out;
1243
1244         /* MDC info */
1245         obd = class_name2obd(mdc);
1246
1247         err = sysfs_create_link(&sbi->ll_kobj, &obd->obd_kobj,
1248                                 obd->obd_type->typ_name);
1249         if (err)
1250                 goto out;
1251
1252         /* OSC */
1253         obd = class_name2obd(osc);
1254
1255         err = sysfs_create_link(&sbi->ll_kobj, &obd->obd_kobj,
1256                                 obd->obd_type->typ_name);
1257 out:
1258         if (err) {
1259                 ldebugfs_remove(&sbi->ll_debugfs_entry);
1260                 lprocfs_free_stats(&sbi->ll_ra_stats);
1261                 lprocfs_free_stats(&sbi->ll_stats);
1262         }
1263         return err;
1264 }
1265
1266 void ldebugfs_unregister_mountpoint(struct ll_sb_info *sbi)
1267 {
1268         if (sbi->ll_debugfs_entry) {
1269                 ldebugfs_remove(&sbi->ll_debugfs_entry);
1270                 kobject_put(&sbi->ll_kobj);
1271                 wait_for_completion(&sbi->ll_kobj_unregister);
1272                 lprocfs_free_stats(&sbi->ll_ra_stats);
1273                 lprocfs_free_stats(&sbi->ll_stats);
1274         }
1275 }
1276
1277 #undef MAX_STRING_SIZE
1278
1279 #define pct(a, b) (b ? a * 100 / b : 0)
1280
1281 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1282                                     struct seq_file *seq, int which)
1283 {
1284         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1285         unsigned long start, end, r, w;
1286         char *unitp = "KMGTPEZY";
1287         int i, units = 10;
1288         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1289
1290         read_cum = 0;
1291         write_cum = 0;
1292         start = 0;
1293
1294         for (i = 0; i < LL_HIST_MAX; i++) {
1295                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1296                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1297         }
1298
1299         for (i = 0; i < LL_HIST_MAX; i++) {
1300                 r = pp_info->pp_r_hist.oh_buckets[i];
1301                 w = pp_info->pp_w_hist.oh_buckets[i];
1302                 read_cum += r;
1303                 write_cum += w;
1304                 end = 1 << (i + LL_HIST_START - units);
1305                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4lu %4lu  | %14lu %4lu %4lu\n",
1306                            start, *unitp, end, *unitp,
1307                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1308                            r, pct(r, read_tot), pct(read_cum, read_tot),
1309                            w, pct(w, write_tot), pct(write_cum, write_tot));
1310                 start = end;
1311                 if (start == 1024) {
1312                         start = 1;
1313                         units += 10;
1314                         unitp++;
1315                 }
1316                 if (read_cum == read_tot && write_cum == write_tot)
1317                         break;
1318         }
1319 }
1320
1321 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1322 {
1323         struct timespec64 now;
1324         struct ll_sb_info *sbi = seq->private;
1325         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1326         int k;
1327
1328         ktime_get_real_ts64(&now);
1329
1330         if (!sbi->ll_rw_stats_on) {
1331                 seq_printf(seq, "disabled\n"
1332                            "write anything in this file to activate, then 0 or \"[D/d]isabled\" to deactivate\n");
1333                 return 0;
1334         }
1335         seq_printf(seq, "snapshot_time:  %llu.%09lu (secs.usecs)\n",
1336                    (s64)now.tv_sec, (unsigned long)now.tv_nsec);
1337         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1338         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1339                    "extents", "calls", "%", "cum%",
1340                    "calls", "%", "cum%");
1341         spin_lock(&sbi->ll_pp_extent_lock);
1342         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1343                 if (io_extents->pp_extents[k].pid != 0) {
1344                         seq_printf(seq, "\nPID: %d\n",
1345                                    io_extents->pp_extents[k].pid);
1346                         ll_display_extents_info(io_extents, seq, k);
1347                 }
1348         }
1349         spin_unlock(&sbi->ll_pp_extent_lock);
1350         return 0;
1351 }
1352
1353 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1354                                                 const char __user *buf,
1355                                                 size_t len,
1356                                                 loff_t *off)
1357 {
1358         struct seq_file *seq = file->private_data;
1359         struct ll_sb_info *sbi = seq->private;
