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 "lprocfs_status.h"
  43 #include "vvp_internal.h"
  44
  45 static struct kobject *llite_kobj;
  46 static struct dentry *llite_root;
  47
  48 static void llite_kobj_release(struct kobject *kobj)
  49 {
  50         if (!IS_ERR_OR_NULL(llite_root)) {
  51                 debugfs_remove(llite_root);
  52                 llite_root = NULL;
  53         }
  54
  55         kfree(kobj);
  56 }
  57
  58 static struct kobj_type llite_kobj_ktype = {
  59         .release        = llite_kobj_release,
  60         .sysfs_ops      = &lustre_sysfs_ops,
  61 };
  62
  63 int llite_tunables_register(void)
  64 {
  65         int rc;
  66
  67         llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
  68         if (!llite_kobj)
  69                 return -ENOMEM;
  70
  71         llite_kobj->kset = lustre_kset;
  72         rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
  73                                   &lustre_kset->kobj, "%s", "llite");
  74         if (rc)
  75                 goto free_kobj;
  76
  77         llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
  78         return 0;
  79
  80 free_kobj:
  81         kobject_put(llite_kobj);
  82         llite_kobj = NULL;
  83
  84         return rc;
  85 }
  86
  87 void llite_tunables_unregister(void)
  88 {
  89         kobject_put(llite_kobj);
  90         llite_kobj = NULL;
  91 }
  92
  93 /* <debugfs>/lustre/llite mount point registration */
  94 static const struct file_operations ll_rw_extents_stats_fops;
  95 static const struct file_operations ll_rw_extents_stats_pp_fops;
  96 static const struct file_operations ll_rw_offset_stats_fops;
  97
  98 /**
  99  * ll_stats_pid_write() - Determine if stats collection should be enabled
 100  * @buf: Buffer containing the data written
 101  * @len: Number of bytes in the buffer
 102  *
 103  * Several proc files begin collecting stats when a value is written, and stop
 104  * collecting when either '0' or 'disable' is written. This function checks the
 105  * written value to see if collection should be enabled or disabled.
 106  *
 107  * Return: If '0' or 'disable' is provided, 0 is returned. If the text
 108  * equivalent of a number is written, that number is returned. Otherwise,
 109  * 1 is returned. Non-zero return values indicate collection should be enabled.
 110  */
 111 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
 112 {
 113         unsigned long long value = 1;
 114         char kernbuf[16];
 115         int rc;
 116
 117         rc = kstrtoull_from_user(buf, len, 0, &value);
 118         if (rc < 0 && len < sizeof(kernbuf)) {
 119                 if (copy_from_user(kernbuf, buf, len))
 120                         return -EFAULT;
 121                 kernbuf[len] = 0;
 122
 123                 if (kernbuf[len - 1] == '\n')
 124                         kernbuf[len - 1] = 0;
 125
 126                 if (strncasecmp(kernbuf, "disable", 7) == 0)
 127                         value = 0;
 128         }
 129
 130         return value;
 131 }
 132
 133 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 134                               char *buf)
 135 {
 136         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 137                                               ll_kset.kobj);
 138         struct obd_statfs osfs;
 139         int rc;
 140
 141         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 142         if (rc)
 143                 return rc;
 144
 145         return sprintf(buf, "%u\n", osfs.os_bsize);
 146 }
 147 LUSTRE_RO_ATTR(blocksize);
 148
 149 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
 150                                    char *buf)
 151 {
 152         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 153                                               ll_kset.kobj);
 154
 155         return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
 156 }
 157
 158 static ssize_t stat_blocksize_store(struct kobject *kobj,
 159                                     struct attribute *attr,
 160                                     const char *buffer,
 161                                     size_t count)
 162 {
 163         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 164                                               ll_kset.kobj);
 165         unsigned int val;
 166         int rc;
 167
 168         rc = kstrtouint(buffer, 10, &val);
 169         if (rc)
 170                 return rc;
 171
 172         if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
 173                 return -ERANGE;
 174
 175         sbi->ll_stat_blksize = val;
 176
 177         return count;
 178 }
 179 LUSTRE_RW_ATTR(stat_blocksize);
 180
 181 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 182                                 char *buf)
 183 {
 184         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 185                                               ll_kset.kobj);
 186         struct obd_statfs osfs;
 187         u32 blk_size;
 188         u64 result;
 189         int rc;
 190
 191         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 192         if (rc)
 193                 return rc;
 194
 195         blk_size = osfs.os_bsize >> 10;
 196         result = osfs.os_blocks;
 197
 198         while (blk_size >>= 1)
 199                 result <<= 1;
 200
 201         return sprintf(buf, "%llu\n", result);
 202 }
 203 LUSTRE_RO_ATTR(kbytestotal);
 204
 205 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 206                                char *buf)
 207 {
 208         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 209                                               ll_kset.kobj);
 210         struct obd_statfs osfs;
 211         u32 blk_size;
 212         u64 result;
 213         int rc;
 214
 215         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 216         if (rc)
 217                 return rc;
 218
 219         blk_size = osfs.os_bsize >> 10;
 220         result = osfs.os_bfree;
 221
 222         while (blk_size >>= 1)
 223                 result <<= 1;
 224
 225         return sprintf(buf, "%llu\n", result);
 226 }
 227 LUSTRE_RO_ATTR(kbytesfree);
 228
 229 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 230                                 char *buf)
 231 {
 232         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 233                                               ll_kset.kobj);
 234         struct obd_statfs osfs;
 235         u32 blk_size;
 236         u64 result;
 237         int rc;
 238
 239         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 240         if (rc)
 241                 return rc;
 242
 243         blk_size = osfs.os_bsize >> 10;
 244         result = osfs.os_bavail;
 245
 246         while (blk_size >>= 1)
 247                 result <<= 1;
 248
 249         return sprintf(buf, "%llu\n", result);
 250 }
 251 LUSTRE_RO_ATTR(kbytesavail);
 252
 253 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 254                                char *buf)
 255 {
 256         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 257                                               ll_kset.kobj);
 258         struct obd_statfs osfs;
 259         int rc;
 260
 261         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 262         if (rc)
 263                 return rc;
 264
 265         return sprintf(buf, "%llu\n", osfs.os_files);
 266 }
 267 LUSTRE_RO_ATTR(filestotal);
 268
 269 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 270                               char *buf)
 271 {
 272         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 273                                               ll_kset.kobj);
 274         struct obd_statfs osfs;
 275         int rc;
 276
 277         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 278         if (rc)
 279                 return rc;
 280
 281         return sprintf(buf, "%llu\n", osfs.os_ffree);
 282 }
 283 LUSTRE_RO_ATTR(filesfree);
 284
 285 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
 286                                 char *buf)
 287 {
 288         return sprintf(buf, "local client\n");
 289 }
 290 LUSTRE_RO_ATTR(client_type);
 291
 292 LUSTRE_RW_ATTR(foreign_symlink_enable);
 293
 294 LUSTRE_RW_ATTR(foreign_symlink_prefix);
 295
 296 LUSTRE_RW_ATTR(foreign_symlink_upcall);
 297
 298 LUSTRE_WO_ATTR(foreign_symlink_upcall_info);
 299
 300 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 301                            char *buf)
 302 {
 303         return sprintf(buf, "lustre\n");
 304 }
 305 LUSTRE_RO_ATTR(fstype);
 306
 307 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 308                          char *buf)
 309 {
 310         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 311                                               ll_kset.kobj);
 312
 313         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 314 }
 315 LUSTRE_RO_ATTR(uuid);
 316
 317 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 318 {
 319         struct super_block *sb = m->private;
 320
 321         /*
 322          * See description of statistical counters in struct cl_site, and
 323          * struct lu_site.
 324          */
 325         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 326 }
 327
 328 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
 329
 330 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
 331                                       struct attribute *attr, char *buf)
 332 {
 333         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 334                                               ll_kset.kobj);
 335
 336         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 337                         PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 338 }
 339
 340 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
 341                                        struct attribute *attr,
 342                                        const char *buffer, size_t count)
 343 {
 344         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 345                                               ll_kset.kobj);
 346         u64 ra_max_mb, pages_number;
 347         int rc;
 348
 349         rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
 350         if (rc)
 351                 return rc;
 352
 353         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
 354         CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
 355                sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
 356         if (pages_number > cfs_totalram_pages() / 2) {
 357                 /* 1/2 of RAM */
 358                 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
 359                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 360                        PAGES_TO_MiB(cfs_totalram_pages() / 2));
 361                 return -ERANGE;
 362         }
 363
 364         spin_lock(&sbi->ll_lock);
 365         sbi->ll_ra_info.ra_max_pages = pages_number;
 366         spin_unlock(&sbi->ll_lock);
 367
 368         return count;
 369 }
 370 LUSTRE_RW_ATTR(max_read_ahead_mb);
 371
 372 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
 373                                                struct attribute *attr,
 374                                                char *buf)
 375 {
 376         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 377                                               ll_kset.kobj);
 378
 379         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 380                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 381 }
 382
 383 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
 384                                                 struct attribute *attr,
 385                                                 const char *buffer,
 386                                                 size_t count)
 387 {
 388         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 389                                               ll_kset.kobj);
 390         u64 ra_max_file_mb, pages_number;
 391         int rc;
 392
 393         rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
 394         if (rc)
 395                 return rc;
 396
 397         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
 398         if (pages_number > sbi->ll_ra_info.ra_max_pages) {
 399                 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
 400                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 401                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 402                 return -ERANGE;
 403         }
 404
 405         spin_lock(&sbi->ll_lock);
 406         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 407         spin_unlock(&sbi->ll_lock);
 408
 409         return count;
 410 }
 411 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
 412
 413 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
 414                                             struct attribute *attr, char *buf)
 415 {
 416         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 417                                               ll_kset.kobj);
 418
 419         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 420                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
 421 }
 422
 423 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
 424                                              struct attribute *attr,
 425                                              const char *buffer, size_t count)
 426 {
 427         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 428                                               ll_kset.kobj);
 429         u64 ra_max_whole_mb, pages_number;
 430         int rc;
 431
 432         rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
 433         if (rc)
 434                 return rc;
 435
 436         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
 437         /* Cap this at the current max readahead window size, the readahead
 438          * algorithm does this anyway so it's pointless to set it larger.
