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) 2012, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  */
  31 #define DEBUG_SUBSYSTEM S_LLITE
  32
  33 #include <linux/version.h>
  34 #include <linux/user_namespace.h>
  35 #include <linux/uidgid.h>
  36
  37 #include <uapi/linux/lustre/lustre_param.h>
  38 #include <lprocfs_status.h>
  39 #include <obd_support.h>
  40
  41 #include "llite_internal.h"
  42 #include "vvp_internal.h"
  43
  44 static struct kobject *llite_kobj;
  45 static struct dentry *llite_root;
  46
  47 static void llite_kobj_release(struct kobject *kobj)
  48 {
  49         if (!IS_ERR_OR_NULL(llite_root)) {
  50                 debugfs_remove(llite_root);
  51                 llite_root = NULL;
  52         }
  53
  54         kfree(kobj);
  55 }
  56
  57 static struct kobj_type llite_kobj_ktype = {
  58         .release        = llite_kobj_release,
  59         .sysfs_ops      = &lustre_sysfs_ops,
  60 };
  61
  62 int llite_tunables_register(void)
  63 {
  64         int rc;
  65
  66         llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
  67         if (!llite_kobj)
  68                 return -ENOMEM;
  69
  70         llite_kobj->kset = lustre_kset;
  71         rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
  72                                   &lustre_kset->kobj, "%s", "llite");
  73         if (rc)
  74                 goto free_kobj;
  75
  76         llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
  77         return 0;
  78
  79 free_kobj:
  80         kobject_put(llite_kobj);
  81         llite_kobj = NULL;
  82
  83         return rc;
  84 }
  85
  86 void llite_tunables_unregister(void)
  87 {
  88         kobject_put(llite_kobj);
  89         llite_kobj = NULL;
  90 }
  91
  92 /* <debugfs>/lustre/llite mount point registration */
  93 static const struct file_operations ll_rw_extents_stats_fops;
  94 static const struct file_operations ll_rw_extents_stats_pp_fops;
  95 static const struct file_operations ll_rw_offset_stats_fops;
  96
  97 /**
  98  * ll_stats_pid_write() - Determine if stats collection should be enabled
  99  * @buf: Buffer containing the data written
 100  * @len: Number of bytes in the buffer
 101  *
 102  * Several proc files begin collecting stats when a value is written, and stop
 103  * collecting when either '0' or 'disable' is written. This function checks the
 104  * written value to see if collection should be enabled or disabled.
 105  *
 106  * Return: If '0' or 'disable' is provided, 0 is returned. If the text
 107  * equivalent of a number is written, that number is returned. Otherwise,
 108  * 1 is returned. Non-zero return values indicate collection should be enabled.
 109  */
 110 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
 111 {
 112         unsigned long long value = 1;
 113         char kernbuf[16];
 114         int rc;
 115
 116         rc = kstrtoull_from_user(buf, len, 0, &value);
 117         if (rc < 0 && len < sizeof(kernbuf)) {
 118                 if (copy_from_user(kernbuf, buf, len))
 119                         return -EFAULT;
 120                 kernbuf[len] = 0;
 121
 122                 if (kernbuf[len - 1] == '\n')
 123                         kernbuf[len - 1] = 0;
 124
 125                 if (strncasecmp(kernbuf, "disable", 7) == 0)
 126                         value = 0;
 127         }
 128
 129         return value;
 130 }
 131
 132 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 133                               char *buf)
 134 {
 135         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 136                                               ll_kset.kobj);
 137         struct obd_statfs osfs;
 138         int rc;
 139
 140         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 141         if (rc)
 142                 return rc;
 143
 144         return sprintf(buf, "%u\n", osfs.os_bsize);
 145 }
 146 LUSTRE_RO_ATTR(blocksize);
 147
 148 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
 149                                    char *buf)
 150 {
 151         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 152                                               ll_kset.kobj);
 153
 154         return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
 155 }
 156
 157 static ssize_t stat_blocksize_store(struct kobject *kobj,
 158                                     struct attribute *attr,
 159                                     const char *buffer,
 160                                     size_t count)
 161 {
 162         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 163                                               ll_kset.kobj);
 164         unsigned int val;
 165         int rc;
 166
 167         rc = kstrtouint(buffer, 10, &val);
 168         if (rc)
 169                 return rc;
 170
 171         if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
 172                 return -ERANGE;
 173
 174         sbi->ll_stat_blksize = val;
 175
 176         return count;
 177 }
 178 LUSTRE_RW_ATTR(stat_blocksize);
 179
 180 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 181                                 char *buf)
 182 {
 183         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 184                                               ll_kset.kobj);
 185         struct obd_statfs osfs;
 186         u32 blk_size;
 187         u64 result;
 188         int rc;
 189
 190         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 191         if (rc)
 192                 return rc;
 193
 194         blk_size = osfs.os_bsize >> 10;
 195         result = osfs.os_blocks;
 196
 197         while (blk_size >>= 1)
 198                 result <<= 1;
 199
 200         return sprintf(buf, "%llu\n", result);
 201 }
 202 LUSTRE_RO_ATTR(kbytestotal);
 203
 204 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 205                                char *buf)
 206 {
 207         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 208                                               ll_kset.kobj);
 209         struct obd_statfs osfs;
 210         u32 blk_size;
 211         u64 result;
 212         int rc;
 213
 214         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 215         if (rc)
 216                 return rc;
 217
 218         blk_size = osfs.os_bsize >> 10;
 219         result = osfs.os_bfree;
 220
 221         while (blk_size >>= 1)
 222                 result <<= 1;
 223
 224         return sprintf(buf, "%llu\n", result);
 225 }
 226 LUSTRE_RO_ATTR(kbytesfree);
 227
 228 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 229                                 char *buf)
 230 {
 231         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 232                                               ll_kset.kobj);
 233         struct obd_statfs osfs;
 234         u32 blk_size;
 235         u64 result;
 236         int rc;
 237
 238         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 239         if (rc)
 240                 return rc;
 241
 242         blk_size = osfs.os_bsize >> 10;
 243         result = osfs.os_bavail;
 244
 245         while (blk_size >>= 1)
 246                 result <<= 1;
 247
 248         return sprintf(buf, "%llu\n", result);
 249 }
 250 LUSTRE_RO_ATTR(kbytesavail);
 251
 252 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 253                                char *buf)
 254 {
 255         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 256                                               ll_kset.kobj);
 257         struct obd_statfs osfs;
 258         int rc;
 259
 260         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 261         if (rc)
 262                 return rc;
 263
 264         return sprintf(buf, "%llu\n", osfs.os_files);
 265 }
 266 LUSTRE_RO_ATTR(filestotal);
 267
 268 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 269                               char *buf)
 270 {
 271         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 272                                               ll_kset.kobj);
 273         struct obd_statfs osfs;
 274         int rc;
 275
 276         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 277         if (rc)
 278                 return rc;
 279
 280         return sprintf(buf, "%llu\n", osfs.os_ffree);
 281 }
 282 LUSTRE_RO_ATTR(filesfree);
 283
 284 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
 285                                 char *buf)
 286 {
 287         return sprintf(buf, "local client\n");
 288 }
 289 LUSTRE_RO_ATTR(client_type);
 290
 291 LUSTRE_RW_ATTR(foreign_symlink_enable);
 292
 293 LUSTRE_RW_ATTR(foreign_symlink_prefix);
 294
 295 LUSTRE_RW_ATTR(foreign_symlink_upcall);
 296
 297 LUSTRE_WO_ATTR(foreign_symlink_upcall_info);
 298
 299 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 300                            char *buf)
 301 {
 302         return sprintf(buf, "lustre\n");
 303 }
 304 LUSTRE_RO_ATTR(fstype);
 305
 306 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 307                          char *buf)
 308 {
 309         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 310                                               ll_kset.kobj);
 311
 312         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 313 }
 314 LUSTRE_RO_ATTR(uuid);
 315
 316 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 317 {
 318         struct super_block *sb = m->private;
 319
 320         /*
 321          * See description of statistical counters in struct cl_site, and
 322          * struct lu_site.