1360         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1361         int i;
1362         int value = 1, rc = 0;
1363
1364         if (len == 0)
1365                 return -EINVAL;
1366
1367         rc = lprocfs_write_helper(buf, len, &value);
1368         if (rc < 0 && len < 16) {
1369                 char kernbuf[16];
1370
1371                 if (copy_from_user(kernbuf, buf, len))
1372                         return -EFAULT;
1373                 kernbuf[len] = 0;
1374
1375                 if (kernbuf[len - 1] == '\n')
1376                         kernbuf[len - 1] = 0;
1377
1378                 if (strcmp(kernbuf, "disabled") == 0 ||
1379                     strcmp(kernbuf, "Disabled") == 0)
1380                         value = 0;
1381         }
1382
1383         if (value == 0)
1384                 sbi->ll_rw_stats_on = 0;
1385         else
1386                 sbi->ll_rw_stats_on = 1;
1387
1388         spin_lock(&sbi->ll_pp_extent_lock);
1389         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1390                 io_extents->pp_extents[i].pid = 0;
1391                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1392                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1393         }
1394         spin_unlock(&sbi->ll_pp_extent_lock);
1395         return len;
1396 }
1397
1398 LPROC_SEQ_FOPS(ll_rw_extents_stats_pp);
1399
1400 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1401 {
1402         struct timespec64 now;
1403         struct ll_sb_info *sbi = seq->private;
1404         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1405
1406         ktime_get_real_ts64(&now);
1407
1408         if (!sbi->ll_rw_stats_on) {
1409                 seq_printf(seq, "disabled\n"
1410                            "write anything in this file to activate, then 0 or \"[D/d]isabled\" to deactivate\n");
1411                 return 0;
1412         }
1413         seq_printf(seq, "snapshot_time:  %llu.%09lu (secs.usecs)\n",
1414                    (u64)now.tv_sec, (unsigned long)now.tv_nsec);
1415
1416         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1417         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1418                    "extents", "calls", "%", "cum%",
1419                    "calls", "%", "cum%");
1420         spin_lock(&sbi->ll_lock);
1421         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1422         spin_unlock(&sbi->ll_lock);
1423
1424         return 0;
1425 }
1426
1427 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1428                                              const char __user *buf,
1429                                              size_t len, loff_t *off)
1430 {
1431         struct seq_file *seq = file->private_data;
1432         struct ll_sb_info *sbi = seq->private;
1433         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1434         int i;
1435         int value = 1, rc = 0;
1436
1437         if (len == 0)
1438                 return -EINVAL;
1439
1440         rc = lprocfs_write_helper(buf, len, &value);
1441         if (rc < 0 && len < 16) {
1442                 char kernbuf[16];
1443
1444                 if (copy_from_user(kernbuf, buf, len))
1445                         return -EFAULT;
1446                 kernbuf[len] = 0;
1447
1448                 if (kernbuf[len - 1] == '\n')
1449                         kernbuf[len - 1] = 0;
1450
1451                 if (strcmp(kernbuf, "disabled") == 0 ||
1452                     strcmp(kernbuf, "Disabled") == 0)
1453                         value = 0;
1454         }
1455
1456         if (value == 0)
1457                 sbi->ll_rw_stats_on = 0;
1458         else
1459                 sbi->ll_rw_stats_on = 1;
1460
1461         spin_lock(&sbi->ll_pp_extent_lock);
1462         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1463                 io_extents->pp_extents[i].pid = 0;
1464                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1465                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1466         }
1467         spin_unlock(&sbi->ll_pp_extent_lock);
1468
1469         return len;
1470 }
1471
1472 LPROC_SEQ_FOPS(ll_rw_extents_stats);
1473
1474 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1475                        struct ll_file_data *file, loff_t pos,
1476                        size_t count, int rw)
1477 {
1478         int i, cur = -1;
1479         struct ll_rw_process_info *process;
1480         struct ll_rw_process_info *offset;
1481         int *off_count = &sbi->ll_rw_offset_entry_count;
1482         int *process_count = &sbi->ll_offset_process_count;
1483         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1484
1485         if (!sbi->ll_rw_stats_on)
1486                 return;
1487         process = sbi->ll_rw_process_info;