 439          */
 440         if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 441                 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
 442                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 443                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 444
 445                 return -ERANGE;
 446         }
 447
 448         spin_lock(&sbi->ll_lock);
 449         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 450         spin_unlock(&sbi->ll_lock);
 451
 452         return count;
 453 }
 454 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
 455
 456 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 457 {
 458         struct super_block     *sb    = m->private;
 459         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 460         struct cl_client_cache *cache = sbi->ll_cache;
 461         struct ll_ra_info *ra = &sbi->ll_ra_info;
 462         long max_cached_mb;
 463         long unused_mb;
 464
 465         mutex_lock(&cache->ccc_max_cache_mb_lock);
 466         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
 467         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
 468         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 469
 470         seq_printf(m, "users: %d\n"
 471                       "max_cached_mb: %ld\n"
 472                       "used_mb: %ld\n"
 473                       "unused_mb: %ld\n"
 474                       "reclaim_count: %u\n"
 475                       "max_read_ahead_mb: %lu\n"
 476                       "used_read_ahead_mb: %d\n",
 477                    refcount_read(&cache->ccc_users),
 478                    max_cached_mb,
 479                    max_cached_mb - unused_mb,
 480                    unused_mb,
 481                    cache->ccc_lru_shrinkers,
 482                    PAGES_TO_MiB(ra->ra_max_pages),
 483                    PAGES_TO_MiB(atomic_read(&ra->ra_cur_pages)));
 484         return 0;
 485 }
 486
 487 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
 488                                           const char __user *buffer,
 489                                           size_t count, loff_t *off)
 490 {
 491         struct seq_file *m = file->private_data;
 492         struct super_block *sb = m->private;
 493         struct ll_sb_info *sbi = ll_s2sbi(sb);
 494         struct cl_client_cache *cache = sbi->ll_cache;
 495         struct lu_env *env;
 496         long diff = 0;
 497         long nrpages = 0;
 498         __u16 refcheck;
 499         u64 value;
 500         u64 pages_number;
 501         int rc;
 502         char kernbuf[128], *ptr;
 503         bool percent = false;
 504
 505         ENTRY;
 506         if (count >= sizeof(kernbuf))
 507                 RETURN(-EINVAL);
 508
 509         if (copy_from_user(kernbuf, buffer, count))
 510                 RETURN(-EFAULT);
 511
 512         if (count > 0 && kernbuf[count - 1] == '%') {
 513                 percent = true;
 514                 /* strip off the % */
 515                 kernbuf[count - 1] = '\0';
 516         } else {
 517                 kernbuf[count] = '\0';
 518         }
 519
 520         ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
 521         if (percent)
 522                 rc = sysfs_memparse(ptr, count, &value, "B");
 523         else
 524                 rc = sysfs_memparse(ptr, count, &value, "MiB");
 525         if (rc)
 526                 RETURN(rc);
 527
 528         if (percent) {
 529                 pages_number = cfs_totalram_pages() * value / 100;
 530         } else {
 531                 pages_number = value >> PAGE_SHIFT;
 532         }
 533
 534         if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
 535                 CERROR("%s: can't set max cache more than %lu MB\n",
 536                        sbi->ll_fsname,
 537                        PAGES_TO_MiB(cfs_totalram_pages()));
 538                 RETURN(-ERANGE);
 539         }
 540         /* Allow enough cache so clients can make well-formed RPCs */
 541         pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
 542
 543         mutex_lock(&cache->ccc_max_cache_mb_lock);
 544         diff = pages_number - cache->ccc_lru_max;
 545
 546         /* easy - add more LRU slots. */
 547         if (diff >= 0) {
 548                 atomic_long_add(diff, &cache->ccc_lru_left);
 549                 GOTO(out, rc = 0);
 550         }
 551
 552         env = cl_env_get(&refcheck);
 553         if (IS_ERR(env))
 554                 GOTO(out_unlock, rc = PTR_ERR(env));
 555
 556         diff = -diff;
 557         while (diff > 0) {
 558                 long tmp;
 559
 560                 /* reduce LRU budget from free slots. */
 561                 do {
 562                         long lru_left_old, lru_left_new, lru_left_ret;
 563
 564                         lru_left_old = atomic_long_read(&cache->ccc_lru_left);
 565                         if (lru_left_old == 0)
 566                                 break;
 567
 568                         lru_left_new = lru_left_old > diff ?
 569                                         lru_left_old - diff : 0;
 570                         lru_left_ret =
 571                                 atomic_long_cmpxchg(&cache->ccc_lru_left,
 572                                                     lru_left_old,
 573                                                     lru_left_new);
 574                         if (likely(lru_left_old == lru_left_ret)) {
 575                                 diff -= lru_left_old - lru_left_new;
 576                                 nrpages += lru_left_old - lru_left_new;
 577                                 break;
 578                         }
 579                 } while (1);
 580
 581                 if (diff <= 0)
 582                         break;
 583
 584                 if (sbi->ll_dt_exp == NULL) { /* being initialized */
 585                         rc = -ENODEV;
 586                         break;
 587                 }
 588
 589                 /* Request extra free slots to avoid them all being used
 590                  * by other processes before this can continue shrinking.
 591                  */
 592                 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
 593                 /* difficult - have to ask OSCs to drop LRU slots. */
 594                 rc = obd_set_info_async(env, sbi->ll_dt_exp,
 595                                 sizeof(KEY_CACHE_LRU_SHRINK),
 596                                 KEY_CACHE_LRU_SHRINK,
 597                                 sizeof(tmp), &tmp, NULL);
 598                 if (rc < 0)
 599                         break;
 600         }
 601         cl_env_put(env, &refcheck);
 602
 603 out:
 604         if (rc >= 0) {
 605                 cache->ccc_lru_max = pages_number;
 606                 rc = count;
 607         } else {
 608                 atomic_long_add(nrpages, &cache->ccc_lru_left);
 609         }
 610 out_unlock:
 611         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 612         return rc;
 613 }
 614 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
 615
 616 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
 617                               char *buf)
 618 {
 619         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 620                                               ll_kset.kobj);
 621
 622         return scnprintf(buf, PAGE_SIZE, "%u\n",
 623                          test_bit(LL_SBI_CHECKSUM, sbi->ll_flags));
 624 }
 625
 626 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
 627                                const char *buffer, size_t count)
 628 {
 629         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 630                                               ll_kset.kobj);
 631         bool val;
 632         int tmp;
 633         int rc;
 634
 635         if (!sbi->ll_dt_exp)
 636                 /* Not set up yet */
 637                 return -EAGAIN;
 638
 639         rc = kstrtobool(buffer, &val);
 640         if (rc)
 641                 return rc;
 642         if (val)
 643                 set_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 644         else
 645                 clear_bit(LL_SBI_CHECKSUM, sbi->ll_flags);
 646         tmp = val;
 647
 648         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 649                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
 650         if (rc)
 651                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 652
 653         return count;
 654 }
 655 LUSTRE_RW_ATTR(checksums);
 656
 657 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
 658
 659 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 660                               enum stats_track_type type)
 661 {
 662         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 663                                               ll_kset.kobj);
 664
 665         if (sbi->ll_stats_track_type == type)
 666                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 667         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 668                 return sprintf(buf, "0 (all)\n");
 669
 670         return sprintf(buf, "untracked\n");
 671 }
 672
 673 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 674                               size_t count, enum stats_track_type type)
 675 {
 676         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 677                                               ll_kset.kobj);
 678         unsigned long pid;
 679         int rc;
 680
 681         rc = kstrtoul(buffer, 10, &pid);
 682         if (rc)
 683                 return rc;
 684
 685         sbi->ll_stats_track_id = pid;
 686         if (pid == 0)
 687                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 688         else
 689                 sbi->ll_stats_track_type = type;
 690         lprocfs_stats_clear(sbi->ll_stats);
 691         return count;
 692 }
 693
 694 static ssize_t stats_track_pid_show(struct kobject *kobj,
 695                                     struct attribute *attr,
 696                                     char *buf)
 697 {
 698         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 699 }
 700
 701 static ssize_t stats_track_pid_store(struct kobject *kobj,
 702                                      struct attribute *attr,
 703                                      const char *buffer,
 704                                      size_t count)
 705 {
 706         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 707 }
 708 LUSTRE_RW_ATTR(stats_track_pid);
 709
 710 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 711                                      struct attribute *attr,
 712                                      char *buf)
 713 {
 714         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 715 }
 716
 717 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 718                                       struct attribute *attr,
 719                                       const char *buffer,
 720                                       size_t count)
 721 {
 722         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 723 }
 724 LUSTRE_RW_ATTR(stats_track_ppid);
 725
 726 static ssize_t stats_track_gid_show(struct kobject *kobj,
 727                                     struct attribute *attr,
 728                                     char *buf)
 729 {
 730         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 731 }
 732
 733 static ssize_t stats_track_gid_store(struct kobject *kobj,
 734                                      struct attribute *attr,
 735                                      const char *buffer,
 736                                      size_t count)
 737 {
 738         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 739 }
 740 LUSTRE_RW_ATTR(stats_track_gid);
 741
 742 static ssize_t statahead_running_max_show(struct kobject *kobj,
 743                                           struct attribute *attr,
 744                                           char *buf)
 745 {
 746         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 747                                               ll_kset.kobj);
 748
 749         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
 750 }
 751
 752 static ssize_t statahead_running_max_store(struct kobject *kobj,
 753                                            struct attribute *attr,
 754                                            const char *buffer,
 755                                            size_t count)
 756 {
 757         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 758                                               ll_kset.kobj);
 759         unsigned long val;
 760         int rc;
 761
 762         rc = kstrtoul(buffer, 0, &val);
 763         if (rc)
 764                 return rc;
 765
 766         if (val <= LL_SA_RUNNING_MAX) {
 767                 sbi->ll_sa_running_max = val;
 768                 return count;
 769         }
 770
 771         CERROR("Bad statahead_running_max value %lu. Valid values "
 772                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 773
 774         return -ERANGE;
 775 }
 776 LUSTRE_RW_ATTR(statahead_running_max);
 777
 778 static ssize_t statahead_batch_max_show(struct kobject *kobj,
 779                                         struct attribute *attr,
 780                                         char *buf)
 781 {
 782         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 783                                               ll_kset.kobj);
 784
 785         return snprintf(buf, 16, "%u\n", sbi->ll_sa_batch_max);
 786 }
 787
 788 static ssize_t statahead_batch_max_store(struct kobject *kobj,
 789                                          struct attribute *attr,
 790                                          const char *buffer,
 791                                          size_t count)
 792 {
 793         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 794                                               ll_kset.kobj);
 795         unsigned long val;
 796         int rc;
 797
 798         rc = kstrtoul(buffer, 0, &val);
 799         if (rc)
 800                 return rc;
 801
 802         if (val > LL_SA_BATCH_MAX) {
 803                 CWARN("%s: statahead_batch_max value %lu limited to maximum %d\n",
 804                       sbi->ll_fsname, val, LL_SA_BATCH_MAX);
 805                 val = LL_SA_BATCH_MAX;
 806         }
 807
 808         sbi->ll_sa_batch_max = val;
 809         return count;
 810 }
 811 LUSTRE_RW_ATTR(statahead_batch_max);