 323          */
 324         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 325 }
 326
 327 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
 328
 329 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
 330                                       struct attribute *attr, char *buf)
 331 {
 332         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 333                                               ll_kset.kobj);
 334
 335         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 336                         PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 337 }
 338
 339 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
 340                                        struct attribute *attr,
 341                                        const char *buffer, size_t count)
 342 {
 343         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 344                                               ll_kset.kobj);
 345         u64 ra_max_mb, pages_number;
 346         int rc;
 347
 348         rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
 349         if (rc)
 350                 return rc;
 351
 352         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
 353         CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
 354                sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
 355         if (pages_number > cfs_totalram_pages() / 2) {
 356                 /* 1/2 of RAM */
 357                 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
 358                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 359                        PAGES_TO_MiB(cfs_totalram_pages() / 2));
 360                 return -ERANGE;
 361         }
 362
 363         spin_lock(&sbi->ll_lock);
 364         sbi->ll_ra_info.ra_max_pages = pages_number;
 365         spin_unlock(&sbi->ll_lock);
 366
 367         return count;
 368 }
 369 LUSTRE_RW_ATTR(max_read_ahead_mb);
 370
 371 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
 372                                                struct attribute *attr,
 373                                                char *buf)
 374 {
 375         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 376                                               ll_kset.kobj);
 377
 378         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 379                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 380 }
 381
 382 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
 383                                                 struct attribute *attr,
 384                                                 const char *buffer,
 385                                                 size_t count)
 386 {
 387         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 388                                               ll_kset.kobj);
 389         u64 ra_max_file_mb, pages_number;
 390         int rc;
 391
 392         rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
 393         if (rc)
 394                 return rc;
 395
 396         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
 397         if (pages_number > sbi->ll_ra_info.ra_max_pages) {
 398                 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
 399                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 400                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 401                 return -ERANGE;
 402         }
 403
 404         spin_lock(&sbi->ll_lock);
 405         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 406         spin_unlock(&sbi->ll_lock);
 407
 408         return count;
 409 }
 410 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
 411
 412 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
 413                                             struct attribute *attr, char *buf)
 414 {
 415         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 416                                               ll_kset.kobj);
 417
 418         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 419                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
 420 }
 421
 422 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
 423                                              struct attribute *attr,
 424                                              const char *buffer, size_t count)
 425 {
 426         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 427                                               ll_kset.kobj);
 428         u64 ra_max_whole_mb, pages_number;
 429         int rc;
 430
 431         rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
 432         if (rc)
 433                 return rc;
 434
 435         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
 436         /* Cap this at the current max readahead window size, the readahead
 437          * algorithm does this anyway so it's pointless to set it larger.
 438          */
 439         if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 440                 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
 441                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 442                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 443
 444                 return -ERANGE;
 445         }
 446
 447         spin_lock(&sbi->ll_lock);
 448         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 449         spin_unlock(&sbi->ll_lock);
 450
 451         return count;
 452 }
 453 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
 454
 455 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 456 {
 457         struct super_block     *sb    = m->private;
 458         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 459         struct cl_client_cache *cache = sbi->ll_cache;
 460         struct ll_ra_info *ra = &sbi->ll_ra_info;
 461         long max_cached_mb;
 462         long unused_mb;
 463
 464         mutex_lock(&cache->ccc_max_cache_mb_lock);
 465         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
 466         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
 467         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 468
 469         seq_printf(m, "users: %d\n"
 470                       "max_cached_mb: %ld\n"
 471                       "used_mb: %ld\n"
 472                       "unused_mb: %ld\n"
 473                       "reclaim_count: %u\n"
 474                       "max_read_ahead_mb: %lu\n"
 475                       "used_read_ahead_mb: %d\n",
 476                    atomic_read(&cache->ccc_users),
 477                    max_cached_mb,
 478                    max_cached_mb - unused_mb,
 479                    unused_mb,
 480                    cache->ccc_lru_shrinkers,
 481                    PAGES_TO_MiB(ra->ra_max_pages),
 482                    PAGES_TO_MiB(atomic_read(&ra->ra_cur_pages)));
 483         return 0;
 484 }
 485
 486 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
 487                                           const char __user *buffer,
 488                                           size_t count, loff_t *off)
 489 {
 490         struct seq_file *m = file->private_data;
 491         struct super_block *sb = m->private;
 492         struct ll_sb_info *sbi = ll_s2sbi(sb);
 493         struct cl_client_cache *cache = sbi->ll_cache;
 494         struct lu_env *env;
 495         long diff = 0;
 496         long nrpages = 0;
 497         __u16 refcheck;
 498         u64 pages_number;
 499         int rc;
 500         char kernbuf[128], *ptr;
 501
 502         ENTRY;
 503         if (count >= sizeof(kernbuf))
 504                 RETURN(-EINVAL);
 505
 506         if (copy_from_user(kernbuf, buffer, count))
 507                 RETURN(-EFAULT);
 508         kernbuf[count] = '\0';
 509
 510         ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
 511         rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
 512         if (rc)
 513                 RETURN(rc);
 514
 515         pages_number >>= PAGE_SHIFT;
 516
 517         if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
 518                 CERROR("%s: can't set max cache more than %lu MB\n",
 519                        sbi->ll_fsname,
 520                        PAGES_TO_MiB(cfs_totalram_pages()));
 521                 RETURN(-ERANGE);
 522         }
 523         /* Allow enough cache so clients can make well-formed RPCs */
 524         pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
 525
 526         mutex_lock(&cache->ccc_max_cache_mb_lock);
 527         diff = pages_number - cache->ccc_lru_max;
 528
 529         /* easy - add more LRU slots. */
 530         if (diff >= 0) {
 531                 atomic_long_add(diff, &cache->ccc_lru_left);
 532                 GOTO(out, rc = 0);
 533         }
 534
 535         env = cl_env_get(&refcheck);
 536         if (IS_ERR(env))
 537                 GOTO(out_unlock, rc = PTR_ERR(env));
 538
 539         diff = -diff;
 540         while (diff > 0) {
 541                 long tmp;
 542
 543                 /* reduce LRU budget from free slots. */
 544                 do {
 545                         long lru_left_old, lru_left_new, lru_left_ret;
 546
 547                         lru_left_old = atomic_long_read(&cache->ccc_lru_left);
 548                         if (lru_left_old == 0)
 549                                 break;
 550
 551                         lru_left_new = lru_left_old > diff ?
 552                                         lru_left_old - diff : 0;
 553                         lru_left_ret =
 554                                 atomic_long_cmpxchg(&cache->ccc_lru_left,
 555                                                     lru_left_old,
 556                                                     lru_left_new);
 557                         if (likely(lru_left_old == lru_left_ret)) {
 558                                 diff -= lru_left_old - lru_left_new;
 559                                 nrpages += lru_left_old - lru_left_new;
 560                                 break;
 561                         }
 562                 } while (1);
 563
 564                 if (diff <= 0)
 565                         break;
 566
 567                 if (sbi->ll_dt_exp == NULL) { /* being initialized */
 568                         rc = -ENODEV;
 569                         break;
 570                 }
 571
 572                 /* Request extra free slots to avoid them all being used
 573                  * by other processes before this can continue shrinking.
