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  * Lustre is a trademark of Sun Microsystems, Inc.
  31  */
  32 #define DEBUG_SUBSYSTEM S_LLITE
  33
  34 #include <linux/version.h>
  35 #include <linux/user_namespace.h>
  36 #include <linux/uidgid.h>
  37
  38 #include <uapi/linux/lustre/lustre_param.h>
  39 #include <lprocfs_status.h>
  40 #include <obd_support.h>
  41
  42 #include "llite_internal.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         if (IS_ERR_OR_NULL(llite_root)) {
  79                 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
  80                 llite_root = NULL;
  81 free_kobj:
  82                 kobject_put(llite_kobj);
  83                 llite_kobj = NULL;
  84         }
  85
  86         return rc;
  87 }
  88
  89 void llite_tunables_unregister(void)
  90 {
  91         kobject_put(llite_kobj);
  92         llite_kobj = NULL;
  93 }
  94
  95 /* <debugfs>/lustre/llite mount point registration */
  96 static const struct file_operations ll_rw_extents_stats_fops;
  97 static const struct file_operations ll_rw_extents_stats_pp_fops;
  98 static const struct file_operations ll_rw_offset_stats_fops;
  99
 100 /**
 101  * ll_stats_pid_write() - Determine if stats collection should be enabled
 102  * @buf: Buffer containing the data written
 103  * @len: Number of bytes in the buffer
 104  *
 105  * Several proc files begin collecting stats when a value is written, and stop
 106  * collecting when either '0' or 'disable' is written. This function checks the
 107  * written value to see if collection should be enabled or disabled.
 108  *
 109  * Return: If '0' or 'disable' is provided, 0 is returned. If the text
 110  * equivalent of a number is written, that number is returned. Otherwise,
 111  * 1 is returned. Non-zero return values indicate collection should be enabled.
 112  */
 113 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
 114 {
 115         unsigned long long value = 1;
 116         char kernbuf[16];
 117         int rc;
 118
 119         rc = kstrtoull_from_user(buf, len, 0, &value);
 120         if (rc < 0 && len < sizeof(kernbuf)) {
 121                 if (copy_from_user(kernbuf, buf, len))
 122                         return -EFAULT;
 123                 kernbuf[len] = 0;
 124
 125                 if (kernbuf[len - 1] == '\n')
 126                         kernbuf[len - 1] = 0;
 127
 128                 if (strncasecmp(kernbuf, "disable", 7) == 0)
 129                         value = 0;
 130         }
 131
 132         return value;
 133 }
 134
 135 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 136                               char *buf)
 137 {
 138         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 139                                               ll_kset.kobj);
 140         struct obd_statfs osfs;
 141         int rc;
 142
 143         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 144         if (rc)
 145                 return rc;
 146
 147         return sprintf(buf, "%u\n", osfs.os_bsize);
 148 }
 149 LUSTRE_RO_ATTR(blocksize);
 150
 151 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
 152                                    char *buf)
 153 {
 154         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 155                                               ll_kset.kobj);
 156
 157         return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
 158 }
 159
 160 static ssize_t stat_blocksize_store(struct kobject *kobj,
 161                                     struct attribute *attr,
 162                                     const char *buffer,
 163                                     size_t count)
 164 {
 165         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 166                                               ll_kset.kobj);
 167         unsigned int val;
 168         int rc;
 169
 170         rc = kstrtouint(buffer, 10, &val);
 171         if (rc)
 172                 return rc;
 173
 174         if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
 175                 return -ERANGE;
 176
 177         sbi->ll_stat_blksize = val;
 178
 179         return count;
 180 }
 181 LUSTRE_RW_ATTR(stat_blocksize);
 182