 812
 813 static ssize_t statahead_max_show(struct kobject *kobj,
 814                                   struct attribute *attr,
 815                                   char *buf)
 816 {
 817         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 818                                               ll_kset.kobj);
 819
 820         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 821 }
 822
 823 static ssize_t statahead_max_store(struct kobject *kobj,
 824                                    struct attribute *attr,
 825                                    const char *buffer,
 826                                    size_t count)
 827 {
 828         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 829                                               ll_kset.kobj);
 830         unsigned long val;
 831         int rc;
 832
 833         rc = kstrtoul(buffer, 0, &val);
 834         if (rc)
 835                 return rc;
 836
 837         if (val > LL_SA_RPC_MAX) {
 838                 CWARN("%s: statahead_max value %lu limited to maximum %d\n",
 839                       sbi->ll_fsname, val, LL_SA_RPC_MAX);
 840                 val = LL_SA_RPC_MAX;
 841         }
 842
 843         sbi->ll_sa_max = val;
 844         return count;
 845 }
 846 LUSTRE_RW_ATTR(statahead_max);
 847
 848 static ssize_t statahead_agl_show(struct kobject *kobj,
 849                                   struct attribute *attr,
 850                                   char *buf)
 851 {
 852         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 853                                               ll_kset.kobj);
 854
 855         return scnprintf(buf, PAGE_SIZE, "%u\n",
 856                          test_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags));
 857 }
 858
 859 static ssize_t statahead_agl_store(struct kobject *kobj,
 860                                    struct attribute *attr,
 861                                    const char *buffer,
 862                                    size_t count)
 863 {
 864         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 865                                               ll_kset.kobj);
 866         bool val;
 867         int rc;
 868
 869         rc = kstrtobool(buffer, &val);
 870         if (rc)
 871                 return rc;
 872
 873         if (val)
 874                 set_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 875         else
 876                 clear_bit(LL_SBI_AGL_ENABLED, sbi->ll_flags);
 877
 878         return count;
 879 }
 880 LUSTRE_RW_ATTR(statahead_agl);
 881
 882 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 883 {
 884         struct super_block *sb = m->private;
 885         struct ll_sb_info *sbi = ll_s2sbi(sb);
 886
 887         seq_printf(m, "statahead total: %u\n"
 888                       "statahead wrong: %u\n"
 889                       "agl total: %u\n"
 890                       "hit_total: %u\n"
 891                       "miss_total: %u\n",
 892                    atomic_read(&sbi->ll_sa_total),
 893                    atomic_read(&sbi->ll_sa_wrong),
 894                    atomic_read(&sbi->ll_agl_total),
 895                    atomic_read(&sbi->ll_sa_hit_total),
 896                    atomic_read(&sbi->ll_sa_miss_total));
 897         return 0;
 898 }
 899
 900 static ssize_t ll_statahead_stats_seq_write(struct file *file,
 901                                             const char __user *buffer,
 902                                             size_t count, loff_t *off)
 903 {
 904         struct seq_file *m = file->private_data;
 905         struct super_block *sb = m->private;
 906         struct ll_sb_info *sbi = ll_s2sbi(sb);
 907
 908         atomic_set(&sbi->ll_sa_total, 0);
 909         atomic_set(&sbi->ll_sa_wrong, 0);
 910         atomic_set(&sbi->ll_agl_total, 0);
 911         atomic_set(&sbi->ll_sa_hit_total, 0);
 912         atomic_set(&sbi->ll_sa_miss_total, 0);
 913
 914         return count;
 915 }
 916 LDEBUGFS_SEQ_FOPS(ll_statahead_stats);
 917
 918 static ssize_t lazystatfs_show(struct kobject *kobj,
 919                                struct attribute *attr,
 920                                char *buf)
 921 {
 922         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 923                                               ll_kset.kobj);
 924
 925         return scnprintf(buf, PAGE_SIZE, "%u\n",
 926                          test_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags));
 927 }
 928
 929 static ssize_t lazystatfs_store(struct kobject *kobj,
 930                                 struct attribute *attr,
 931                                 const char *buffer,
 932                                 size_t count)
 933 {
 934         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 935                                               ll_kset.kobj);
 936         bool val;
 937         int rc;
 938
 939         rc = kstrtobool(buffer, &val);
 940         if (rc)
 941                 return rc;
 942
 943         if (val)
 944                 set_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
 945         else
 946                 clear_bit(LL_SBI_LAZYSTATFS, sbi->ll_flags);
 947
 948         return count;
 949 }
 950 LUSTRE_RW_ATTR(lazystatfs);
 951
 952 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
 953                                    char *buf)
 954 {
 955         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 956                                               ll_kset.kobj);
 957
 958         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
 959 }
 960
 961 static ssize_t statfs_max_age_store(struct kobject *kobj,
 962                                     struct attribute *attr, const char *buffer,
 963                                     size_t count)
 964 {
 965         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 966                                               ll_kset.kobj);
 967         unsigned int val;
 968         int rc;
 969
 970         rc = kstrtouint(buffer, 10, &val);
 971         if (rc)
 972                 return rc;
 973         if (val > OBD_STATFS_CACHE_MAX_AGE)
 974                 return -EINVAL;
 975
 976         sbi->ll_statfs_max_age = val;
 977
 978         return count;
 979 }
 980 LUSTRE_RW_ATTR(statfs_max_age);
 981
 982 static ssize_t max_easize_show(struct kobject *kobj,
 983                                struct attribute *attr,
 984                                char *buf)
 985 {
 986         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 987                                               ll_kset.kobj);
 988         unsigned int ealen;
 989         int rc;
 990
 991         rc = ll_get_max_mdsize(sbi, &ealen);
 992         if (rc)
 993                 return rc;
 994
 995         /* Limit xattr size returned to userspace based on kernel maximum */
 996         return scnprintf(buf, PAGE_SIZE, "%u\n",
 997                          ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 998 }
 999 LUSTRE_RO_ATTR(max_easize);
1000
1001 /**
1002  * Get default_easize.
1003  *
1004  * \see client_obd::cl_default_mds_easize
1005  *
1006  * \param[in] m         seq_file handle
1007  * \param[in] v         unused for single entry
1008  *
1009  * \retval 0            on success
1010  * \retval negative     negated errno on failure
1011  */
1012 static ssize_t default_easize_show(struct kobject *kobj,
1013                                    struct attribute *attr,
1014                                    char *buf)
1015 {
1016         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1017                                               ll_kset.kobj);
1018         unsigned int ealen;
1019         int rc;
1020
1021         rc = ll_get_default_mdsize(sbi, &ealen);
1022         if (rc)
1023                 return rc;
1024
1025         /* Limit xattr size returned to userspace based on kernel maximum */
1026         return scnprintf(buf, PAGE_SIZE, "%u\n",
1027                          ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
1028 }
1029
1030 /**
1031  * Set default_easize.
1032  *
1033  * Range checking on the passed value is handled by
1034  * ll_set_default_mdsize().
1035  *
1036  * \see client_obd::cl_default_mds_easize
1037  *
1038  * \param[in] file      proc file
1039  * \param[in] buffer    string passed from user space
1040  * \param[in] count     \a buffer length
1041  * \param[in] off       unused for single entry
1042  *
1043  * \retval positive     \a count on success
1044  * \retval negative     negated errno on failure
1045  */
1046 static ssize_t default_easize_store(struct kobject *kobj,
1047                                     struct attribute *attr,
1048                                     const char *buffer,
1049                                     size_t count)
1050 {
1051         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1052                                               ll_kset.kobj);
1053         unsigned int val;
1054         int rc;
1055
1056         if (count == 0)
1057                 return 0;
1058
1059         rc = kstrtouint(buffer, 10, &val);
1060         if (rc)
1061                 return rc;
1062
1063         rc = ll_set_default_mdsize(sbi, val);
1064         if (rc)
1065                 return rc;
1066
1067         return count;
1068 }
1069 LUSTRE_RW_ATTR(default_easize);
1070
1071 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1072
1073 static ssize_t xattr_cache_show(struct kobject *kobj,
1074                                 struct attribute *attr,
1075                                 char *buf)
1076 {
1077         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1078                                               ll_kset.kobj);
1079
1080         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1081 }
1082
1083 static ssize_t xattr_cache_store(struct kobject *kobj,
1084                                  struct attribute *attr,
1085                                  const char *buffer,
1086                                  size_t count)
1087 {
1088         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1089                                               ll_kset.kobj);
1090         bool val;
1091         int rc;
1092
1093         rc = kstrtobool(buffer, &val);
1094         if (rc)
1095                 return rc;
1096
1097         if (val && !test_bit(LL_SBI_XATTR_CACHE, sbi->ll_flags))
1098                 return -EOPNOTSUPP;
1099
1100         sbi->ll_xattr_cache_enabled = val;
1101         sbi->ll_xattr_cache_set = 1;
1102
1103         return count;
1104 }
1105 LUSTRE_RW_ATTR(xattr_cache);
1106
1107 static ssize_t tiny_write_show(struct kobject *kobj,
1108                                struct attribute *attr,
1109                                char *buf)
1110 {
1111         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1112                                               ll_kset.kobj);
1113
1114         return scnprintf(buf, PAGE_SIZE, "%u\n",
1115                          test_bit(LL_SBI_TINY_WRITE, sbi->ll_flags));
1116 }
1117
1118 static ssize_t tiny_write_store(struct kobject *kobj,
1119                                 struct attribute *attr,
1120                                 const char *buffer,
1121                                 size_t count)
1122 {
1123         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1124                                               ll_kset.kobj);
1125         bool val;
1126         int rc;
1127
1128         rc = kstrtobool(buffer, &val);
1129         if (rc)
1130                 return rc;
1131
1132         spin_lock(&sbi->ll_lock);
1133         if (val)
1134                 set_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1135         else
1136                 clear_bit(LL_SBI_TINY_WRITE, sbi->ll_flags);
1137         spin_unlock(&sbi->ll_lock);
1138
1139         return count;
1140 }
1141 LUSTRE_RW_ATTR(tiny_write);
1142
1143 static ssize_t parallel_dio_show(struct kobject *kobj,
1144                                  struct attribute *attr,
1145                                  char *buf)
1146 {
1147         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1148                                               ll_kset.kobj);
1149
1150         return snprintf(buf, PAGE_SIZE, "%u\n",
1151                         test_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags));
1152 }
1153
1154 static ssize_t parallel_dio_store(struct kobject *kobj,
1155                                   struct attribute *attr,
1156                                   const char *buffer,
1157                                   size_t count)
1158 {
1159         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1160                                               ll_kset.kobj);
1161         bool val;
1162         int rc;
1163
1164         rc = kstrtobool(buffer, &val);
1165         if (rc)
1166                 return rc;
1167
1168         spin_lock(&sbi->ll_lock);
1169         if (val)
1170                 set_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1171         else
1172                 clear_bit(LL_SBI_PARALLEL_DIO, sbi->ll_flags);
1173         spin_unlock(&sbi->ll_lock);
1174
1175         return count;
1176 }
1177 LUSTRE_RW_ATTR(parallel_dio);
1178
1179 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1180                                                struct attribute *attr,
1181                                                char *buf)
1182 {
1183         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1184                                               ll_kset.kobj);
1185
1186         return scnprintf(buf, PAGE_SIZE, "%u\n",
1187                          sbi->ll_ra_info.ra_async_max_active);
1188 }
1189
1190 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1191                                                  struct attribute *attr,
1192                                                  const char *buffer,
1193                                                  size_t count)
1194 {
1195         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1196                                               ll_kset.kobj);
1197         unsigned int val;
1198         int rc;
1199
1200         rc = kstrtouint(buffer, 10, &val);
1201         if (rc)
1202                 return rc;
1203
1204         /**
1205          * It doesn't make any sense to make it exceed what
1206          * workqueue could acutally support. This can easily
1207          * over subscripe the cores but Lustre internally
1208          * throttles to avoid those impacts.