 574                  */
 575                 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
 576                 /* difficult - have to ask OSCs to drop LRU slots. */
 577                 rc = obd_set_info_async(env, sbi->ll_dt_exp,
 578                                 sizeof(KEY_CACHE_LRU_SHRINK),
 579                                 KEY_CACHE_LRU_SHRINK,
 580                                 sizeof(tmp), &tmp, NULL);
 581                 if (rc < 0)
 582                         break;
 583         }
 584         cl_env_put(env, &refcheck);
 585
 586 out:
 587         if (rc >= 0) {
 588                 cache->ccc_lru_max = pages_number;
 589                 rc = count;
 590         } else {
 591                 atomic_long_add(nrpages, &cache->ccc_lru_left);
 592         }
 593 out_unlock:
 594         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 595         return rc;
 596 }
 597 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
 598
 599 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
 600                               char *buf)
 601 {
 602         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 603                                               ll_kset.kobj);
 604
 605         return scnprintf(buf, PAGE_SIZE, "%u\n",
 606                          test_bit(LL_SBI_CHECKSUM, sbi->ll_flags));
 607 }
 608
 609 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
 610                                const char *buffer, size_t count)
 611 {
 612         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 613                                               ll_kset.kobj);
 614         bool val;
 615         int tmp;
 616         int rc;
 617
 618         if (!sbi->ll_dt_exp)
 619                 /* Not set up yet */
 620                 return -EAGAIN;
 621
 622         rc = kstrtobool(buffer, &val);
 623         if (rc)
 624                 return rc;
 625         if (val)
 626                 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 627         else
 628                 clear_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 629         tmp = val;
 630
 631         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 632                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
 633         if (rc)
 634                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 635
 636         return count;
 637 }
 638 LUSTRE_RW_ATTR(checksums);
 639
 640 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
 641
 642 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 643                               enum stats_track_type type)
 644 {
 645         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 646                                               ll_kset.kobj);
 647
 648         if (sbi->ll_stats_track_type == type)
 649                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 650         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 651                 return sprintf(buf, "0 (all)\n");
 652
 653         return sprintf(buf, "untracked\n");
 654 }
 655
 656 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 657                               size_t count, enum stats_track_type type)
 658 {
 659         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 660                                               ll_kset.kobj);
 661         unsigned long pid;
 662         int rc;
 663
 664         rc = kstrtoul(buffer, 10, &pid);
 665         if (rc)
 666                 return rc;
 667
 668         sbi->ll_stats_track_id = pid;
 669         if (pid == 0)
 670                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 671         else
 672                 sbi->ll_stats_track_type = type;
 673         lprocfs_clear_stats(sbi->ll_stats);
 674         return count;
 675 }
 676
 677 static ssize_t stats_track_pid_show(struct kobject *kobj,
 678                                     struct attribute *attr,
 679                                     char *buf)
 680 {
 681         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 682 }
 683
 684 static ssize_t stats_track_pid_store(struct kobject *kobj,
 685                                      struct attribute *attr,
 686                                      const char *buffer,
 687                                      size_t count)
 688 {
 689         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 690 }
 691 LUSTRE_RW_ATTR(stats_track_pid);
 692
 693 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 694                                      struct attribute *attr,
 695                                      char *buf)
 696 {
 697         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 698 }
 699
 700 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 701                                       struct attribute *attr,
 702                                       const char *buffer,
 703                                       size_t count)
 704 {
 705         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 706 }
 707 LUSTRE_RW_ATTR(stats_track_ppid);
 708
 709 static ssize_t stats_track_gid_show(struct kobject *kobj,
 710                                     struct attribute *attr,
 711                                     char *buf)
 712 {
 713         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 714 }
 715
 716 static ssize_t stats_track_gid_store(struct kobject *kobj,
 717                                      struct attribute *attr,
 718                                      const char *buffer,
 719                                      size_t count)
 720 {
 721         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 722 }
 723 LUSTRE_RW_ATTR(stats_track_gid);
 724
 725 static ssize_t statahead_running_max_show(struct kobject *kobj,
 726                                           struct attribute *attr,
 727                                           char *buf)
 728 {
 729         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 730                                               ll_kset.kobj);
 731
 732         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
 733 }
 734
 735 static ssize_t statahead_running_max_store(struct kobject *kobj,
 736                                            struct attribute *attr,
 737                                            const char *buffer,
 738                                            size_t count)
 739 {
 740         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 741                                               ll_kset.kobj);
 742         unsigned long val;
 743         int rc;
 744
 745         rc = kstrtoul(buffer, 0, &val);
 746         if (rc)
 747                 return rc;
 748
 749         if (val <= LL_SA_RUNNING_MAX) {
 750                 sbi->ll_sa_running_max = val;
 751                 return count;
 752         }
 753
 754         CERROR("Bad statahead_running_max value %lu. Valid values "
 755                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 756
 757         return -ERANGE;
 758 }
 759 LUSTRE_RW_ATTR(statahead_running_max);
 760
 761 static ssize_t statahead_max_show(struct kobject *kobj,
 762                                   struct attribute *attr,
 763                                   char *buf)
 764 {
 765         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 766                                               ll_kset.kobj);
 767
 768         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 769 }
 770
 771 static ssize_t statahead_max_store(struct kobject *kobj,
 772                                    struct attribute *attr,
 773                                    const char *buffer,
 774                                    size_t count)
 775 {
 776         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 777                                               ll_kset.kobj);
 778         unsigned long val;
 779         int rc;
 780
 781         rc = kstrtoul(buffer, 0, &val);
 782         if (rc)
 783                 return rc;
 784
 785         if (val <= LL_SA_RPC_MAX)
 786                 sbi->ll_sa_max = val;
 787         else
 788                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 789                        val, LL_SA_RPC_MAX);
 790
 791         return count;
 792 }
 793 LUSTRE_RW_ATTR(statahead_max);
 794
 795 static ssize_t statahead_agl_show(struct kobject *kobj,
 796                                   struct attribute *attr,
 797                                   char *buf)
 798 {
 799         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 800                                               ll_kset.kobj);
 801
 802         return scnprintf(buf, PAGE_SIZE, "%u\n",
 803                          test_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags));
 804 }
 805
 806 static ssize_t statahead_agl_store(struct kobject *kobj,
 807                                    struct attribute *attr,
 808                                    const char *buffer,
 809                                    size_t count)
 810 {
 811         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 812                                               ll_kset.kobj);
 813         bool val;
 814         int rc;
 815
 816         rc = kstrtobool(buffer, &val);
 817         if (rc)
 818                 return rc;
 819
 820         if (val)
 821                 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 822         else
 823                 clear_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 824
 825         return count;
 826 }
 827 LUSTRE_RW_ATTR(statahead_agl);
 828
 829 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 830 {
 831         struct super_block *sb = m->private;
 832         struct ll_sb_info *sbi = ll_s2sbi(sb);
 833
 834         seq_printf(m, "statahead total: %u\n"
 835                       "statahead wrong: %u\n"
 836                       "agl total: %u\n",
 837                    atomic_read(&sbi->ll_sa_total),
 838                    atomic_read(&sbi->ll_sa_wrong),
 839                    atomic_read(&sbi->ll_agl_total));
 840         return 0;
 841 }
 842
 843 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
 844
 845 static ssize_t lazystatfs_show(struct kobject *kobj,
 846                                struct attribute *attr,
 847                                char *buf)
 848 {
 849         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 850                                               ll_kset.kobj);
 851
 852         return scnprintf(buf, PAGE_SIZE, "%u\n",
 853                          test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags));
 854 }
 855
 856 static ssize_t lazystatfs_store(struct kobject *kobj,
 857                                 struct attribute *attr,
 858                                 const char *buffer,
 859                                 size_t count)
 860 {
 861         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 862                                               ll_kset.kobj);
 863         bool val;
 864         int rc;
 865
 866         rc = kstrtobool(buffer, &val);
 867         if (rc)
 868                 return rc;
 869
 870         if (val)
 871                 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
 872         else
 873                 clear_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
 874
 875         return count;
 876 }
 877 LUSTRE_RW_ATTR(lazystatfs);
 878
 879 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
 880                                    char *buf)
 881 {
 882         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 883                                               ll_kset.kobj);
 884
 885         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
 886 }
 887
 888 static ssize_t statfs_max_age_store(struct kobject *kobj,
 889                                     struct attribute *attr, const char *buffer,
 890                                     size_t count)
 891 {
 892         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 893                                               ll_kset.kobj);
 894         unsigned int val;
 895         int rc;
 896
 897         rc = kstrtouint(buffer, 10, &val);
 898         if (rc)
 899                 return rc;
 900         if (val > OBD_STATFS_CACHE_MAX_AGE)
 901                 return -EINVAL;
 902
 903         sbi->ll_statfs_max_age = val;
 904
 905         return count;
 906 }
 907 LUSTRE_RW_ATTR(statfs_max_age);
 908
 909 static ssize_t max_easize_show(struct kobject *kobj,
 910                                struct attribute *attr,
 911                                char *buf)
 912 {
 913         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 914                                               ll_kset.kobj);
 915         unsigned int ealen;
 916         int rc;
 917
 918         rc = ll_get_max_mdsize(sbi, &ealen);
 919         if (rc)
 920                 return rc;
 921
 922         /* Limit xattr size returned to userspace based on kernel maximum */
 923         return scnprintf(buf, PAGE_SIZE, "%u\n",
 924                          ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 925 }
 926 LUSTRE_RO_ATTR(max_easize);
 927
 928 /**
 929  * Get default_easize.
 930  *
 931  * \see client_obd::cl_default_mds_easize
 932  *
 933  * \param[in] m         seq_file handle
 934  * \param[in] v         unused for single entry
 935  *
 936  * \retval 0            on success
 937  * \retval negative     negated errno on failure
 938  */
 939 static ssize_t default_easize_show(struct kobject *kobj,
 940                                    struct attribute *attr,
 941                                    char *buf)
 942 {
 943         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 944                                               ll_kset.kobj);
 945         unsigned int ealen;
 946         int rc;
 947
 948         rc = ll_get_default_mdsize(sbi, &ealen);
 949         if (rc)
 950                 return rc;
 951
 952         /* Limit xattr size returned to userspace based on kernel maximum */
 953         return scnprintf(buf, PAGE_SIZE, "%u\n",
 954                          ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 955 }
 956
 957 /**
 958  * Set default_easize.
 959  *
 960  * Range checking on the passed value is handled by
 961  * ll_set_default_mdsize().