 183 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 184                                 char *buf)
 185 {
 186         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 187                                               ll_kset.kobj);
 188         struct obd_statfs osfs;
 189         u32 blk_size;
 190         u64 result;
 191         int rc;
 192
 193         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 194         if (rc)
 195                 return rc;
 196
 197         blk_size = osfs.os_bsize >> 10;
 198         result = osfs.os_blocks;
 199
 200         while (blk_size >>= 1)
 201                 result <<= 1;
 202
 203         return sprintf(buf, "%llu\n", result);
 204 }
 205 LUSTRE_RO_ATTR(kbytestotal);
 206
 207 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 208                                char *buf)
 209 {
 210         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 211                                               ll_kset.kobj);
 212         struct obd_statfs osfs;
 213         u32 blk_size;
 214         u64 result;
 215         int rc;
 216
 217         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 218         if (rc)
 219                 return rc;
 220
 221         blk_size = osfs.os_bsize >> 10;
 222         result = osfs.os_bfree;
 223
 224         while (blk_size >>= 1)
 225                 result <<= 1;
 226
 227         return sprintf(buf, "%llu\n", result);
 228 }
 229 LUSTRE_RO_ATTR(kbytesfree);
 230
 231 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 232                                 char *buf)
 233 {
 234         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 235                                               ll_kset.kobj);
 236         struct obd_statfs osfs;
 237         u32 blk_size;
 238         u64 result;
 239         int rc;
 240
 241         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 242         if (rc)
 243                 return rc;
 244
 245         blk_size = osfs.os_bsize >> 10;
 246         result = osfs.os_bavail;
 247
 248         while (blk_size >>= 1)
 249                 result <<= 1;
 250
 251         return sprintf(buf, "%llu\n", result);
 252 }
 253 LUSTRE_RO_ATTR(kbytesavail);
 254
 255 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 256                                char *buf)
 257 {
 258         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 259                                               ll_kset.kobj);
 260         struct obd_statfs osfs;
 261         int rc;
 262
 263         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 264         if (rc)
 265                 return rc;
 266
 267         return sprintf(buf, "%llu\n", osfs.os_files);
 268 }
 269 LUSTRE_RO_ATTR(filestotal);
 270
 271 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 272                               char *buf)
 273 {
 274         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 275                                               ll_kset.kobj);
 276         struct obd_statfs osfs;
 277         int rc;
 278
 279         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 280         if (rc)
 281                 return rc;
 282
 283         return sprintf(buf, "%llu\n", osfs.os_ffree);
 284 }
 285 LUSTRE_RO_ATTR(filesfree);
 286
 287 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
 288                                 char *buf)
 289 {
 290         return sprintf(buf, "local client\n");
 291 }
 292 LUSTRE_RO_ATTR(client_type);
 293
 294 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 295                            char *buf)
 296 {
 297         return sprintf(buf, "lustre\n");
 298 }
 299 LUSTRE_RO_ATTR(fstype);
 300
 301 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 302                          char *buf)
 303 {
 304         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 305                                               ll_kset.kobj);
 306
 307         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 308 }
 309 LUSTRE_RO_ATTR(uuid);
 310
 311 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 312 {
 313         struct super_block *sb = m->private;
 314
 315         /*
 316          * See description of statistical counters in struct cl_site, and
 317          * struct lu_site.