1209          */
1210         if (val > WQ_UNBOUND_MAX_ACTIVE) {
1211                 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1212                        sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1213                 return -ERANGE;
1214         }
1215
1216         spin_lock(&sbi->ll_lock);
1217         sbi->ll_ra_info.ra_async_max_active = val;
1218         spin_unlock(&sbi->ll_lock);
1219
1220         return count;
1221 }
1222 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1223
1224 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1225                                                        struct attribute *attr,
1226                                                        char *buf)
1227 {
1228         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1229                                               ll_kset.kobj);
1230
1231         return scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1232                          sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1233 }
1234
1235 static ssize_t
1236 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1237                                          struct attribute *attr,
1238                                          const char *buffer, size_t count)
1239 {
1240         unsigned long pages_number;
1241         unsigned long max_ra_per_file;
1242         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1243                                               ll_kset.kobj);
1244         int rc;
1245
1246         rc = kstrtoul(buffer, 10, &pages_number);
1247         if (rc)
1248                 return rc;
1249
1250         pages_number = MiB_TO_PAGES(pages_number);
1251         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1252         if (pages_number < 0 || pages_number > max_ra_per_file) {
1253                 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1254                        "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1255                        PAGES_TO_MiB(pages_number),
1256                        PAGES_TO_MiB(max_ra_per_file));
1257                 return -ERANGE;
1258         }
1259         sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1260
1261         return count;
1262 }
1263 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1264
1265 static ssize_t read_ahead_range_kb_show(struct kobject *kobj,
1266                                         struct attribute *attr,char *buf)
1267 {
1268         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1269                                               ll_kset.kobj);
1270
1271         return snprintf(buf, PAGE_SIZE, "%lu\n",
1272                         sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1273 }
1274
1275 static ssize_t
1276 read_ahead_range_kb_store(struct kobject *kobj,
1277                                struct attribute *attr,
1278                                const char *buffer, size_t count)
1279 {
1280         unsigned long pages_number;
1281         unsigned long max_ra_per_file;
1282         u64 val;
1283         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1284                                               ll_kset.kobj);
1285         int rc;
1286
1287         rc = sysfs_memparse(buffer, count, &val, "KiB");
1288         if (rc < 0)
1289                 return rc;
1290
1291         pages_number = val >> PAGE_SHIFT;
1292         /* Disable mmap range read */
1293         if (pages_number == 0)
1294                 goto out;
1295
1296         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1297         if (pages_number > max_ra_per_file ||
1298             pages_number < RA_MIN_MMAP_RANGE_PAGES)
1299                 return -ERANGE;
1300
1301 out:
1302         spin_lock(&sbi->ll_lock);
1303         sbi->ll_ra_info.ra_range_pages = pages_number;
1304         spin_unlock(&sbi->ll_lock);
1305
1306         return count;
1307 }
1308 LUSTRE_RW_ATTR(read_ahead_range_kb);
1309
1310 static ssize_t fast_read_show(struct kobject *kobj,
1311                               struct attribute *attr,
1312                               char *buf)
1313 {
1314         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1315                                               ll_kset.kobj);
1316
1317         return scnprintf(buf, PAGE_SIZE, "%u\n",
1318                          test_bit(LL_SBI_FAST_READ, sbi->ll_flags));
1319 }
1320
1321 static ssize_t fast_read_store(struct kobject *kobj,
1322                                struct attribute *attr,
1323                                const char *buffer,
1324                                size_t count)
1325 {
1326         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1327                                               ll_kset.kobj);
1328         bool val;
1329         int rc;
1330
1331         rc = kstrtobool(buffer, &val);
1332         if (rc)
1333                 return rc;
1334
1335         spin_lock(&sbi->ll_lock);
1336         if (val)
1337                 set_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1338         else
1339                 clear_bit(LL_SBI_FAST_READ, sbi->ll_flags);
1340         spin_unlock(&sbi->ll_lock);
1341
1342         return count;
1343 }
1344 LUSTRE_RW_ATTR(fast_read);
1345
1346 static ssize_t file_heat_show(struct kobject *kobj,
1347                               struct attribute *attr,
1348                               char *buf)
1349 {
1350         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1351                                               ll_kset.kobj);
1352
1353         return scnprintf(buf, PAGE_SIZE, "%u\n",
1354                          test_bit(LL_SBI_FILE_HEAT, sbi->ll_flags));
1355 }
1356
1357 static ssize_t file_heat_store(struct kobject *kobj,
1358                                struct attribute *attr,
1359                                const char *buffer,
1360                                size_t count)
1361 {
1362         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1363                                               ll_kset.kobj);
1364         bool val;
1365         int rc;
1366
1367         rc = kstrtobool(buffer, &val);
1368         if (rc)
1369                 return rc;
1370
1371         spin_lock(&sbi->ll_lock);
1372         if (val)
1373                 set_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1374         else
1375                 clear_bit(LL_SBI_FILE_HEAT, sbi->ll_flags);
1376         spin_unlock(&sbi->ll_lock);
1377
1378         return count;
1379 }
1380 LUSTRE_RW_ATTR(file_heat);
1381
1382 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1383                                           struct attribute *attr,
1384                                           char *buf)
1385 {
1386         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1387                                               ll_kset.kobj);
1388
1389         return scnprintf(buf, PAGE_SIZE, "%u\n",
1390                          (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1391 }
1392
1393 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1394                                            struct attribute *attr,
1395                                            const char *buffer,
1396                                            size_t count)
1397 {
1398         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1399                                               ll_kset.kobj);
1400         unsigned long val;
1401         int rc;
1402
1403         rc = kstrtoul(buffer, 10, &val);
1404         if (rc)
1405                 return rc;
1406
1407         if (val < 0 || val > 100)
1408                 return -ERANGE;
1409
1410         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1411
1412         return count;
1413 }
1414 LUSTRE_RW_ATTR(heat_decay_percentage);
1415
1416 static ssize_t heat_period_second_show(struct kobject *kobj,
1417                                        struct attribute *attr,
1418                                        char *buf)
1419 {
1420         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1421                                               ll_kset.kobj);
1422
1423         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1424 }
1425
1426 static ssize_t heat_period_second_store(struct kobject *kobj,
1427                                         struct attribute *attr,
1428                                         const char *buffer,
1429                                         size_t count)
1430 {
1431         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1432                                               ll_kset.kobj);
1433         unsigned long val;
1434         int rc;
1435
1436         rc = kstrtoul(buffer, 10, &val);
1437         if (rc)
1438                 return rc;
1439
1440         if (val <= 0)
1441                 return -ERANGE;
1442
1443         sbi->ll_heat_period_second = val;
1444
1445         return count;
1446 }
1447 LUSTRE_RW_ATTR(heat_period_second);
1448
1449 static ssize_t opencache_threshold_count_show(struct kobject *kobj,
1450                                               struct attribute *attr,
1451                                               char *buf)
1452 {
1453         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1454                                               ll_kset.kobj);
1455
1456         if (sbi->ll_oc_thrsh_count)
1457                 return snprintf(buf, PAGE_SIZE, "%u\n",
1458                                 sbi->ll_oc_thrsh_count);
1459         else
1460                 return snprintf(buf, PAGE_SIZE, "off\n");
1461 }
1462
1463 static ssize_t opencache_threshold_count_store(struct kobject *kobj,
1464                                                struct attribute *attr,
1465                                                const char *buffer,
1466                                                size_t count)
1467 {
1468         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1469                                               ll_kset.kobj);
1470         unsigned int val;
1471         int rc;
1472
1473         rc = kstrtouint(buffer, 10, &val);
1474         if (rc) {
1475                 bool enable;
1476                 /* also accept "off" to disable and "on" to always cache */
1477                 rc = kstrtobool(buffer, &enable);
1478                 if (rc)
1479                         return rc;
1480                 val = enable;
1481         }
1482         sbi->ll_oc_thrsh_count = val;
1483
1484         return count;
1485 }
1486 LUSTRE_RW_ATTR(opencache_threshold_count);
1487
1488 static ssize_t opencache_threshold_ms_show(struct kobject *kobj,
1489                                            struct attribute *attr,
1490                                            char *buf)
1491 {
1492         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1493                                               ll_kset.kobj);
1494
1495         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_thrsh_ms);
1496 }
1497
1498 static ssize_t opencache_threshold_ms_store(struct kobject *kobj,
1499                                             struct attribute *attr,
1500                                             const char *buffer,
1501                                             size_t count)
1502 {
1503         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1504                                               ll_kset.kobj);
1505         unsigned int val;
1506         int rc;
1507
1508         rc = kstrtouint(buffer, 10, &val);
1509         if (rc)
1510                 return rc;
1511