 962  *
 963  * \see client_obd::cl_default_mds_easize
 964  *
 965  * \param[in] file      proc file
 966  * \param[in] buffer    string passed from user space
 967  * \param[in] count     \a buffer length
 968  * \param[in] off       unused for single entry
 969  *
 970  * \retval positive     \a count on success
 971  * \retval negative     negated errno on failure
 972  */
 973 static ssize_t default_easize_store(struct kobject *kobj,
 974                                     struct attribute *attr,
 975                                     const char *buffer,
 976                                     size_t count)
 977 {
 978         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 979                                               ll_kset.kobj);
 980         unsigned int val;
 981         int rc;
 982
 983         if (count == 0)
 984                 return 0;
 985
 986         rc = kstrtouint(buffer, 10, &val);
 987         if (rc)
 988                 return rc;
 989
 990         rc = ll_set_default_mdsize(sbi, val);
 991         if (rc)
 992                 return rc;
 993
 994         return count;
 995 }
 996 LUSTRE_RW_ATTR(default_easize);
 997
 998 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
 999
1000 static ssize_t xattr_cache_show(struct kobject *kobj,
1001                                 struct attribute *attr,
1002                                 char *buf)
1003 {
1004         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1005                                               ll_kset.kobj);
1006
1007         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1008 }
1009
1010 static ssize_t xattr_cache_store(struct kobject *kobj,
1011                                  struct attribute *attr,
1012                                  const char *buffer,
1013                                  size_t count)
1014 {
1015         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1016                                               ll_kset.kobj);
1017         bool val;
1018         int rc;
1019
1020         rc = kstrtobool(buffer, &val);
1021         if (rc)
1022                 return rc;
1023
1024         if (val && !test_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags))
1025                 return -ENOTSUPP;
1026
1027         sbi->ll_xattr_cache_enabled = val;
1028         sbi->ll_xattr_cache_set = 1;
1029
1030         return count;
1031 }
1032 LUSTRE_RW_ATTR(xattr_cache);
1033
1034 static ssize_t tiny_write_show(struct kobject *kobj,
1035                                struct attribute *attr,
1036                                char *buf)
1037 {
1038         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1039                                               ll_kset.kobj);
1040
1041         return scnprintf(buf, PAGE_SIZE, "%u\n",
1042                          test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags));
1043 }
1044
1045 static ssize_t tiny_write_store(struct kobject *kobj,
1046                                 struct attribute *attr,
1047                                 const char *buffer,
1048                                 size_t count)
1049 {
1050         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1051                                               ll_kset.kobj);
1052         bool val;
1053         int rc;
1054
1055         rc = kstrtobool(buffer, &val);
1056         if (rc)
1057                 return rc;
1058
1059         spin_lock(&sbi->ll_lock);
1060         if (val)
1061                 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1062         else
1063                 clear_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1064         spin_unlock(&sbi->ll_lock);
1065
1066         return count;
1067 }
1068 LUSTRE_RW_ATTR(tiny_write);
1069
1070 static ssize_t parallel_dio_show(struct kobject *kobj,
1071                                  struct attribute *attr,
1072                                  char *buf)
1073 {
1074         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1075                                               ll_kset.kobj);
1076
1077         return snprintf(buf, PAGE_SIZE, "%u\n",
1078                         test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags));
1079 }
1080
1081 static ssize_t parallel_dio_store(struct kobject *kobj,
1082                                   struct attribute *attr,
1083                                   const char *buffer,
1084                                   size_t count)
1085 {
1086         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1087                                               ll_kset.kobj);
1088         bool val;
1089         int rc;
1090
1091         rc = kstrtobool(buffer, &val);
1092         if (rc)
1093                 return rc;
1094
1095         spin_lock(&sbi->ll_lock);
1096         if (val)
1097                 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1098         else
1099                 clear_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1100         spin_unlock(&sbi->ll_lock);
1101
1102         return count;
1103 }
1104 LUSTRE_RW_ATTR(parallel_dio);
1105
1106 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1107                                                struct attribute *attr,
1108                                                char *buf)
1109 {
1110         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1111                                               ll_kset.kobj);
1112
1113         return scnprintf(buf, PAGE_SIZE, "%u\n",
1114                          sbi->ll_ra_info.ra_async_max_active);
1115 }
1116
1117 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1118                                                  struct attribute *attr,
1119                                                  const char *buffer,
1120                                                  size_t count)
1121 {
1122         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1123                                               ll_kset.kobj);
1124         unsigned int val;
1125         int rc;
1126
1127         rc = kstrtouint(buffer, 10, &val);
1128         if (rc)
1129                 return rc;
1130
1131         /**
1132          * It doesn't make any sense to make it exceed what
1133          * workqueue could acutally support. This can easily
1134          * over subscripe the cores but Lustre internally
1135          * throttles to avoid those impacts.
1136          */
1137         if (val > WQ_UNBOUND_MAX_ACTIVE) {
1138                 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1139                        sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1140                 return -ERANGE;
1141         }
1142
1143         spin_lock(&sbi->ll_lock);
1144         sbi->ll_ra_info.ra_async_max_active = val;
1145         spin_unlock(&sbi->ll_lock);
1146
1147         return count;
1148 }
1149 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1150
1151 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1152                                                        struct attribute *attr,
1153                                                        char *buf)
1154 {
1155         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1156                                               ll_kset.kobj);
1157
1158         return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1159                          sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1160 }
1161
1162 static ssize_t
1163 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1164                                          struct attribute *attr,
1165                                          const char *buffer, size_t count)
1166 {
1167         unsigned long pages_number;
1168         unsigned long max_ra_per_file;
1169         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1170                                               ll_kset.kobj);
1171         int rc;
1172
1173         rc = kstrtoul(buffer, 10, &pages_number);
1174         if (rc)
1175                 return rc;
1176
1177         pages_number = MiB_TO_PAGES(pages_number);
1178         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1179         if (pages_number < 0 || pages_number > max_ra_per_file) {
1180                 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1181                        "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1182                        PAGES_TO_MiB(pages_number),
1183                        PAGES_TO_MiB(max_ra_per_file));
1184                 return -ERANGE;
1185         }
1186         sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1187
1188         return count;
1189 }
1190 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1191
1192 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1193                                         struct attribute *attr,char *buf)
1194 {
1195         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1196                                               ll_kset.kobj);
1197
1198         return snprintf(buf, PAGE_SIZE, "%lu\n",
1199                         sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1200 }
1201
1202 static ssize_t
1203 read_ahead_range_kb_store(struct kobject *kobj,
1204                                struct attribute *attr,
1205                                const char *buffer, size_t count)
1206 {
1207         unsigned long pages_number;
1208         unsigned long max_ra_per_file;
1209         u64 val;
1210         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1211                                               ll_kset.kobj);
1212         int rc;
1213
1214         rc = sysfs_memparse(buffer, count, &val, "KiB");
1215         if (rc < 0)
1216                 return rc;
1217
1218         pages_number = val >> PAGE_SHIFT;
1219         /* Disable mmap range read */
1220         if (pages_number == 0)
1221                 goto out;
1222
1223         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1224         if (pages_number > max_ra_per_file ||
1225             pages_number < RA_MIN_MMAP_RANGE_PAGES)
1226                 return -ERANGE;
1227
1228 out:
1229         spin_lock(&sbi->ll_lock);
1230         sbi->ll_ra_info.ra_range_pages = pages_number;
1231         spin_unlock(&sbi->ll_lock);
1232
1233         return count;
1234 }
1235 LUSTRE_RW_ATTR(read_ahead_range_kb);
1236
1237 static ssize_t fast_read_show(struct kobject *kobj,
1238                               struct attribute *attr,
1239                               char *buf)
1240 {
1241         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1242                                               ll_kset.kobj);
1243
1244         return scnprintf(buf, PAGE_SIZE, "%u\n",
1245                          test_bit(LL_SBI_FAST_READ, sbi->ll_flags));
