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