1512         sbi->ll_oc_thrsh_ms = val;
1513
1514         return count;
1515 }
1516 LUSTRE_RW_ATTR(opencache_threshold_ms);
1517
1518 static ssize_t opencache_max_ms_show(struct kobject *kobj,
1519                                      struct attribute *attr,
1520                                      char *buf)
1521 {
1522         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1523                                               ll_kset.kobj);
1524
1525         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_oc_max_ms);
1526 }
1527
1528 static ssize_t opencache_max_ms_store(struct kobject *kobj,
1529                                       struct attribute *attr,
1530                                       const char *buffer,
1531                                       size_t count)
1532 {
1533         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1534                                               ll_kset.kobj);
1535         unsigned int val;
1536         int rc;
1537
1538         rc = kstrtouint(buffer, 10, &val);
1539         if (rc)
1540                 return rc;
1541
1542         sbi->ll_oc_max_ms = val;
1543
1544         return count;
1545 }
1546 LUSTRE_RW_ATTR(opencache_max_ms);
1547
1548 static ssize_t inode_cache_show(struct kobject *kobj,
1549                                 struct attribute *attr,
1550                                 char *buf)
1551 {
1552         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1553                                               ll_kset.kobj);
1554
1555         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_inode_cache_enabled);
1556 }
1557
1558 static ssize_t inode_cache_store(struct kobject *kobj,
1559                                  struct attribute *attr,
1560                                  const char *buffer,
1561                                  size_t count)
1562 {
1563         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1564                                               ll_kset.kobj);
1565         bool val;
1566         int rc;
1567
1568         rc = kstrtobool(buffer, &val);
1569         if (rc)
1570                 return rc;
1571
1572         sbi->ll_inode_cache_enabled = val;
1573
1574         return count;
1575 }
1576 LUSTRE_RW_ATTR(inode_cache);
1577
1578 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1579 {
1580         struct super_block      *sb    = m->private;
1581         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1582         struct cl_client_cache  *cache = sbi->ll_cache;
1583         long pages;
1584         int mb;
1585
1586         pages = atomic_long_read(&cache->ccc_unstable_nr);
1587         mb    = (pages * PAGE_SIZE) >> 20;
1588
1589         seq_printf(m, "unstable_check:     %8d\n"
1590                       "unstable_pages: %12ld\n"
1591                       "unstable_mb:        %8d\n",
1592                    cache->ccc_unstable_check, pages, mb);
1593         return 0;
1594 }
1595
1596 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1597                                            const char __user *buffer,
1598                                            size_t count, loff_t *unused)
1599 {
1600         struct seq_file *seq = file->private_data;
1601         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1602         char kernbuf[128];
1603         bool val;
1604         int rc;
1605
1606         if (count == 0)
1607                 return 0;
1608         if (count >= sizeof(kernbuf))
1609                 return -EINVAL;
1610
1611         if (copy_from_user(kernbuf, buffer, count))
1612                 return -EFAULT;
1613         kernbuf[count] = 0;
1614
1615         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1616                   kernbuf;
1617         rc = kstrtobool_from_user(buffer, count, &val);
1618         if (rc < 0)
1619                 return rc;
1620
1621         /* borrow lru lock to set the value */
1622         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1623         sbi->ll_cache->ccc_unstable_check = val;
1624         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1625
1626         return count;
1627 }
1628
1629 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1630
1631 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1632 {
1633         struct super_block *sb = m->private;
1634         struct ll_sb_info *sbi = ll_s2sbi(sb);
1635         struct root_squash_info *squash = &sbi->ll_squash;
1636
1637         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1638         return 0;
1639 }
1640
1641 static ssize_t ll_root_squash_seq_write(struct file *file,
1642                                         const char __user *buffer,
1643                                         size_t count, loff_t *off)
1644 {
1645         struct seq_file *m = file->private_data;
1646         struct super_block *sb = m->private;
1647         struct ll_sb_info *sbi = ll_s2sbi(sb);
1648         struct root_squash_info *squash = &sbi->ll_squash;
1649
1650         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1651 }
1652
1653 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1654
1655 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1656 {
1657         struct super_block *sb = m->private;
1658         struct ll_sb_info *sbi = ll_s2sbi(sb);
1659         struct root_squash_info *squash = &sbi->ll_squash;
1660         int len;
1661
1662         spin_lock(&squash->rsi_lock);
1663         if (!list_empty(&squash->rsi_nosquash_nids)) {
1664                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1665                                         &squash->rsi_nosquash_nids);
1666                 m->count += len;
1667                 seq_putc(m, '\n');
1668         } else {
1669                 seq_puts(m, "NONE\n");
1670         }
1671         spin_unlock(&squash->rsi_lock);
1672
1673         return 0;
1674 }
1675
1676 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1677                                           const char __user *buffer,
1678                                           size_t count, loff_t *off)
1679 {
1680         struct seq_file *m = file->private_data;
1681         struct super_block *sb = m->private;
1682         struct ll_sb_info *sbi = ll_s2sbi(sb);
1683         struct root_squash_info *squash = &sbi->ll_squash;
1684         int rc;
1685
1686         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1687         if (rc < 0)
1688                 return rc;
1689
1690         ll_compute_rootsquash_state(sbi);
1691
1692         return rc;
1693 }
1694
1695 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1696
1697 #if defined(CONFIG_LL_ENCRYPTION)
1698 static ssize_t enable_filename_encryption_show(struct kobject *kobj,
1699                                                struct attribute *attr,
1700                                                char *buffer)
1701 {
1702         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1703                                               ll_kset.kobj);
1704         struct lustre_sb_info *lsi = sbi->lsi;
1705
1706         return snprintf(buffer, PAGE_SIZE,  "%u\n",
1707                         lsi->lsi_flags & LSI_FILENAME_ENC ? 1 : 0);
1708 }
1709
1710 static ssize_t enable_filename_encryption_store(struct kobject *kobj,
1711                                                 struct attribute *attr,
1712                                                 const char *buffer,
1713                                                 size_t count)
1714 {
1715         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1716                                               ll_kset.kobj);
1717         struct lustre_sb_info *lsi = sbi->lsi;
1718         bool val;
1719         int rc;
1720
1721         rc = kstrtobool(buffer, &val);
1722         if (rc)
1723                 return rc;
1724
1725         if (val) {
1726                 if (!ll_sbi_has_name_encrypt(sbi)) {
1727                         /* server does not support name encryption,
1728                          * so force it to NULL on client
1729                          */
1730                         CDEBUG(D_SEC, "%s: server does not support name encryption\n",
1731                                sbi->ll_fsname);
1732                         lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1733                         return -EOPNOTSUPP;
1734                 }
1735
1736                 lsi->lsi_flags |= LSI_FILENAME_ENC;
1737         } else {
1738                 lsi->lsi_flags &= ~LSI_FILENAME_ENC;
1739         }
1740
1741         return count;
1742 }
1743
1744 LUSTRE_RW_ATTR(enable_filename_encryption);
1745 #endif /* CONFIG_LL_ENCRYPTION */
1746
1747 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
1748 static ssize_t filename_enc_use_old_base64_show(struct kobject *kobj,
1749                                                 struct attribute *attr,
1750                                                 char *buffer)
1751 {
1752         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1753                                               ll_kset.kobj);
1754         struct lustre_sb_info *lsi = sbi->lsi;
1755
1756         return snprintf(buffer, PAGE_SIZE, "%u\n",
1757                         lsi->lsi_flags & LSI_FILENAME_ENC_B64_OLD_CLI ? 1 : 0);
1758 }
1759
1760 static ssize_t filename_enc_use_old_base64_store(struct kobject *kobj,
1761                                                  struct attribute *attr,
1762                                                  const char *buffer,
1763                                                  size_t count)
1764 {
1765         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1766                                               ll_kset.kobj);
1767         struct lustre_sb_info *lsi = sbi->lsi;
1768         bool val;
1769         int rc;
1770
1771         rc = kstrtobool(buffer, &val);
1772         if (rc)
1773                 return rc;
1774
1775         if (val) {
1776                 if (!ll_sbi_has_name_encrypt(sbi)) {
1777                         /* server does not support name encryption,
1778                          * so force it to NULL on client
1779                          */
1780                         CDEBUG(D_SEC,
1781                                "%s: server does not support name encryption\n",
1782                                sbi->ll_fsname);
1783                         lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1784                         return -EOPNOTSUPP;
1785                 }
1786
1787                 lsi->lsi_flags |= LSI_FILENAME_ENC_B64_OLD_CLI;
1788         } else {
1789                 lsi->lsi_flags &= ~LSI_FILENAME_ENC_B64_OLD_CLI;
1790         }
1791
1792         return count;
1793 }
1794
1795 LUSTRE_RW_ATTR(filename_enc_use_old_base64);
1796 #endif /* CONFIG_LL_ENCRYPTION || HAVE_LUSTRE_CRYPTO */
1797
1798 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1799 {
1800         struct super_block *sb = m->private;
1801         struct ll_sb_info *sbi = ll_s2sbi(sb);
1802
1803         return pcc_super_dump(&sbi->ll_pcc_super, m);
1804 }
1805
1806 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1807                                 size_t count, loff_t *off)
1808 {
1809         struct seq_file *m = file->private_data;
1810         struct super_block *sb = m->private;
1811         struct ll_sb_info *sbi = ll_s2sbi(sb);
1812         int rc;
1813         char *kernbuf;
1814
1815         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1816                 return -EINVAL;
1817
1818         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1819                 return -EOPNOTSUPP;