1246 }
1247
1248 static ssize_t fast_read_store(struct kobject *kobj,
1249                                struct attribute *attr,
1250                                const char *buffer,
1251                                size_t count)
1252 {
1253         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1254                                               ll_kset.kobj);
1255         bool val;
1256         int rc;
1257
1258         rc = kstrtobool(buffer, &val);
1259         if (rc)
1260                 return rc;
1261
1262         spin_lock(&sbi->ll_lock);
1263         if (val)
1264                 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1265         else
1266                 clear_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1267         spin_unlock(&sbi->ll_lock);
1268
1269         return count;
1270 }
1271 LUSTRE_RW_ATTR(fast_read);
1272
1273 static ssize_t file_heat_show(struct kobject *kobj,
1274                               struct attribute *attr,
1275                               char *buf)
1276 {
1277         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1278                                               ll_kset.kobj);
1279
1280         return scnprintf(buf, PAGE_SIZE, "%u\n",
1281                          test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags));
1282 }
1283
1284 static ssize_t file_heat_store(struct kobject *kobj,
1285                                struct attribute *attr,
1286                                const char *buffer,
1287                                size_t count)
1288 {
1289         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1290                                               ll_kset.kobj);
1291         bool val;
1292         int rc;
1293
1294         rc = kstrtobool(buffer, &val);
1295         if (rc)
1296                 return rc;
1297
1298         spin_lock(&sbi->ll_lock);
1299         if (val)
1300                 set_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1301         else
1302                 clear_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1303         spin_unlock(&sbi->ll_lock);
1304
1305         return count;
1306 }
1307 LUSTRE_RW_ATTR(file_heat);
1308
1309 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1310                                           struct attribute *attr,
1311                                           char *buf)
1312 {
1313         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1314                                               ll_kset.kobj);
1315
1316         return scnprintf(buf, PAGE_SIZE, "%u\n",
1317                          (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1318 }
1319
1320 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1321                                            struct attribute *attr,
1322                                            const char *buffer,
1323                                            size_t count)
1324 {
1325         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1326                                               ll_kset.kobj);
1327         unsigned long val;
1328         int rc;
1329
1330         rc = kstrtoul(buffer, 10, &val);
1331         if (rc)
1332                 return rc;
1333
1334         if (val < 0 || val > 100)
1335                 return -ERANGE;
1336
1337         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1338
1339         return count;
1340 }
1341 LUSTRE_RW_ATTR(heat_decay_percentage);
1342
1343 static ssize_t heat_period_second_show(struct kobject *kobj,
1344                                        struct attribute *attr,
1345                                        char *buf)
1346 {
1347         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1348                                               ll_kset.kobj);
1349
1350         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1351 }
1352
1353 static ssize_t heat_period_second_store(struct kobject *kobj,
1354                                         struct attribute *attr,
1355                                         const char *buffer,
1356                                         size_t count)
1357 {
1358         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1359                                               ll_kset.kobj);
1360         unsigned long val;
1361         int rc;
1362
1363         rc = kstrtoul(buffer, 10, &val);
1364         if (rc)
1365                 return rc;
1366
1367         if (val <= 0)
1368                 return -ERANGE;
1369
1370         sbi->ll_heat_period_second = val;
1371
1372         return count;
1373 }
1374 LUSTRE_RW_ATTR(heat_period_second);
1375
1376 static ssize_t opencache_threshold_count_show(struct kobject *kobj,
1377                                               struct attribute *attr,
1378                                               char *buf)
1379 {
1380         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1381                                               ll_kset.kobj);
1382
1383         if (sbi->ll_oc_thrsh_count)
1384                 return snprintf(buf, PAGE_SIZE, "%u\n",
1385                                 sbi->ll_oc_thrsh_count);
1386         else
1387                 return snprintf(buf, PAGE_SIZE, "off\n");
1388 }
1389
1390 static ssize_t opencache_threshold_count_store(struct kobject *kobj,
1391                                                struct attribute *attr,
1392                                                const char *buffer,
1393                                                size_t count)
1394 {
1395         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1396                                               ll_kset.kobj);
1397         unsigned int val;
1398         int rc;
1399
1400         rc = kstrtouint(buffer, 10, &val);
1401         if (rc) {
1402                 bool enable;
1403                 /* also accept "off" to disable and "on" to always cache */
1404                 rc = kstrtobool(buffer, &enable);
1405                 if (rc)
1406                         return rc;
1407                 val = enable;
1408         }
1409         sbi->ll_oc_thrsh_count = val;
1410
1411         return count;
1412 }
1413 LUSTRE_RW_ATTR(opencache_threshold_count);
1414
1415 static ssize_t opencache_threshold_ms_show(struct kobject *kobj,
1416                                            struct attribute *attr,
1417                                            char *buf)
1418 {
1419         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1420                                               ll_kset.kobj);
1421
1422         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_thrsh_ms);
1423 }
1424
1425 static ssize_t opencache_threshold_ms_store(struct kobject *kobj,
1426                                             struct attribute *attr,
1427                                             const char *buffer,
1428                                             size_t count)
1429 {
1430         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1431                                               ll_kset.kobj);
1432         unsigned int val;
1433         int rc;
1434
1435         rc = kstrtouint(buffer, 10, &val);
1436         if (rc)
1437                 return rc;
1438
1439         sbi->ll_oc_thrsh_ms = val;
1440
1441         return count;
1442 }
1443 LUSTRE_RW_ATTR(opencache_threshold_ms);
1444
1445 static ssize_t opencache_max_ms_show(struct kobject *kobj,
1446                                      struct attribute *attr,
1447                                      char *buf)
1448 {
1449         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1450                                               ll_kset.kobj);
1451
1452         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_max_ms);
1453 }
1454
1455 static ssize_t opencache_max_ms_store(struct kobject *kobj,
1456                                       struct attribute *attr,
1457                                       const char *buffer,
1458                                       size_t count)
1459 {
1460         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1461                                               ll_kset.kobj);
1462         unsigned int val;
1463         int rc;
1464
1465         rc = kstrtouint(buffer, 10, &val);
1466         if (rc)
1467                 return rc;
1468
1469         sbi->ll_oc_max_ms = val;
1470
1471         return count;
1472 }
1473 LUSTRE_RW_ATTR(opencache_max_ms);
1474
1475 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1476 {
1477         struct super_block      *sb    = m->private;
1478         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1479         struct cl_client_cache  *cache = sbi->ll_cache;
1480         long pages;
1481         int mb;
1482
1483         pages = atomic_long_read(&cache->ccc_unstable_nr);
1484         mb    = (pages * PAGE_SIZE) >> 20;
1485
1486         seq_printf(m, "unstable_check:     %8d\n"
1487                       "unstable_pages: %12ld\n"
1488                       "unstable_mb:        %8d\n",
1489                    cache->ccc_unstable_check, pages, mb);
1490         return 0;
1491 }
1492
1493 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1494                                            const char __user *buffer,
1495                                            size_t count, loff_t *unused)
1496 {
1497         struct seq_file *seq = file->private_data;
1498         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1499         char kernbuf[128];
1500         bool val;
1501         int rc;
1502
1503         if (count == 0)
1504                 return 0;
1505         if (count >= sizeof(kernbuf))
1506                 return -EINVAL;
1507
1508         if (copy_from_user(kernbuf, buffer, count))
1509                 return -EFAULT;
1510         kernbuf[count] = 0;
1511
1512         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1513                   kernbuf;
1514         rc = kstrtobool_from_user(buffer, count, &val);
1515         if (rc < 0)
1516                 return rc;
1517
1518         /* borrow lru lock to set the value */
1519         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1520         sbi->ll_cache->ccc_unstable_check = val;
1521         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1522
1523         return count;
1524 }
1525
1526 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1527
1528 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1529 {
1530         struct super_block *sb = m->private;
1531         struct ll_sb_info *sbi = ll_s2sbi(sb);
1532         struct root_squash_info *squash = &sbi->ll_squash;
1533
1534         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1535         return 0;
1536 }
1537
1538 static ssize_t ll_root_squash_seq_write(struct file *file,