1488         offset = sbi->ll_rw_offset_info;
1489
1490         spin_lock(&sbi->ll_pp_extent_lock);
1491         /* Extent statistics */
1492         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1493                 if (io_extents->pp_extents[i].pid == pid) {
1494                         cur = i;
1495                         break;
1496                 }
1497         }
1498
1499         if (cur == -1) {
1500                 /* new process */
1501                 sbi->ll_extent_process_count =
1502                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1503                 cur = sbi->ll_extent_process_count;
1504                 io_extents->pp_extents[cur].pid = pid;
1505                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1506                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1507         }
1508
1509         for (i = 0; (count >= (1 << LL_HIST_START << i)) &&
1510              (i < (LL_HIST_MAX - 1)); i++)
1511                 ;
1512         if (rw == 0) {
1513                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1514                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1515         } else {
1516                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1517                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1518         }
1519         spin_unlock(&sbi->ll_pp_extent_lock);
1520
1521         spin_lock(&sbi->ll_process_lock);
1522         /* Offset statistics */
1523         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1524                 if (process[i].rw_pid == pid) {
1525                         if (process[i].rw_last_file != file) {
1526                                 process[i].rw_range_start = pos;
1527                                 process[i].rw_last_file_pos = pos + count;
1528                                 process[i].rw_smallest_extent = count;
1529                                 process[i].rw_largest_extent = count;
1530                                 process[i].rw_offset = 0;
1531                                 process[i].rw_last_file = file;
1532                                 spin_unlock(&sbi->ll_process_lock);
1533                                 return;
1534                         }
1535                         if (process[i].rw_last_file_pos != pos) {
1536                                 *off_count =
1537                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1538                                 offset[*off_count].rw_op = process[i].rw_op;
1539                                 offset[*off_count].rw_pid = pid;
1540                                 offset[*off_count].rw_range_start =
1541                                         process[i].rw_range_start;
1542                                 offset[*off_count].rw_range_end =
1543                                         process[i].rw_last_file_pos;
1544                                 offset[*off_count].rw_smallest_extent =
1545                                         process[i].rw_smallest_extent;
1546                                 offset[*off_count].rw_largest_extent =
1547                                         process[i].rw_largest_extent;
1548                                 offset[*off_count].rw_offset =
1549                                         process[i].rw_offset;
1550                                 process[i].rw_op = rw;
1551                                 process[i].rw_range_start = pos;
1552                                 process[i].rw_smallest_extent = count;
1553                                 process[i].rw_largest_extent = count;
1554                                 process[i].rw_offset = pos -
1555                                         process[i].rw_last_file_pos;
1556                         }
1557                         if (process[i].rw_smallest_extent > count)
1558                                 process[i].rw_smallest_extent = count;
1559                         if (process[i].rw_largest_extent < count)
1560                                 process[i].rw_largest_extent = count;
1561                         process[i].rw_last_file_pos = pos + count;
1562                         spin_unlock(&sbi->ll_process_lock);
1563                         return;
1564                 }
1565         }
1566         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1567         process[*process_count].rw_pid = pid;
1568         process[*process_count].rw_op = rw;
1569         process[*process_count].rw_range_start = pos;
1570         process[*process_count].rw_last_file_pos = pos + count;
1571         process[*process_count].rw_smallest_extent = count;
1572         process[*process_count].rw_largest_extent = count;
1573         process[*process_count].rw_offset = 0;