 318          */
 319         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 320 }
 321
 322 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
 323
 324 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
 325                                       struct attribute *attr, char *buf)
 326 {
 327         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 328                                               ll_kset.kobj);
 329         unsigned long ra_max_mb;
 330
 331         spin_lock(&sbi->ll_lock);
 332         ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
 333         spin_unlock(&sbi->ll_lock);
 334
 335         return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_mb);
 336 }
 337
 338 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
 339                                        struct attribute *attr,
 340                                        const char *buffer, size_t count)
 341 {
 342         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 343                                               ll_kset.kobj);
 344         u64 ra_max_mb, pages_number;
 345         int rc;
 346
 347         rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
 348         if (rc)
 349                 return rc;
 350
 351         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
 352         CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
 353                sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
 354         if (pages_number > cfs_totalram_pages() / 2) {
 355                 /* 1/2 of RAM */
 356                 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
 357                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 358                        PAGES_TO_MiB(cfs_totalram_pages() / 2));
 359                 return -ERANGE;
 360         }
 361
 362         spin_lock(&sbi->ll_lock);
 363         sbi->ll_ra_info.ra_max_pages = pages_number;
 364         spin_unlock(&sbi->ll_lock);
 365
 366         return count;
 367 }
 368 LUSTRE_RW_ATTR(max_read_ahead_mb);
 369
 370 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
 371                                                struct attribute *attr,
 372                                                char *buf)
 373 {
 374         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 375                                               ll_kset.kobj);
 376         unsigned long ra_max_file_mb;
 377
 378         spin_lock(&sbi->ll_lock);
 379         ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
 380         spin_unlock(&sbi->ll_lock);
 381
 382         return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_file_mb);
 383 }
 384
 385 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
 386                                                 struct attribute *attr,
 387                                                 const char *buffer,
 388                                                 size_t count)
 389 {
 390         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 391                                               ll_kset.kobj);
 392         u64 ra_max_file_mb, pages_number;
 393         int rc;
 394
 395         rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
 396         if (rc)
 397                 return rc;
 398
 399         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
 400         if (pages_number > sbi->ll_ra_info.ra_max_pages) {
 401                 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
 402                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 403                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 404                 return -ERANGE;
 405         }
 406
 407         spin_lock(&sbi->ll_lock);
 408         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 409         spin_unlock(&sbi->ll_lock);
 410
 411         return count;
 412 }
 413 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
 414
 415 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
 416                                             struct attribute *attr, char *buf)
 417 {
 418         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 419                                               ll_kset.kobj);
 420         unsigned long ra_max_whole_mb;
 421
 422         spin_lock(&sbi->ll_lock);
 423         ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
 424         spin_unlock(&sbi->ll_lock);
 425
 426         return snprintf(buf, PAGE_SIZE, "%lu\n", ra_max_whole_mb);
 427 }
 428
 429 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
 430                                              struct attribute *attr,
 431                                              const char *buffer, size_t count)
 432 {
 433         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 434                                               ll_kset.kobj);
 435         u64 ra_max_whole_mb, pages_number;
 436         int rc;
 437
 438         rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
 439         if (rc)
 440                 return rc;
 441
 442         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
 443         /* Cap this at the current max readahead window size, the readahead
 444          * algorithm does this anyway so it's pointless to set it larger.
 445          */
 446         if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 447                 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
 448                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 449                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 450
 451                 return -ERANGE;
 452         }
 453
 454         spin_lock(&sbi->ll_lock);
 455         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 456         spin_unlock(&sbi->ll_lock);
 457
 458         return count;
 459 }
 460 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
 461
 462 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 463 {
 464         struct super_block     *sb    = m->private;
 465         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 466         struct cl_client_cache *cache = sbi->ll_cache;
 467         long max_cached_mb;
 468         long unused_mb;
 469
 470         mutex_lock(&cache->ccc_max_cache_mb_lock);
 471         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
 472         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
 473         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 474         seq_printf(m, "users: %d\n"
 475                       "max_cached_mb: %ld\n"
 476                       "used_mb: %ld\n"