1539                                         const char __user *buffer,
1540                                         size_t count, loff_t *off)
1541 {
1542         struct seq_file *m = file->private_data;
1543         struct super_block *sb = m->private;
1544         struct ll_sb_info *sbi = ll_s2sbi(sb);
1545         struct root_squash_info *squash = &sbi->ll_squash;
1546
1547         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1548 }
1549
1550 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1551
1552 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1553 {
1554         struct super_block *sb = m->private;
1555         struct ll_sb_info *sbi = ll_s2sbi(sb);
1556         struct root_squash_info *squash = &sbi->ll_squash;
1557         int len;
1558
1559         spin_lock(&squash->rsi_lock);
1560         if (!list_empty(&squash->rsi_nosquash_nids)) {
1561                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1562                                         &squash->rsi_nosquash_nids);
1563                 m->count += len;
1564                 seq_putc(m, '\n');
1565         } else {
1566                 seq_puts(m, "NONE\n");
1567         }
1568         spin_unlock(&squash->rsi_lock);
1569
1570         return 0;
1571 }
1572
1573 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1574                                           const char __user *buffer,
1575                                           size_t count, loff_t *off)
1576 {
1577         struct seq_file *m = file->private_data;
1578         struct super_block *sb = m->private;
1579         struct ll_sb_info *sbi = ll_s2sbi(sb);
1580         struct root_squash_info *squash = &sbi->ll_squash;
1581         int rc;
1582
1583         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1584         if (rc < 0)
1585                 return rc;
1586
1587         ll_compute_rootsquash_state(sbi);
1588
1589         return rc;
1590 }
1591
1592 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1593
1594 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1595 {
1596         struct super_block *sb = m->private;
1597         struct ll_sb_info *sbi = ll_s2sbi(sb);
1598
1599         return pcc_super_dump(&sbi->ll_pcc_super, m);
1600 }
1601
1602 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1603                                 size_t count, loff_t *off)
1604 {
1605         struct seq_file *m = file->private_data;
1606         struct super_block *sb = m->private;
1607         struct ll_sb_info *sbi = ll_s2sbi(sb);
1608         int rc;
1609         char *kernbuf;
1610
1611         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1612                 return -EINVAL;
1613
1614         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1615                 return -EOPNOTSUPP;
1616
1617         OBD_ALLOC(kernbuf, count + 1);
1618         if (kernbuf == NULL)
1619                 return -ENOMEM;
1620
1621         if (copy_from_user(kernbuf, buffer, count))
1622                 GOTO(out_free_kernbuff, rc = -EFAULT);
1623
1624         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1625 out_free_kernbuff:
1626         OBD_FREE(kernbuf, count + 1);
1627         return rc ? rc : count;
1628 }
1629 LDEBUGFS_SEQ_FOPS(ll_pcc);
1630
1631 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1632         { .name =       "site",
1633           .fops =       &ll_site_stats_fops                     },
1634         { .name =       "max_cached_mb",
1635           .fops =       &ll_max_cached_mb_fops                  },
1636         { .name =       "statahead_stats",
1637           .fops =       &ll_statahead_stats_fops                },
1638         { .name =       "unstable_stats",
1639           .fops =       &ll_unstable_stats_fops                 },
1640         { .name =       "sbi_flags",
1641           .fops =       &ll_sbi_flags_fops                      },
1642         { .name =       "root_squash",
1643           .fops =       &ll_root_squash_fops                    },
1644         { .name =       "nosquash_nids",
1645           .fops =       &ll_nosquash_nids_fops                  },
1646         { .name =       "pcc",
1647           .fops =       &ll_pcc_fops,                           },
1648         { NULL }
1649 };
1650
1651 #define MAX_STRING_SIZE 128
1652
1653 static struct attribute *llite_attrs[] = {
1654         &lustre_attr_blocksize.attr,
1655         &lustre_attr_stat_blocksize.attr,
1656         &lustre_attr_kbytestotal.attr,
1657         &lustre_attr_kbytesfree.attr,
1658         &lustre_attr_kbytesavail.attr,
1659         &lustre_attr_filestotal.attr,
1660         &lustre_attr_filesfree.attr,
1661         &lustre_attr_client_type.attr,
1662         &lustre_attr_foreign_symlink_enable.attr,
1663         &lustre_attr_foreign_symlink_prefix.attr,
1664         &lustre_attr_foreign_symlink_upcall.attr,
1665         &lustre_attr_foreign_symlink_upcall_info.attr,
1666         &lustre_attr_fstype.attr,
1667         &lustre_attr_uuid.attr,
1668         &lustre_attr_checksums.attr,
1669         &lustre_attr_checksum_pages.attr,
1670         &lustre_attr_max_read_ahead_mb.attr,
1671         &lustre_attr_max_read_ahead_per_file_mb.attr,
1672         &lustre_attr_max_read_ahead_whole_mb.attr,
1673         &lustre_attr_max_read_ahead_async_active.attr,
1674         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1675         &lustre_attr_read_ahead_range_kb.attr,
1676         &lustre_attr_stats_track_pid.attr,
1677         &lustre_attr_stats_track_ppid.attr,
1678         &lustre_attr_stats_track_gid.attr,
1679         &lustre_attr_statahead_running_max.attr,
1680         &lustre_attr_statahead_max.attr,
1681         &lustre_attr_statahead_agl.attr,
1682         &lustre_attr_lazystatfs.attr,
1683         &lustre_attr_statfs_max_age.attr,
1684         &lustre_attr_max_easize.attr,
1685         &lustre_attr_default_easize.attr,
1686         &lustre_attr_xattr_cache.attr,
1687         &lustre_attr_fast_read.attr,
1688         &lustre_attr_tiny_write.attr,
1689         &lustre_attr_parallel_dio.attr,
1690         &lustre_attr_file_heat.attr,
1691         &lustre_attr_heat_decay_percentage.attr,
1692         &lustre_attr_heat_period_second.attr,
1693         &lustre_attr_opencache_threshold_count.attr,
1694         &lustre_attr_opencache_threshold_ms.attr,
1695         &lustre_attr_opencache_max_ms.attr,
1696         NULL,
1697 };
1698
1699 static void sbi_kobj_release(struct kobject *kobj)
1700 {
1701         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1702                                               ll_kset.kobj);
1703         complete(&sbi->ll_kobj_unregister);
1704 }
1705
1706 static struct kobj_type sbi_ktype = {
1707         .default_attrs  = llite_attrs,
1708         .sysfs_ops      = &lustre_sysfs_ops,
1709         .release        = sbi_kobj_release,
1710 };
1711
1712 static const struct llite_file_opcode {
1713         __u32           opcode;
1714         __u32           type;
1715         const char      *opname;
1716 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1717         /* file operation */
1718         { LPROC_LL_READ_BYTES,  LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1719         { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1720         { LPROC_LL_READ,        LPROCFS_TYPE_LATENCY,   "read" },
1721         { LPROC_LL_WRITE,       LPROCFS_TYPE_LATENCY,   "write" },
1722         { LPROC_LL_IOCTL,       LPROCFS_TYPE_REQS,      "ioctl" },
1723         { LPROC_LL_OPEN,        LPROCFS_TYPE_LATENCY,   "open" },
1724         { LPROC_LL_RELEASE,     LPROCFS_TYPE_LATENCY,   "close" },
1725         { LPROC_LL_MMAP,        LPROCFS_TYPE_LATENCY,   "mmap" },
1726         { LPROC_LL_FAULT,       LPROCFS_TYPE_LATENCY,   "page_fault" },
1727         { LPROC_LL_MKWRITE,     LPROCFS_TYPE_LATENCY,   "page_mkwrite" },
1728         { LPROC_LL_LLSEEK,      LPROCFS_TYPE_LATENCY,   "seek" },
1729         { LPROC_LL_FSYNC,       LPROCFS_TYPE_LATENCY,   "fsync" },
1730         { LPROC_LL_READDIR,     LPROCFS_TYPE_LATENCY,   "readdir" },
1731         { LPROC_LL_INODE_OCOUNT,LPROCFS_TYPE_REQS |
1732                                 LPROCFS_CNTR_AVGMINMAX |
1733                                 LPROCFS_CNTR_STDDEV,    "opencount" },
1734         { LPROC_LL_INODE_OPCLTM,LPROCFS_TYPE_LATENCY,   "openclosetime" },
1735         /* inode operation */
1736         { LPROC_LL_SETATTR,     LPROCFS_TYPE_LATENCY,   "setattr" },
1737         { LPROC_LL_TRUNC,       LPROCFS_TYPE_LATENCY,   "truncate" },
1738         { LPROC_LL_FLOCK,       LPROCFS_TYPE_LATENCY,   "flock" },
1739         { LPROC_LL_GETATTR,     LPROCFS_TYPE_LATENCY,   "getattr" },
1740         { LPROC_LL_FALLOCATE,   LPROCFS_TYPE_LATENCY, "fallocate"},
1741         /* dir inode operation */
1742         { LPROC_LL_CREATE,      LPROCFS_TYPE_LATENCY,   "create" },
1743         { LPROC_LL_LINK,        LPROCFS_TYPE_LATENCY,   "link" },
1744         { LPROC_LL_UNLINK,      LPROCFS_TYPE_LATENCY,   "unlink" },
1745         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_LATENCY,   "symlink" },
1746         { LPROC_LL_MKDIR,       LPROCFS_TYPE_LATENCY,   "mkdir" },
1747         { LPROC_LL_RMDIR,       LPROCFS_TYPE_LATENCY,   "rmdir" },
1748         { LPROC_LL_MKNOD,       LPROCFS_TYPE_LATENCY,   "mknod" },
1749         { LPROC_LL_RENAME,      LPROCFS_TYPE_LATENCY,   "rename" },
1750         /* special inode operation */
1751         { LPROC_LL_STATFS,      LPROCFS_TYPE_LATENCY,   "statfs" },
1752         { LPROC_LL_SETXATTR,    LPROCFS_TYPE_LATENCY,   "setxattr" },
1753         { LPROC_LL_GETXATTR,    LPROCFS_TYPE_LATENCY,   "getxattr" },
1754         { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS,    "getxattr_hits" },
1755         { LPROC_LL_LISTXATTR,   LPROCFS_TYPE_LATENCY,   "listxattr" },
1756         { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY,   "removexattr" },
1757         { LPROC_LL_INODE_PERM,  LPROCFS_TYPE_LATENCY,   "inode_permission" },
1758 };
1759
1760 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1761 {
1762         if (!sbi->ll_stats)
1763                 return;
1764
1765         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1766                 lprocfs_counter_add(sbi->ll_stats, op, count);
1767         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1768                  sbi->ll_stats_track_id == current->pid)
1769                 lprocfs_counter_add(sbi->ll_stats, op, count);
1770         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1771                  sbi->ll_stats_track_id == current->parent->pid)
1772                 lprocfs_counter_add(sbi->ll_stats, op, count);
1773         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1774                  sbi->ll_stats_track_id ==
1775                         from_kgid(&init_user_ns, current_gid()))