1820
1821         OBD_ALLOC(kernbuf, count + 1);
1822         if (kernbuf == NULL)
1823                 return -ENOMEM;
1824
1825         if (copy_from_user(kernbuf, buffer, count))
1826                 GOTO(out_free_kernbuff, rc = -EFAULT);
1827
1828         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1829 out_free_kernbuff:
1830         OBD_FREE(kernbuf, count + 1);
1831         return rc ? rc : count;
1832 }
1833 LDEBUGFS_SEQ_FOPS(ll_pcc);
1834
1835 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1836         { .name =       "site",
1837           .fops =       &ll_site_stats_fops                     },
1838         { .name =       "max_cached_mb",
1839           .fops =       &ll_max_cached_mb_fops                  },
1840         { .name =       "statahead_stats",
1841           .fops =       &ll_statahead_stats_fops                },
1842         { .name =       "unstable_stats",
1843           .fops =       &ll_unstable_stats_fops                 },
1844         { .name =       "sbi_flags",
1845           .fops =       &ll_sbi_flags_fops                      },
1846         { .name =       "root_squash",
1847           .fops =       &ll_root_squash_fops                    },
1848         { .name =       "nosquash_nids",
1849           .fops =       &ll_nosquash_nids_fops                  },
1850         { .name =       "pcc",
1851           .fops =       &ll_pcc_fops,                           },
1852         { NULL }
1853 };
1854
1855 #define MAX_STRING_SIZE 128
1856
1857 static struct attribute *llite_attrs[] = {
1858         &lustre_attr_blocksize.attr,
1859         &lustre_attr_stat_blocksize.attr,
1860         &lustre_attr_kbytestotal.attr,
1861         &lustre_attr_kbytesfree.attr,
1862         &lustre_attr_kbytesavail.attr,
1863         &lustre_attr_filestotal.attr,
1864         &lustre_attr_filesfree.attr,
1865         &lustre_attr_client_type.attr,
1866         &lustre_attr_foreign_symlink_enable.attr,
1867         &lustre_attr_foreign_symlink_prefix.attr,
1868         &lustre_attr_foreign_symlink_upcall.attr,
1869         &lustre_attr_foreign_symlink_upcall_info.attr,
1870         &lustre_attr_fstype.attr,
1871         &lustre_attr_uuid.attr,
1872         &lustre_attr_checksums.attr,
1873         &lustre_attr_checksum_pages.attr,
1874         &lustre_attr_max_read_ahead_mb.attr,
1875         &lustre_attr_max_read_ahead_per_file_mb.attr,
1876         &lustre_attr_max_read_ahead_whole_mb.attr,
1877         &lustre_attr_max_read_ahead_async_active.attr,
1878         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1879         &lustre_attr_read_ahead_range_kb.attr,
1880         &lustre_attr_stats_track_pid.attr,
1881         &lustre_attr_stats_track_ppid.attr,
1882         &lustre_attr_stats_track_gid.attr,
1883         &lustre_attr_statahead_running_max.attr,
1884         &lustre_attr_statahead_batch_max.attr,
1885         &lustre_attr_statahead_max.attr,
1886         &lustre_attr_statahead_agl.attr,
1887         &lustre_attr_lazystatfs.attr,
1888         &lustre_attr_statfs_max_age.attr,
1889         &lustre_attr_max_easize.attr,
1890         &lustre_attr_default_easize.attr,
1891         &lustre_attr_xattr_cache.attr,
1892         &lustre_attr_fast_read.attr,
1893         &lustre_attr_tiny_write.attr,
1894         &lustre_attr_parallel_dio.attr,
1895         &lustre_attr_file_heat.attr,
1896         &lustre_attr_heat_decay_percentage.attr,
1897         &lustre_attr_heat_period_second.attr,
1898         &lustre_attr_opencache_threshold_count.attr,
1899         &lustre_attr_opencache_threshold_ms.attr,
1900         &lustre_attr_opencache_max_ms.attr,
1901         &lustre_attr_inode_cache.attr,
1902 #ifdef CONFIG_LL_ENCRYPTION
1903         &lustre_attr_enable_filename_encryption.attr,
1904 #endif
1905 #if defined(CONFIG_LL_ENCRYPTION) || defined(HAVE_LUSTRE_CRYPTO)
1906         &lustre_attr_filename_enc_use_old_base64.attr,
1907 #endif
1908         NULL,
1909 };
1910
1911 KOBJ_ATTRIBUTE_GROUPS(llite); /* creates llite_groups */
1912
1913 static void sbi_kobj_release(struct kobject *kobj)
1914 {
1915         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1916                                               ll_kset.kobj);
1917         complete(&sbi->ll_kobj_unregister);
1918 }
1919
1920 static struct kobj_type sbi_ktype = {
1921         .default_groups = KOBJ_ATTR_GROUPS(llite),
1922         .sysfs_ops      = &lustre_sysfs_ops,
1923         .release        = sbi_kobj_release,
1924 };
1925
1926 static const struct llite_file_opcode {
1927         __u32                           lfo_opcode;
1928         enum lprocfs_counter_config     lfo_config;
1929         const char                      *lfo_opname;
1930 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1931         /* file operation */
1932         { LPROC_LL_READ_BYTES,  LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1933         { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1934         { LPROC_LL_READ,        LPROCFS_TYPE_LATENCY,   "read" },
1935         { LPROC_LL_WRITE,       LPROCFS_TYPE_LATENCY,   "write" },
1936         { LPROC_LL_IOCTL,       LPROCFS_TYPE_REQS,      "ioctl" },
1937         { LPROC_LL_OPEN,        LPROCFS_TYPE_LATENCY,   "open" },
1938         { LPROC_LL_RELEASE,     LPROCFS_TYPE_LATENCY,   "close" },
1939         { LPROC_LL_MMAP,        LPROCFS_TYPE_LATENCY,   "mmap" },
1940         { LPROC_LL_FAULT,       LPROCFS_TYPE_LATENCY,   "page_fault" },
1941         { LPROC_LL_MKWRITE,     LPROCFS_TYPE_LATENCY,   "page_mkwrite" },
1942         { LPROC_LL_LLSEEK,      LPROCFS_TYPE_LATENCY,   "seek" },
1943         { LPROC_LL_FSYNC,       LPROCFS_TYPE_LATENCY,   "fsync" },
1944         { LPROC_LL_READDIR,     LPROCFS_TYPE_LATENCY,   "readdir" },
1945         { LPROC_LL_INODE_OCOUNT, LPROCFS_TYPE_REQS | LPROCFS_CNTR_AVGMINMAX |
1946                                 LPROCFS_CNTR_STDDEV,    "opencount" },
1947         { LPROC_LL_INODE_OPCLTM,LPROCFS_TYPE_LATENCY,   "openclosetime" },
1948         /* inode operation */
1949         { LPROC_LL_SETATTR,     LPROCFS_TYPE_LATENCY,   "setattr" },
1950         { LPROC_LL_TRUNC,       LPROCFS_TYPE_LATENCY,   "truncate" },
1951         { LPROC_LL_FLOCK,       LPROCFS_TYPE_LATENCY,   "flock" },
1952         { LPROC_LL_GETATTR,     LPROCFS_TYPE_LATENCY,   "getattr" },
1953         { LPROC_LL_FALLOCATE,   LPROCFS_TYPE_LATENCY,   "fallocate"},
1954         /* dir inode operation */
1955         { LPROC_LL_CREATE,      LPROCFS_TYPE_LATENCY,   "create" },
1956         { LPROC_LL_LINK,        LPROCFS_TYPE_LATENCY,   "link" },
1957         { LPROC_LL_UNLINK,      LPROCFS_TYPE_LATENCY,   "unlink" },
1958         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_LATENCY,   "symlink" },
1959         { LPROC_LL_MKDIR,       LPROCFS_TYPE_LATENCY,   "mkdir" },
1960         { LPROC_LL_RMDIR,       LPROCFS_TYPE_LATENCY,   "rmdir" },
1961         { LPROC_LL_MKNOD,       LPROCFS_TYPE_LATENCY,   "mknod" },
1962         { LPROC_LL_RENAME,      LPROCFS_TYPE_LATENCY,   "rename" },
1963         /* special inode operation */
1964         { LPROC_LL_STATFS,      LPROCFS_TYPE_LATENCY,   "statfs" },
1965         { LPROC_LL_SETXATTR,    LPROCFS_TYPE_LATENCY,   "setxattr" },
1966         { LPROC_LL_GETXATTR,    LPROCFS_TYPE_LATENCY,   "getxattr" },
1967         { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS,    "getxattr_hits" },
1968         { LPROC_LL_LISTXATTR,   LPROCFS_TYPE_LATENCY,   "listxattr" },
1969         { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY,   "removexattr" },
1970         { LPROC_LL_INODE_PERM,  LPROCFS_TYPE_LATENCY,   "inode_permission" },
1971 };
1972
1973 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1974 {
1975         if (!sbi->ll_stats)
1976                 return;
1977
1978         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1979                 lprocfs_counter_add(sbi->ll_stats, op, count);
1980         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1981                  sbi->ll_stats_track_id == current->pid)
1982                 lprocfs_counter_add(sbi->ll_stats, op, count);
1983         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1984                  sbi->ll_stats_track_id == current->real_parent->pid)
1985                 lprocfs_counter_add(sbi->ll_stats, op, count);
1986         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1987                  sbi->ll_stats_track_id ==
1988                         from_kgid(&init_user_ns, current_gid()))
1989                 lprocfs_counter_add(sbi->ll_stats, op, count);
1990 }
1991 EXPORT_SYMBOL(ll_stats_ops_tally);
1992
1993 static const char *const ra_stat_string[] = {
1994         [RA_STAT_HIT]                   = "hits",
1995         [RA_STAT_MISS]                  = "misses",
1996         [RA_STAT_DISTANT_READPAGE]      = "readpage_not_consecutive",
1997         [RA_STAT_MISS_IN_WINDOW]        = "miss_inside_window",
1998         [RA_STAT_FAILED_GRAB_PAGE]      = "failed_grab_cache_page",
1999         [RA_STAT_FAILED_MATCH]          = "failed_lock_match",
2000         [RA_STAT_DISCARDED]             = "read_but_discarded",
2001         [RA_STAT_ZERO_LEN]              = "zero_length_file",
2002         [RA_STAT_ZERO_WINDOW]           = "zero_size_window",
2003         [RA_STAT_EOF]                   = "readahead_to_eof",
2004         [RA_STAT_MAX_IN_FLIGHT]         = "hit_max_readahead_issue",
2005         [RA_STAT_WRONG_GRAB_PAGE]       = "wrong_page_from_grab_cache_page",
2006         [RA_STAT_FAILED_REACH_END]      = "failed_to_reach_end",
2007         [RA_STAT_ASYNC]                 = "async_readahead",
2008         [RA_STAT_FAILED_FAST_READ]      = "failed_to_fast_read",
2009         [RA_STAT_MMAP_RANGE_READ]       = "mmap_range_read",
2010         [RA_STAT_READAHEAD_PAGES]       = "readahead_pages"
2011 };
2012
2013 int ll_debugfs_register_super(struct super_block *sb, const char *name)
2014 {
2015         struct lustre_sb_info *lsi = s2lsi(sb);
2016         struct ll_sb_info *sbi = ll_s2sbi(sb);
2017         int err, id;
2018
2019         ENTRY;
2020         LASSERT(sbi);
2021
2022         if (IS_ERR_OR_NULL(llite_root))
2023                 goto out_ll_kset;
2024
2025         sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
2026         ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
2027
2028         debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
2029                             &vvp_dump_pgcache_file_ops);
2030
2031         debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
2032                                  &ll_rw_extents_stats_fops);
2033
2034         debugfs_create_file("extents_stats_per_process", 0644,
2035                             sbi->ll_debugfs_entry, sbi,
2036                             &ll_rw_extents_stats_pp_fops);
2037
2038         debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
2039                             &ll_rw_offset_stats_fops);
2040
2041         /* File operations stats */
2042         sbi->ll_stats = lprocfs_stats_alloc(LPROC_LL_FILE_OPCODES,
2043                                             LPROCFS_STATS_FLAG_NONE);
2044         if (sbi->ll_stats == NULL)
2045                 GOTO(out_debugfs, err = -ENOMEM);
2046
2047         /* do counter init */
2048         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++)
2049                 lprocfs_counter_init(sbi->ll_stats,
2050                                      llite_opcode_table[id].lfo_opcode,
2051                                      llite_opcode_table[id].lfo_config,