 477                       "unused_mb: %ld\n"
 478                       "reclaim_count: %u\n",
 479                    atomic_read(&cache->ccc_users),
 480                    max_cached_mb,
 481                    max_cached_mb - unused_mb,
 482                    unused_mb,
 483                    cache->ccc_lru_shrinkers);
 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 sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 607 }
 608
 609 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
 610                                const char *buffer, size_t count)
 611 {
 612         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 613                                               ll_kset.kobj);
 614         bool val;
 615         int tmp;
 616         int rc;
 617
 618         if (!sbi->ll_dt_exp)
 619                 /* Not set up yet */
 620                 return -EAGAIN;
 621
 622         rc = kstrtobool(buffer, &val);
 623         if (rc)
 624                 return rc;
 625         if (val)
 626                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 627         else
 628                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 629         tmp = val;
 630
 631         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 632                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
 633         if (rc)
 634                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 635
 636         return count;
 637 }
 638 LUSTRE_RW_ATTR(checksums);
 639
 640 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
 641
 642 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 643                               enum stats_track_type type)
 644 {
 645         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 646                                               ll_kset.kobj);
 647
 648         if (sbi->ll_stats_track_type == type)
 649                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 650         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 651                 return sprintf(buf, "0 (all)\n");
 652
 653         return sprintf(buf, "untracked\n");
 654 }
 655
 656 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 657                               size_t count, enum stats_track_type type)
 658 {
 659         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 660                                               ll_kset.kobj);
 661         unsigned long pid;
 662         int rc;
 663
 664         rc = kstrtoul(buffer, 10, &pid);
 665         if (rc)
 666                 return rc;
 667
 668         sbi->ll_stats_track_id = pid;
 669         if (pid == 0)
 670                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 671         else
 672                 sbi->ll_stats_track_type = type;
 673         lprocfs_clear_stats(sbi->ll_stats);
 674         return count;
 675 }
 676
 677 static ssize_t stats_track_pid_show(struct kobject *kobj,
 678                                     struct attribute *attr,
 679                                     char *buf)
 680 {
 681         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 682 }
 683
 684 static ssize_t stats_track_pid_store(struct kobject *kobj,
 685                                      struct attribute *attr,
 686                                      const char *buffer,
 687                                      size_t count)
 688 {
 689         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 690 }
 691 LUSTRE_RW_ATTR(stats_track_pid);
 692
 693 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 694                                      struct attribute *attr,
 695                                      char *buf)
 696 {
 697         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 698 }
 699
 700 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 701                                       struct attribute *attr,
 702                                       const char *buffer,
 703                                       size_t count)
 704 {
 705         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 706 }
 707 LUSTRE_RW_ATTR(stats_track_ppid);
 708
 709 static ssize_t stats_track_gid_show(struct kobject *kobj,
 710                                     struct attribute *attr,
 711                                     char *buf)
 712 {
 713         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 714 }
 715
 716 static ssize_t stats_track_gid_store(struct kobject *kobj,
 717                                      struct attribute *attr,
 718                                      const char *buffer,
 719                                      size_t count)
 720 {
 721         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 722 }
 723 LUSTRE_RW_ATTR(stats_track_gid);
 724
 725 static ssize_t statahead_running_max_show(struct kobject *kobj,
 726                                           struct attribute *attr,
 727                                           char *buf)
 728 {
 729         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 730                                               ll_kset.kobj);
 731
 732         return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
 733 }
 734
 735 static ssize_t statahead_running_max_store(struct kobject *kobj,
 736                                            struct attribute *attr,
 737                                            const char *buffer,
 738                                            size_t count)
 739 {
 740         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 741                                               ll_kset.kobj);
 742         unsigned long val;
 743         int rc;
 744
 745         rc = kstrtoul(buffer, 0, &val);
 746         if (rc)
 747                 return rc;
 748
 749         if (val <= LL_SA_RUNNING_MAX) {
 750                 sbi->ll_sa_running_max = val;
 751                 return count;
 752         }
 753
 754         CERROR("Bad statahead_running_max value %lu. Valid values "
 755                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 756
 757         return -ERANGE;
 758 }
 759 LUSTRE_RW_ATTR(statahead_running_max);
 760
 761 static ssize_t statahead_max_show(struct kobject *kobj,
 762                                   struct attribute *attr,
 763                                   char *buf)
 764 {
 765         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 766                                               ll_kset.kobj);
 767
 768         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 769 }
 770