1574         process[*process_count].rw_last_file = file;
1575         spin_unlock(&sbi->ll_process_lock);
1576 }
1577
1578 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1579 {
1580         struct timespec64 now;
1581         struct ll_sb_info *sbi = seq->private;
1582         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1583         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1584         int i;
1585
1586         ktime_get_real_ts64(&now);
1587
1588         if (!sbi->ll_rw_stats_on) {
1589                 seq_printf(seq, "disabled\n"
1590                            "write anything in this file to activate, then 0 or \"[D/d]isabled\" to deactivate\n");
1591                 return 0;
1592         }
1593         spin_lock(&sbi->ll_process_lock);
1594
1595         seq_printf(seq, "snapshot_time:  %llu.%09lu (secs.usecs)\n",
1596                    (s64)now.tv_sec, (unsigned long)now.tv_nsec);
1597         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
1598                    "R/W", "PID", "RANGE START", "RANGE END",
1599                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
1600         /* We stored the discontiguous offsets here; print them first */
1601         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
1602                 if (offset[i].rw_pid != 0)
1603                         seq_printf(seq,
1604                                    "%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1605                                    offset[i].rw_op == READ ? 'R' : 'W',
1606                                    offset[i].rw_pid,
1607                                    offset[i].rw_range_start,
1608                                    offset[i].rw_range_end,
1609                                    (unsigned long)offset[i].rw_smallest_extent,
1610                                    (unsigned long)offset[i].rw_largest_extent,
1611                                    offset[i].rw_offset);
1612         }
1613         /* Then print the current offsets for each process */
1614         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1615                 if (process[i].rw_pid != 0)
1616                         seq_printf(seq,
1617                                    "%3c %10d %14Lu %14Lu %17lu %17lu %14Lu",
1618                                    process[i].rw_op == READ ? 'R' : 'W',
1619                                    process[i].rw_pid,
1620                                    process[i].rw_range_start,
1621                                    process[i].rw_last_file_pos,
1622                                    (unsigned long)process[i].rw_smallest_extent,
1623                                    (unsigned long)process[i].rw_largest_extent,
1624                                    process[i].rw_offset);
1625         }
1626         spin_unlock(&sbi->ll_process_lock);
1627
1628         return 0;
1629 }
1630
1631 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
1632                                             const char __user *buf,
1633                                             size_t len, loff_t *off)
1634 {
1635         struct seq_file *seq = file->private_data;
1636         struct ll_sb_info *sbi = seq->private;
1637         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
1638         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
1639         int value = 1, rc = 0;
1640
1641         if (len == 0)
1642                 return -EINVAL;
1643
1644         rc = lprocfs_write_helper(buf, len, &value);
1645
1646         if (rc < 0 && len < 16) {
1647                 char kernbuf[16];
1648
1649                 if (copy_from_user(kernbuf, buf, len))
1650                         return -EFAULT;
1651                 kernbuf[len] = 0;
1652
1653                 if (kernbuf[len - 1] == '\n')
1654                         kernbuf[len - 1] = 0;
1655
1656                 if (strcmp(kernbuf, "disabled") == 0 ||
1657                     strcmp(kernbuf, "Disabled") == 0)
1658                         value = 0;
1659         }
1660
1661         if (value == 0)
1662                 sbi->ll_rw_stats_on = 0;
1663         else
1664                 sbi->ll_rw_stats_on = 1;
1665
1666         spin_lock(&sbi->ll_process_lock);
1667         sbi->ll_offset_process_count = 0;
1668         sbi->ll_rw_offset_entry_count = 0;
1669         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
1670                LL_PROCESS_HIST_MAX);
1671         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
1672                LL_OFFSET_HIST_MAX);
1673         spin_unlock(&sbi->ll_process_lock);
1674
1675         return len;
1676 }
1677
1678 LPROC_SEQ_FOPS(ll_rw_offset_stats);
1679
1680 void lprocfs_llite_init_vars(struct lprocfs_static_vars *lvars)
1681 {
1682         lvars->obd_vars = lprocfs_llite_obd_vars;
1683 }