2052                                      llite_opcode_table[id].lfo_opname);
2053
2054         debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
2055                             sbi->ll_stats, &ldebugfs_stats_seq_fops);
2056
2057         sbi->ll_ra_stats = lprocfs_stats_alloc(ARRAY_SIZE(ra_stat_string),
2058                                                LPROCFS_STATS_FLAG_NONE);
2059         if (sbi->ll_ra_stats == NULL)
2060                 GOTO(out_stats, err = -ENOMEM);
2061
2062         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++) {
2063                 if (id == RA_STAT_READAHEAD_PAGES)
2064                         lprocfs_counter_init(sbi->ll_ra_stats, id,
2065                                              LPROCFS_TYPE_PAGES |
2066                                              LPROCFS_CNTR_AVGMINMAX,
2067                                              ra_stat_string[id]);
2068                 else
2069                         lprocfs_counter_init(sbi->ll_ra_stats, id,
2070                                              LPROCFS_TYPE_PAGES,
2071                                              ra_stat_string[id]);
2072         }
2073
2074         debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
2075                             sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
2076
2077 out_ll_kset:
2078         /* Yes we also register sysfs mount kset here as well */
2079         sbi->ll_kset.kobj.parent = llite_kobj;
2080         sbi->ll_kset.kobj.ktype = &sbi_ktype;
2081         init_completion(&sbi->ll_kobj_unregister);
2082         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
2083         if (err)
2084                 GOTO(out_ra_stats, err);
2085
2086         err = kset_register(&sbi->ll_kset);
2087         if (err)
2088                 GOTO(out_ra_stats, err);
2089
2090         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
2091
2092         RETURN(0);
2093 out_ra_stats:
2094         lprocfs_stats_free(&sbi->ll_ra_stats);
2095 out_stats:
2096         lprocfs_stats_free(&sbi->ll_stats);
2097 out_debugfs:
2098         debugfs_remove_recursive(sbi->ll_debugfs_entry);
2099
2100         RETURN(err);
2101 }
2102
2103 void ll_debugfs_unregister_super(struct super_block *sb)
2104 {
2105         struct lustre_sb_info *lsi = s2lsi(sb);
2106         struct ll_sb_info *sbi = ll_s2sbi(sb);
2107
2108         debugfs_remove_recursive(sbi->ll_debugfs_entry);
2109
2110         if (sbi->ll_dt_obd)
2111                 sysfs_remove_link(&sbi->ll_kset.kobj,
2112                                   sbi->ll_dt_obd->obd_type->typ_name);
2113
2114         if (sbi->ll_md_obd)
2115                 sysfs_remove_link(&sbi->ll_kset.kobj,
2116                                   sbi->ll_md_obd->obd_type->typ_name);
2117
2118         kobject_put(lsi->lsi_kobj);
2119
2120         kset_unregister(&sbi->ll_kset);
2121         wait_for_completion(&sbi->ll_kobj_unregister);
2122
2123         lprocfs_stats_free(&sbi->ll_ra_stats);
2124         lprocfs_stats_free(&sbi->ll_stats);
2125 }
2126 #undef MAX_STRING_SIZE
2127
2128 static void ll_display_extents_info(struct ll_rw_extents_info *rw_extents,
2129                                     struct seq_file *seq, int which)
2130 {
2131         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
2132         unsigned long start, end, r, w;
2133         char *unitp = "KMGTPEZY";
2134         int i, units = 10;
2135         struct per_process_info *pp_info;
2136
2137         pp_info = &rw_extents->pp_extents[which];
2138         read_cum = 0;
2139         write_cum = 0;
2140         start = 0;
2141
2142         for (i = 0; i < LL_HIST_MAX; i++) {
2143                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
2144                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
2145         }
2146
2147         for (i = 0; i < LL_HIST_MAX; i++) {
2148                 r = pp_info->pp_r_hist.oh_buckets[i];
2149                 w = pp_info->pp_w_hist.oh_buckets[i];
2150                 read_cum += r;
2151                 write_cum += w;
2152                 end = 1 << (i + LL_HIST_START - units);
2153                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
2154                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
2155                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
2156                            r, pct(r, read_tot), pct(read_cum, read_tot),
2157                            w, pct(w, write_tot), pct(write_cum, write_tot));
2158                 start = end;
2159                 if (start == (1 << 10)) {
2160                         start = 1;
2161                         units += 10;
2162                         unitp++;
2163                 }
2164                 if (read_cum == read_tot && write_cum == write_tot)
2165                         break;
2166         }
2167 }
2168
2169 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
2170 {
2171         struct ll_sb_info *sbi = seq->private;
2172         struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2173         int k;
2174
2175         if (!sbi->ll_rw_stats_on || !rw_extents) {
2176                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2177                 return 0;
2178         }
2179
2180         spin_lock(&sbi->ll_pp_extent_lock);
2181         lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2182                              ":", true, "");
2183         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
2184         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
2185                    "extents", "calls", "%", "cum%", "calls", "%", "cum%");
2186
2187         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
2188                 if (rw_extents->pp_extents[k].pid != 0) {
2189                         seq_printf(seq, "\nPID: %d\n",
2190                                    rw_extents->pp_extents[k].pid);
2191                         ll_display_extents_info(rw_extents, seq, k);
2192                 }
2193         }
2194         spin_unlock(&sbi->ll_pp_extent_lock);
2195         return 0;
2196 }
2197
2198 static int alloc_rw_stats_info(struct ll_sb_info *sbi)
2199 {
2200         struct ll_rw_extents_info *rw_extents;
2201         struct ll_rw_process_info *offset;
2202         struct ll_rw_process_info *process;
2203         int i, rc = 0;
2204
2205         OBD_ALLOC(rw_extents, sizeof(*rw_extents));
2206         if (!rw_extents)
2207                 return -ENOMEM;
2208
2209         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2210                 spin_lock_init(&rw_extents->pp_extents[i].pp_r_hist.oh_lock);
2211                 spin_lock_init(&rw_extents->pp_extents[i].pp_w_hist.oh_lock);
2212         }
2213         rw_extents->pp_init = ktime_get_real();
2214
2215         spin_lock(&sbi->ll_pp_extent_lock);
2216         if (!sbi->ll_rw_extents_info)
2217                 sbi->ll_rw_extents_info = rw_extents;
2218         spin_unlock(&sbi->ll_pp_extent_lock);
2219         /* another writer allocated the struct before we got the lock */
2220         if (sbi->ll_rw_extents_info != rw_extents)
2221                 OBD_FREE(rw_extents, sizeof(*rw_extents));
2222
2223         OBD_ALLOC(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2224         if (!process)
2225                 GOTO(out, rc = -ENOMEM);
2226         OBD_ALLOC(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2227         if (!offset)
2228                 GOTO(out_free, rc = -ENOMEM);
2229
2230         spin_lock(&sbi->ll_process_lock);
2231         if (!sbi->ll_rw_process_info)
2232                 sbi->ll_rw_process_info = process;
2233         if (!sbi->ll_rw_offset_info)
2234                 sbi->ll_rw_offset_info = offset;
2235         spin_unlock(&sbi->ll_process_lock);
2236         sbi->ll_process_stats_init = ktime_get_real();
2237
2238         /* another writer allocated the structs before we got the lock */
2239         if (sbi->ll_rw_offset_info != offset)
2240                 OBD_FREE(offset, sizeof(*offset) * LL_OFFSET_HIST_MAX);
2241         if (sbi->ll_rw_process_info != process) {
2242 out_free:
2243                 OBD_FREE(process, sizeof(*process) * LL_PROCESS_HIST_MAX);
2244         }
2245
2246 out:
2247         return rc;
2248 }
2249
2250 void ll_free_rw_stats_info(struct ll_sb_info *sbi)
2251 {
2252         if (sbi->ll_rw_extents_info) {
2253                 OBD_FREE(sbi->ll_rw_extents_info,
2254                          sizeof(*sbi->ll_rw_extents_info));
2255                 sbi->ll_rw_extents_info = NULL;
2256         }
2257         if (sbi->ll_rw_offset_info) {
2258                 OBD_FREE(sbi->ll_rw_offset_info,
2259                          sizeof(*sbi->ll_rw_offset_info) * LL_OFFSET_HIST_MAX);
2260                 sbi->ll_rw_offset_info = NULL;
2261         }
2262         if (sbi->ll_rw_process_info) {
2263                 OBD_FREE(sbi->ll_rw_process_info,
2264                         sizeof(*sbi->ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2265                 sbi->ll_rw_process_info = NULL;
2266         }
2267 }
2268
2269 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
2270                                                 const char __user *buf,
2271                                                 size_t len, loff_t *off)
2272 {
2273         struct seq_file *seq = file->private_data;
2274         struct ll_sb_info *sbi = seq->private;
2275         struct ll_rw_extents_info *rw_extents;
2276         int i;
2277         __s64 value;
2278
2279         if (len == 0)
2280                 return -EINVAL;
2281
2282         value = ll_stats_pid_write(buf, len);
2283
2284         if (value == 0) {
2285                 sbi->ll_rw_stats_on = 0;
2286         } else {
2287                 if (!sbi->ll_rw_extents_info) {
2288                         int rc = alloc_rw_stats_info(sbi);
2289
2290                         if (rc)
2291                                 return rc;
2292                 }
2293                 sbi->ll_rw_stats_on = 1;
2294         }
2295
2296
2297         spin_lock(&sbi->ll_pp_extent_lock);
2298         rw_extents = sbi->ll_rw_extents_info;
2299         if (rw_extents) {
2300                 rw_extents->pp_init = ktime_get_real();
2301                 for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2302                         rw_extents->pp_extents[i].pid = 0;
2303                         lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2304                         lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2305                 }
2306         }
2307         spin_unlock(&sbi->ll_pp_extent_lock);
2308
2309         return len;
2310 }
2311
2312 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
2313
2314 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
2315 {
2316         struct ll_sb_info *sbi = seq->private;
2317         struct ll_rw_extents_info *rw_extents = sbi->ll_rw_extents_info;
2318
2319         if (!sbi->ll_rw_stats_on || !rw_extents) {
2320                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2321                 return 0;
2322         }
2323
2324         spin_lock(&sbi->ll_lock);
2325         lprocfs_stats_header(seq, ktime_get_real(), rw_extents->pp_init, 25,
2326                              ":", true, "");
2327
2328         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
2329         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
2330                    "extents", "calls", "%", "cum%",
2331                    "calls", "%", "cum%");
2332
2333         ll_display_extents_info(rw_extents, seq, LL_PROCESS_HIST_MAX);
2334         spin_unlock(&sbi->ll_lock);