 771 static ssize_t statahead_max_store(struct kobject *kobj,
 772                                    struct attribute *attr,
 773                                    const char *buffer,
 774                                    size_t count)
 775 {
 776         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 777                                               ll_kset.kobj);
 778         unsigned long val;
 779         int rc;
 780
 781         rc = kstrtoul(buffer, 0, &val);
 782         if (rc)
 783                 return rc;
 784
 785         if (val <= LL_SA_RPC_MAX)
 786                 sbi->ll_sa_max = val;
 787         else
 788                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 789                        val, LL_SA_RPC_MAX);
 790
 791         return count;
 792 }
 793 LUSTRE_RW_ATTR(statahead_max);
 794
 795 static ssize_t statahead_agl_show(struct kobject *kobj,
 796                                   struct attribute *attr,
 797                                   char *buf)
 798 {
 799         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 800                                               ll_kset.kobj);
 801
 802         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
 803 }
 804
 805 static ssize_t statahead_agl_store(struct kobject *kobj,
 806                                    struct attribute *attr,
 807                                    const char *buffer,
 808                                    size_t count)
 809 {
 810         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 811                                               ll_kset.kobj);
 812         bool val;
 813         int rc;
 814
 815         rc = kstrtobool(buffer, &val);
 816         if (rc)
 817                 return rc;
 818
 819         if (val)
 820                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 821         else
 822                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
 823
 824         return count;
 825 }
 826 LUSTRE_RW_ATTR(statahead_agl);
 827
 828 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 829 {
 830         struct super_block *sb = m->private;
 831         struct ll_sb_info *sbi = ll_s2sbi(sb);
 832
 833         seq_printf(m, "statahead total: %u\n"
 834                       "statahead wrong: %u\n"
 835                       "agl total: %u\n",
 836                    atomic_read(&sbi->ll_sa_total),
 837                    atomic_read(&sbi->ll_sa_wrong),
 838                    atomic_read(&sbi->ll_agl_total));
 839         return 0;
 840 }
 841
 842 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
 843
 844 static ssize_t lazystatfs_show(struct kobject *kobj,
 845                                struct attribute *attr,
 846                                char *buf)
 847 {
 848         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 849                                               ll_kset.kobj);
 850
 851         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 852 }
 853
 854 static ssize_t lazystatfs_store(struct kobject *kobj,
 855                                 struct attribute *attr,
 856                                 const char *buffer,
 857                                 size_t count)
 858 {
 859         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 860                                               ll_kset.kobj);
 861         bool val;
 862         int rc;
 863
 864         rc = kstrtobool(buffer, &val);
 865         if (rc)
 866                 return rc;
 867
 868         if (val)
 869                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 870         else
 871                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 872
 873         return count;
 874 }
 875 LUSTRE_RW_ATTR(lazystatfs);
 876
 877 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
 878                                    char *buf)
 879 {
 880         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 881                                               ll_kset.kobj);
 882
 883         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
 884 }
 885
 886 static ssize_t statfs_max_age_store(struct kobject *kobj,
 887                                     struct attribute *attr, const char *buffer,
 888                                     size_t count)
 889 {
 890         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 891                                               ll_kset.kobj);
 892         unsigned int val;
 893         int rc;
 894
 895         rc = kstrtouint(buffer, 10, &val);
 896         if (rc)
 897                 return rc;
 898         if (val > OBD_STATFS_CACHE_MAX_AGE)
 899                 return -EINVAL;
 900
 901         sbi->ll_statfs_max_age = val;
 902
 903         return count;
 904 }
 905 LUSTRE_RW_ATTR(statfs_max_age);
 906
 907 static ssize_t max_easize_show(struct kobject *kobj,
 908                                struct attribute *attr,
 909                                char *buf)
 910 {
 911         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 912                                               ll_kset.kobj);
 913         unsigned int ealen;
 914         int rc;
 915
 916         rc = ll_get_max_mdsize(sbi, &ealen);
 917         if (rc)
 918                 return rc;
 919
 920         /* Limit xattr size returned to userspace based on kernel maximum */
 921         return snprintf(buf, PAGE_SIZE, "%u\n",
 922                         ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 923 }
 924 LUSTRE_RO_ATTR(max_easize);
 925
 926 /**
 927  * Get default_easize.
 928  *
 929  * \see client_obd::cl_default_mds_easize
 930  *
 931  * \param[in] m         seq_file handle
 932  * \param[in] v         unused for single entry
 933  *
 934  * \retval 0            on success
 935  * \retval negative     negated errno on failure
 936  */
 937 static ssize_t default_easize_show(struct kobject *kobj,
 938                                    struct attribute *attr,
 939                                    char *buf)
 940 {
 941         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 942                                               ll_kset.kobj);
 943         unsigned int ealen;
 944         int rc;
 945
 946         rc = ll_get_default_mdsize(sbi, &ealen);
 947         if (rc)
 948                 return rc;
 949
 950         /* Limit xattr size returned to userspace based on kernel maximum */
 951         return snprintf(buf, PAGE_SIZE, "%u\n",
 952                         ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 953 }
 954
 955 /**
 956  * Set default_easize.