1776                 lprocfs_counter_add(sbi->ll_stats, op, count);
1777 }
1778 EXPORT_SYMBOL(ll_stats_ops_tally);
1779
1780 static const char *const ra_stat_string[] = {
1781         [RA_STAT_HIT]                   = "hits",
1782         [RA_STAT_MISS]                  = "misses",
1783         [RA_STAT_DISTANT_READPAGE]      = "readpage_not_consecutive",
1784         [RA_STAT_MISS_IN_WINDOW]        = "miss_inside_window",
1785         [RA_STAT_FAILED_GRAB_PAGE]      = "failed_grab_cache_page",
1786         [RA_STAT_FAILED_MATCH]          = "failed_lock_match",
1787         [RA_STAT_DISCARDED]             = "read_but_discarded",
1788         [RA_STAT_ZERO_LEN]              = "zero_length_file",
1789         [RA_STAT_ZERO_WINDOW]           = "zero_size_window",
1790         [RA_STAT_EOF]                   = "readahead_to_eof",
1791         [RA_STAT_MAX_IN_FLIGHT]         = "hit_max_readahead_issue",
1792         [RA_STAT_WRONG_GRAB_PAGE]       = "wrong_page_from_grab_cache_page",
1793         [RA_STAT_FAILED_REACH_END]      = "failed_to_reach_end",
1794         [RA_STAT_ASYNC]                 = "async_readahead",
1795         [RA_STAT_FAILED_FAST_READ]      = "failed_to_fast_read",
1796         [RA_STAT_MMAP_RANGE_READ]       = "mmap_range_read",
1797 };
1798
1799 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1800 {
1801         struct lustre_sb_info *lsi = s2lsi(sb);
1802         struct ll_sb_info *sbi = ll_s2sbi(sb);
1803         int err, id;
1804
1805         ENTRY;
1806         LASSERT(sbi);
1807
1808         if (IS_ERR_OR_NULL(llite_root))
1809                 goto out_ll_kset;
1810
1811         sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1812         ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1813
1814         debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1815                             &vvp_dump_pgcache_file_ops);
1816
1817         debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1818                                  &ll_rw_extents_stats_fops);
1819
1820         debugfs_create_file("extents_stats_per_process", 0644,
1821                             sbi->ll_debugfs_entry, sbi,
1822                             &ll_rw_extents_stats_pp_fops);
1823
1824         debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1825                             &ll_rw_offset_stats_fops);
1826
1827         /* File operations stats */
1828         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1829                                             LPROCFS_STATS_FLAG_NONE);
1830         if (sbi->ll_stats == NULL)
1831                 GOTO(out_debugfs, err = -ENOMEM);
1832
1833         /* do counter init */
1834         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1835                 u32 type = llite_opcode_table[id].type;
1836                 void *ptr = "unknown";
1837
1838                 if (type & LPROCFS_TYPE_REQS)
1839                         ptr = "reqs";
1840                 else if (type & LPROCFS_TYPE_BYTES)
1841                         ptr = "bytes";
1842                 else if (type & LPROCFS_TYPE_USEC)
1843                         ptr = "usec";
1844                 lprocfs_counter_init(sbi->ll_stats,
1845                                      llite_opcode_table[id].opcode, type,
1846                                      llite_opcode_table[id].opname, ptr);
1847         }
1848
1849         debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1850                             sbi->ll_stats, &ldebugfs_stats_seq_fops);
1851
1852         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1853                                                LPROCFS_STATS_FLAG_NONE);
1854         if (sbi->ll_ra_stats == NULL)
1855                 GOTO(out_stats, err = -ENOMEM);
1856
1857         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1858                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1859                                      ra_stat_string[id], "pages");
1860
1861         debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1862                             sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1863
1864 out_ll_kset:
1865         /* Yes we also register sysfs mount kset here as well */
1866         sbi->ll_kset.kobj.parent = llite_kobj;
1867         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1868         init_completion(&sbi->ll_kobj_unregister);
1869         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1870         if (err)
1871                 GOTO(out_ra_stats, err);
1872
1873         err = kset_register(&sbi->ll_kset);
1874         if (err)
1875                 GOTO(out_ra_stats, err);
1876
1877         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1878
1879         RETURN(0);
1880 out_ra_stats:
1881         lprocfs_free_stats(&sbi->ll_ra_stats);
1882 out_stats:
1883         lprocfs_free_stats(&sbi->ll_stats);
1884 out_debugfs:
1885         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1886
1887         RETURN(err);
1888 }
1889
1890 void ll_debugfs_unregister_super(struct super_block *sb)
1891 {
1892         struct lustre_sb_info *lsi = s2lsi(sb);
1893         struct ll_sb_info *sbi = ll_s2sbi(sb);
1894
1895         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1896
1897         if (sbi->ll_dt_obd)
1898                 sysfs_remove_link(&sbi->ll_kset.kobj,
1899                                   sbi->ll_dt_obd->obd_type->typ_name);
1900
1901         if (sbi->ll_md_obd)
1902                 sysfs_remove_link(&sbi->ll_kset.kobj,
1903                                   sbi->ll_md_obd->obd_type->typ_name);
1904
1905         kobject_put(lsi->lsi_kobj);
1906
1907         kset_unregister(&sbi->ll_kset);
1908         wait_for_completion(&sbi->ll_kobj_unregister);
1909
1910         lprocfs_free_stats(&sbi->ll_ra_stats);
1911         lprocfs_free_stats(&sbi->ll_stats);
1912 }
1913 #undef MAX_STRING_SIZE
1914
1915 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1916                                    struct seq_file *seq, int which)
1917 {
1918         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1919         unsigned long start, end, r, w;
1920         char *unitp = "KMGTPEZY";
1921         int i, units = 10;
1922         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1923
1924         read_cum = 0;
1925         write_cum = 0;
1926         start = 0;
1927
1928         for(i = 0; i < LL_HIST_MAX; i++) {
1929                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1930                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1931         }
1932
1933         for(i = 0; i < LL_HIST_MAX; i++) {
1934                 r = pp_info->pp_r_hist.oh_buckets[i];
1935                 w = pp_info->pp_w_hist.oh_buckets[i];
1936                 read_cum += r;
1937                 write_cum += w;
1938                 end = 1 << (i + LL_HIST_START - units);
1939                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1940                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1941                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1942                            r, pct(r, read_tot), pct(read_cum, read_tot),
1943                            w, pct(w, write_tot), pct(write_cum, write_tot));
1944                 start = end;
1945                 if (start == (1 << 10)) {
1946                         start = 1;
1947                         units += 10;
1948                         unitp++;
1949                 }
1950                 if (read_cum == read_tot && write_cum == write_tot)
1951                         break;
1952         }
1953 }
1954
1955 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1956 {
1957         struct ll_sb_info *sbi = seq->private;
1958         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1959         int k;
1960
1961         if (!sbi->ll_rw_stats_on) {
1962                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1963                 return 0;
1964         }
1965         lprocfs_stats_header(seq, ktime_get(), io_extents->pp_init, 25, ":", 1);
1966         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1967         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1968                    "extents", "calls", "%", "cum%",
1969                    "calls", "%", "cum%");
1970         spin_lock(&sbi->ll_pp_extent_lock);
1971         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1972                 if (io_extents->pp_extents[k].pid != 0) {
1973                         seq_printf(seq, "\nPID: %d\n",
1974                                    io_extents->pp_extents[k].pid);
1975                         ll_display_extents_info(io_extents, seq, k);
1976                 }
1977         }
1978         spin_unlock(&sbi->ll_pp_extent_lock);
1979         return 0;
1980 }
1981
1982 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1983                                                 const char __user *buf,
1984                                                 size_t len,
1985                                                 loff_t *off)
1986 {
1987         struct seq_file *seq = file->private_data;
1988         struct ll_sb_info *sbi = seq->private;
1989         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1990         int i;
1991         __s64 value;
1992
1993         if (len == 0)
1994                 return -EINVAL;
1995
1996         value = ll_stats_pid_write(buf, len);
1997
1998         if (value == 0)
1999                 sbi->ll_rw_stats_on = 0;
2000         else
2001                 sbi->ll_rw_stats_on = 1;
2002
2003         spin_lock(&sbi->ll_pp_extent_lock);
2004         io_extents->pp_init = ktime_get();
2005         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2006                 io_extents->pp_extents[i].pid = 0;
2007                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
2008                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
2009         }
2010         spin_unlock(&sbi->ll_pp_extent_lock);
2011         return len;
2012 }
2013
2014 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
2015
2016 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
2017 {
2018         struct ll_sb_info *sbi = seq->private;
2019         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
2020
2021         if (!sbi->ll_rw_stats_on) {
2022                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2023                 return 0;
2024         }
2025
2026         lprocfs_stats_header(seq, ktime_get(), io_extents->pp_init, 25, ":", 1);
2027
2028         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
2029         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
2030                    "extents", "calls", "%", "cum%",
2031                    "calls", "%", "cum%");