2335
2336         return 0;
2337 }
2338
2339 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
2340                                              const char __user *buf,
2341                                              size_t len, loff_t *off)
2342 {
2343         struct seq_file *seq = file->private_data;
2344         struct ll_sb_info *sbi = seq->private;
2345         struct ll_rw_extents_info *rw_extents;
2346         int i;
2347         __s64 value;
2348
2349         if (len == 0)
2350                 return -EINVAL;
2351
2352         value = ll_stats_pid_write(buf, len);
2353
2354         if (value == 0) {
2355                 sbi->ll_rw_stats_on = 0;
2356         } else {
2357                 if (!sbi->ll_rw_extents_info) {
2358                         int rc = alloc_rw_stats_info(sbi);
2359
2360                         if (rc)
2361                                 return rc;
2362                 }
2363                 sbi->ll_rw_stats_on = 1;
2364         }
2365
2366         spin_lock(&sbi->ll_pp_extent_lock);
2367         rw_extents = sbi->ll_rw_extents_info;
2368         if (rw_extents) {
2369                 rw_extents->pp_init = ktime_get_real();
2370                 for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
2371                         rw_extents->pp_extents[i].pid = 0;
2372                         lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_r_hist);
2373                         lprocfs_oh_clear(&rw_extents->pp_extents[i].pp_w_hist);
2374                 }
2375         }
2376         spin_unlock(&sbi->ll_pp_extent_lock);
2377
2378         return len;
2379 }
2380
2381 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
2382
2383 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
2384                        struct ll_file_data *file, loff_t pos,
2385                        size_t count, int rw)
2386 {
2387         int i, cur = -1;
2388         struct ll_rw_process_info *process;
2389         struct ll_rw_process_info *offset;
2390         int *off_count = &sbi->ll_rw_offset_entry_count;
2391         int *process_count = &sbi->ll_offset_process_count;
2392         struct ll_rw_extents_info *rw_extents;
2393
2394         if (!sbi->ll_rw_stats_on)
2395                 return;
2396
2397         spin_lock(&sbi->ll_pp_extent_lock);
2398         rw_extents = sbi->ll_rw_extents_info;
2399         if (!rw_extents) {
2400                 spin_unlock(&sbi->ll_pp_extent_lock);
2401                 return;
2402         }
2403
2404         /* Extent statistics */
2405         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2406                 if (rw_extents->pp_extents[i].pid == pid) {
2407                         cur = i;
2408                         break;
2409                 }
2410         }
2411
2412         if (cur == -1) {
2413                 /* new process */
2414                 sbi->ll_extent_process_count =
2415                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
2416                 cur = sbi->ll_extent_process_count;
2417                 rw_extents->pp_extents[cur].pid = pid;
2418                 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_r_hist);
2419                 lprocfs_oh_clear(&rw_extents->pp_extents[cur].pp_w_hist);
2420         }
2421
2422         for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
2423              (i < (LL_HIST_MAX - 1)); i++);
2424         if (rw == 0) {
2425                 rw_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
2426                 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
2427         } else {
2428                 rw_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
2429                 rw_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
2430         }
2431         spin_unlock(&sbi->ll_pp_extent_lock);
2432
2433         spin_lock(&sbi->ll_process_lock);
2434         process = sbi->ll_rw_process_info;
2435         offset = sbi->ll_rw_offset_info;
2436         if (!process || !offset)
2437                 goto out_unlock;
2438
2439         /* Offset statistics */
2440         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2441                 if (process[i].rw_pid == pid) {
2442                         if (process[i].rw_last_file != file) {
2443                                 process[i].rw_range_start = pos;
2444                                 process[i].rw_last_file_pos = pos + count;
2445                                 process[i].rw_smallest_extent = count;
2446                                 process[i].rw_largest_extent = count;
2447                                 process[i].rw_offset = 0;
2448                                 process[i].rw_last_file = file;
2449                                 goto out_unlock;
2450                         }
2451                         if (process[i].rw_last_file_pos != pos) {
2452                                 *off_count =
2453                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
2454                                 offset[*off_count].rw_op = process[i].rw_op;
2455                                 offset[*off_count].rw_pid = pid;
2456                                 offset[*off_count].rw_range_start =
2457                                         process[i].rw_range_start;
2458                                 offset[*off_count].rw_range_end =
2459                                         process[i].rw_last_file_pos;
2460                                 offset[*off_count].rw_smallest_extent =
2461                                         process[i].rw_smallest_extent;
2462                                 offset[*off_count].rw_largest_extent =
2463                                         process[i].rw_largest_extent;
2464                                 offset[*off_count].rw_offset =
2465                                         process[i].rw_offset;
2466                                 process[i].rw_op = rw;
2467                                 process[i].rw_range_start = pos;
2468                                 process[i].rw_smallest_extent = count;
2469                                 process[i].rw_largest_extent = count;
2470                                 process[i].rw_offset = pos -
2471                                         process[i].rw_last_file_pos;
2472                         }
2473                         if (process[i].rw_smallest_extent > count)
2474                                 process[i].rw_smallest_extent = count;
2475                         if (process[i].rw_largest_extent < count)
2476                                 process[i].rw_largest_extent = count;
2477                         process[i].rw_last_file_pos = pos + count;
2478                         goto out_unlock;
2479                 }
2480         }
2481         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2482         process[*process_count].rw_pid = pid;
2483         process[*process_count].rw_op = rw;
2484         process[*process_count].rw_range_start = pos;
2485         process[*process_count].rw_last_file_pos = pos + count;
2486         process[*process_count].rw_smallest_extent = count;
2487         process[*process_count].rw_largest_extent = count;
2488         process[*process_count].rw_offset = 0;
2489         process[*process_count].rw_last_file = file;
2490
2491 out_unlock:
2492         spin_unlock(&sbi->ll_process_lock);
2493 }
2494
2495 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2496 {
2497         struct ll_sb_info *sbi = seq->private;
2498         struct ll_rw_process_info *offset;
2499         struct ll_rw_process_info *process;
2500         int i;
2501
2502         if (!sbi->ll_rw_stats_on) {
2503                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2504                 return 0;
2505         }
2506
2507         spin_lock(&sbi->ll_process_lock);
2508         lprocfs_stats_header(seq, ktime_get_real(), sbi->ll_process_stats_init,
2509                              25, ":", true, "");
2510         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2511                    "R/W", "PID", "RANGE START", "RANGE END",
2512                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2513
2514         /* We stored the discontiguous offsets here; print them first */
2515         offset = sbi->ll_rw_offset_info;
2516         for (i = 0; offset && i < LL_OFFSET_HIST_MAX; i++) {
2517                 if (offset[i].rw_pid != 0)
2518                         seq_printf(seq,
2519                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2520                                    offset[i].rw_op == READ ? 'R' : 'W',
2521                                    offset[i].rw_pid,
2522                                    offset[i].rw_range_start,
2523                                    offset[i].rw_range_end,
2524                                    (unsigned long)offset[i].rw_smallest_extent,
2525                                    (unsigned long)offset[i].rw_largest_extent,
2526                                    offset[i].rw_offset);
2527         }
2528
2529         /* Then print the current offsets for each process */
2530         process = sbi->ll_rw_process_info;
2531         for (i = 0; process && i < LL_PROCESS_HIST_MAX; i++) {
2532                 if (process[i].rw_pid != 0)
2533                         seq_printf(seq,
2534                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2535                                    process[i].rw_op == READ ? 'R' : 'W',
2536                                    process[i].rw_pid,
2537                                    process[i].rw_range_start,
2538                                    process[i].rw_last_file_pos,
2539                                    (unsigned long)process[i].rw_smallest_extent,
2540                                    (unsigned long)process[i].rw_largest_extent,
2541                                    process[i].rw_offset);
2542         }
2543         spin_unlock(&sbi->ll_process_lock);
2544
2545         return 0;
2546 }
2547
2548 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2549                                             const char __user *buf,
2550                                             size_t len, loff_t *off)
2551 {
2552         struct seq_file *seq = file->private_data;
2553         struct ll_sb_info *sbi = seq->private;
2554         __s64 value;
2555
2556         if (len == 0)
2557                 return -EINVAL;
2558
2559         value = ll_stats_pid_write(buf, len);
2560
2561         if (value == 0) {
2562                 sbi->ll_rw_stats_on = 0;
2563         } else {
2564                 if (!sbi->ll_rw_process_info || !sbi->ll_rw_offset_info) {
2565                         int rc = alloc_rw_stats_info(sbi);
2566
2567                         if (rc)
2568                                 return rc;
2569                 }
2570                 sbi->ll_rw_stats_on = 1;
2571         }
2572
2573         spin_lock(&sbi->ll_process_lock);
2574         sbi->ll_offset_process_count = 0;
2575         sbi->ll_rw_offset_entry_count = 0;
2576         sbi->ll_process_stats_init = ktime_get_real();
2577         if (sbi->ll_rw_process_info)
2578                 memset(sbi->ll_rw_process_info, 0,
2579                        sizeof(struct ll_rw_process_info) * LL_PROCESS_HIST_MAX);
2580         if (sbi->ll_rw_offset_info)
2581                 memset(sbi->ll_rw_offset_info, 0,
2582                        sizeof(struct ll_rw_process_info) * LL_OFFSET_HIST_MAX);
2583         spin_unlock(&sbi->ll_process_lock);
2584
2585         return len;
2586 }
2587
2588 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);