 957  *
 958  * Range checking on the passed value is handled by
 959  * ll_set_default_mdsize().
 960  *
 961  * \see client_obd::cl_default_mds_easize
 962  *
 963  * \param[in] file      proc file
 964  * \param[in] buffer    string passed from user space
 965  * \param[in] count     \a buffer length
 966  * \param[in] off       unused for single entry
 967  *
 968  * \retval positive     \a count on success
 969  * \retval negative     negated errno on failure
 970  */
 971 static ssize_t default_easize_store(struct kobject *kobj,
 972                                     struct attribute *attr,
 973                                     const char *buffer,
 974                                     size_t count)
 975 {
 976         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 977                                               ll_kset.kobj);
 978         unsigned int val;
 979         int rc;
 980
 981         if (count == 0)
 982                 return 0;
 983
 984         rc = kstrtouint(buffer, 10, &val);
 985         if (rc)
 986                 return rc;
 987
 988         rc = ll_set_default_mdsize(sbi, val);
 989         if (rc)
 990                 return rc;
 991
 992         return count;
 993 }
 994 LUSTRE_RW_ATTR(default_easize);
 995
 996 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 997 {
 998         const char *str[] = LL_SBI_FLAGS;
 999         struct super_block *sb = m->private;
1000         int flags = ll_s2sbi(sb)->ll_flags;
1001         int i = 0;
1002
1003         while (flags != 0) {
1004                 if (ARRAY_SIZE(str) <= i) {
1005                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
1006                                 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
1007                         return -EINVAL;
1008                 }
1009
1010                 if (flags & 0x1)
1011                         seq_printf(m, "%s ", str[i]);
1012                 flags >>= 1;
1013                 ++i;
1014         }
1015         seq_printf(m, "\b\n");
1016         return 0;
1017 }
1018
1019 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1020
1021 static ssize_t xattr_cache_show(struct kobject *kobj,
1022                                 struct attribute *attr,
1023                                 char *buf)
1024 {
1025         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1026                                               ll_kset.kobj);
1027
1028         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1029 }
1030
1031 static ssize_t xattr_cache_store(struct kobject *kobj,
1032                                  struct attribute *attr,
1033                                  const char *buffer,
1034                                  size_t count)
1035 {
1036         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1037                                               ll_kset.kobj);
1038         bool val;
1039         int rc;
1040
1041         rc = kstrtobool(buffer, &val);
1042         if (rc)
1043                 return rc;
1044
1045         if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1046                 return -ENOTSUPP;
1047
1048         sbi->ll_xattr_cache_enabled = val;
1049         sbi->ll_xattr_cache_set = 1;
1050
1051         return count;
1052 }
1053 LUSTRE_RW_ATTR(xattr_cache);
1054
1055 static ssize_t tiny_write_show(struct kobject *kobj,
1056                                struct attribute *attr,
1057                                char *buf)
1058 {
1059         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1060                                               ll_kset.kobj);
1061
1062         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1063 }
1064
1065 static ssize_t tiny_write_store(struct kobject *kobj,
1066                                 struct attribute *attr,
1067                                 const char *buffer,
1068                                 size_t count)
1069 {
1070         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1071                                               ll_kset.kobj);
1072         bool val;
1073         int rc;
1074
1075         rc = kstrtobool(buffer, &val);
1076         if (rc)
1077                 return rc;
1078
1079         spin_lock(&sbi->ll_lock);
1080         if (val)
1081                 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1082         else
1083                 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1084         spin_unlock(&sbi->ll_lock);
1085
1086         return count;
1087 }
1088 LUSTRE_RW_ATTR(tiny_write);
1089
1090 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1091                                                struct attribute *attr,
1092                                                char *buf)
1093 {
1094         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1095                                               ll_kset.kobj);
1096
1097         return snprintf(buf, PAGE_SIZE, "%u\n",
1098                         sbi->ll_ra_info.ra_async_max_active);
1099 }
1100
1101 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1102                                                  struct attribute *attr,
1103                                                  const char *buffer,
1104                                                  size_t count)
1105 {
1106         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1107                                               ll_kset.kobj);
1108         unsigned int val;
1109         int rc;
1110
1111         rc = kstrtouint(buffer, 10, &val);
1112         if (rc)
1113                 return rc;
1114
1115         /**
1116          * It doesn't make any sense to make it exceed what
1117          * workqueue could acutally support. This can easily
1118          * over subscripe the cores but Lustre internally
1119          * throttles to avoid those impacts.