2032         spin_lock(&sbi->ll_lock);
2033         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
2034         spin_unlock(&sbi->ll_lock);
2035
2036         return 0;
2037 }
2038
2039 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
2040                                              const char __user *buf,
2041                                              size_t len, loff_t *off)
2042 {
2043         struct seq_file *seq = file->private_data;
2044         struct ll_sb_info *sbi = seq->private;
2045         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
2046         int i;
2047         __s64 value;
2048
2049         if (len == 0)
2050                 return -EINVAL;
2051
2052         value = ll_stats_pid_write(buf, len);
2053
2054         if (value == 0)
2055                 sbi->ll_rw_stats_on = 0;
2056         else
2057                 sbi->ll_rw_stats_on = 1;
2058
2059         spin_lock(&sbi->ll_pp_extent_lock);
2060         io_extents->pp_init = ktime_get();
2061         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2062                 io_extents->pp_extents[i].pid = 0;
2063                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
2064                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
2065         }
2066         spin_unlock(&sbi->ll_pp_extent_lock);
2067
2068         return len;
2069 }
2070
2071 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
2072
2073 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
2074                        struct ll_file_data *file, loff_t pos,
2075                        size_t count, int rw)
2076 {
2077         int i, cur = -1;
2078         struct ll_rw_process_info *process;
2079         struct ll_rw_process_info *offset;
2080         int *off_count = &sbi->ll_rw_offset_entry_count;
2081         int *process_count = &sbi->ll_offset_process_count;
2082         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
2083
2084         if(!sbi->ll_rw_stats_on)
2085                 return;
2086         process = sbi->ll_rw_process_info;
2087         offset = sbi->ll_rw_offset_info;
2088
2089         spin_lock(&sbi->ll_pp_extent_lock);
2090         /* Extent statistics */
2091         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2092                 if(io_extents->pp_extents[i].pid == pid) {
2093                         cur = i;
2094                         break;
2095                 }
2096         }
2097
2098         if (cur == -1) {
2099                 /* new process */
2100                 sbi->ll_extent_process_count =
2101                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
2102                 cur = sbi->ll_extent_process_count;
2103                 io_extents->pp_extents[cur].pid = pid;
2104                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
2105                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
2106         }
2107
2108         for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
2109              (i < (LL_HIST_MAX - 1)); i++);
2110         if (rw == 0) {
2111                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
2112                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
2113         } else {
2114                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
2115                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
2116         }
2117         spin_unlock(&sbi->ll_pp_extent_lock);
2118
2119         spin_lock(&sbi->ll_process_lock);
2120         /* Offset statistics */
2121         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2122                 if (process[i].rw_pid == pid) {
2123                         if (process[i].rw_last_file != file) {
2124                                 process[i].rw_range_start = pos;
2125                                 process[i].rw_last_file_pos = pos + count;
2126                                 process[i].rw_smallest_extent = count;
2127                                 process[i].rw_largest_extent = count;
2128                                 process[i].rw_offset = 0;
2129                                 process[i].rw_last_file = file;
2130                                 spin_unlock(&sbi->ll_process_lock);
2131                                 return;
2132                         }
2133                         if (process[i].rw_last_file_pos != pos) {
2134                                 *off_count =
2135                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
2136                                 offset[*off_count].rw_op = process[i].rw_op;
2137                                 offset[*off_count].rw_pid = pid;
2138                                 offset[*off_count].rw_range_start =
2139                                         process[i].rw_range_start;
2140                                 offset[*off_count].rw_range_end =
2141                                         process[i].rw_last_file_pos;
2142                                 offset[*off_count].rw_smallest_extent =
2143                                         process[i].rw_smallest_extent;
2144                                 offset[*off_count].rw_largest_extent =
2145                                         process[i].rw_largest_extent;
2146                                 offset[*off_count].rw_offset =
2147                                         process[i].rw_offset;
2148                                 process[i].rw_op = rw;
2149                                 process[i].rw_range_start = pos;
2150                                 process[i].rw_smallest_extent = count;
2151                                 process[i].rw_largest_extent = count;
2152                                 process[i].rw_offset = pos -
2153                                         process[i].rw_last_file_pos;
2154                         }
2155                         if(process[i].rw_smallest_extent > count)
2156                                 process[i].rw_smallest_extent = count;
2157                         if(process[i].rw_largest_extent < count)
2158                                 process[i].rw_largest_extent = count;
2159                         process[i].rw_last_file_pos = pos + count;
2160                         spin_unlock(&sbi->ll_process_lock);
2161                         return;
2162                 }
2163         }
2164         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2165         process[*process_count].rw_pid = pid;
2166         process[*process_count].rw_op = rw;
2167         process[*process_count].rw_range_start = pos;
2168         process[*process_count].rw_last_file_pos = pos + count;
2169         process[*process_count].rw_smallest_extent = count;
2170         process[*process_count].rw_largest_extent = count;
2171         process[*process_count].rw_offset = 0;
2172         process[*process_count].rw_last_file = file;
2173         spin_unlock(&sbi->ll_process_lock);
2174 }
2175
2176 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2177 {
2178         struct ll_sb_info *sbi = seq->private;
2179         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2180         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2181         int i;
2182
2183         if (!sbi->ll_rw_stats_on) {
2184                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2185                 return 0;
2186         }
2187
2188         spin_lock(&sbi->ll_process_lock);
2189         lprocfs_stats_header(seq, ktime_get(), sbi->ll_process_stats_init, 25,
2190                              ":", true);
2191
2192         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2193                    "R/W", "PID", "RANGE START", "RANGE END",
2194                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2195
2196         /* We stored the discontiguous offsets here; print them first */
2197         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2198                 if (offset[i].rw_pid != 0)
2199                         seq_printf(seq,
2200                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2201                                    offset[i].rw_op == READ ? 'R' : 'W',
2202                                    offset[i].rw_pid,
2203                                    offset[i].rw_range_start,
2204                                    offset[i].rw_range_end,
2205                                    (unsigned long)offset[i].rw_smallest_extent,
2206                                    (unsigned long)offset[i].rw_largest_extent,
2207                                    offset[i].rw_offset);
2208         }
2209
2210         /* Then print the current offsets for each process */
2211         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2212                 if (process[i].rw_pid != 0)
2213                         seq_printf(seq,
2214                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2215                                    process[i].rw_op == READ ? 'R' : 'W',
2216                                    process[i].rw_pid,
2217                                    process[i].rw_range_start,
2218                                    process[i].rw_last_file_pos,
2219                                    (unsigned long)process[i].rw_smallest_extent,
2220                                    (unsigned long)process[i].rw_largest_extent,
2221                                    process[i].rw_offset);
2222         }
2223         spin_unlock(&sbi->ll_process_lock);
2224
2225         return 0;
2226 }
2227
2228 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2229                                             const char __user *buf,
2230                                             size_t len, loff_t *off)
2231 {
2232         struct seq_file *seq = file->private_data;
2233         struct ll_sb_info *sbi = seq->private;
2234         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2235         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2236         __s64 value;
2237
2238         if (len == 0)
2239                 return -EINVAL;
2240
2241         value = ll_stats_pid_write(buf, len);
2242
2243         if (value == 0)
2244                 sbi->ll_rw_stats_on = 0;
2245         else
2246                 sbi->ll_rw_stats_on = 1;
2247
2248         spin_lock(&sbi->ll_process_lock);
2249         sbi->ll_offset_process_count = 0;
2250         sbi->ll_rw_offset_entry_count = 0;
2251         sbi->ll_process_stats_init = ktime_get();
2252         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2253                LL_PROCESS_HIST_MAX);
2254         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2255                LL_OFFSET_HIST_MAX);
2256         spin_unlock(&sbi->ll_process_lock);
2257
2258         return len;
2259 }
2260
2261 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);