1120          */
1121         if (val > WQ_UNBOUND_MAX_ACTIVE) {
1122                 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1123                        sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1124                 return -ERANGE;
1125         }
1126
1127         spin_lock(&sbi->ll_lock);
1128         sbi->ll_ra_info.ra_async_max_active = val;
1129         spin_unlock(&sbi->ll_lock);
1130
1131         return count;
1132 }
1133 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1134
1135 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1136                                                        struct attribute *attr,
1137                                                        char *buf)
1138 {
1139         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1140                                               ll_kset.kobj);
1141
1142         return snprintf(buf, PAGE_SIZE, "%lu\n",
1143              PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1144 }
1145
1146 static ssize_t
1147 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1148                                          struct attribute *attr,
1149                                          const char *buffer, size_t count)
1150 {
1151         unsigned long pages_number;
1152         unsigned long max_ra_per_file;
1153         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1154                                               ll_kset.kobj);
1155         int rc;
1156
1157         rc = kstrtoul(buffer, 10, &pages_number);
1158         if (rc)
1159                 return rc;
1160
1161         pages_number = MiB_TO_PAGES(pages_number);
1162         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1163         if (pages_number < 0 || pages_number > max_ra_per_file) {
1164                 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1165                        "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1166                        PAGES_TO_MiB(pages_number),
1167                        PAGES_TO_MiB(max_ra_per_file));
1168                 return -ERANGE;
1169         }
1170         sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1171
1172         return count;
1173 }
1174 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1175
1176 static ssize_t fast_read_show(struct kobject *kobj,
1177                               struct attribute *attr,
1178                               char *buf)
1179 {
1180         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1181                                               ll_kset.kobj);
1182
1183         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1184 }
1185
1186 static ssize_t fast_read_store(struct kobject *kobj,
1187                                struct attribute *attr,
1188                                const char *buffer,
1189                                size_t count)
1190 {
1191         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1192                                               ll_kset.kobj);
1193         bool val;
1194         int rc;
1195
1196         rc = kstrtobool(buffer, &val);
1197         if (rc)
1198                 return rc;
1199
1200         spin_lock(&sbi->ll_lock);
1201         if (val)
1202                 sbi->ll_flags |= LL_SBI_FAST_READ;
1203         else
1204                 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1205         spin_unlock(&sbi->ll_lock);
1206
1207         return count;
1208 }
1209 LUSTRE_RW_ATTR(fast_read);
1210
1211 static ssize_t file_heat_show(struct kobject *kobj,
1212                               struct attribute *attr,
1213                               char *buf)
1214 {
1215         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1216                                               ll_kset.kobj);
1217
1218         return snprintf(buf, PAGE_SIZE, "%u\n",
1219                         !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1220 }
1221
1222 static ssize_t file_heat_store(struct kobject *kobj,
1223                                struct attribute *attr,
1224                                const char *buffer,
1225                                size_t count)
1226 {
1227         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1228                                               ll_kset.kobj);
1229         bool val;
1230         int rc;
1231
1232         rc = kstrtobool(buffer, &val);
1233         if (rc)
1234                 return rc;
1235
1236         spin_lock(&sbi->ll_lock);
1237         if (val)
1238                 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1239         else
1240                 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1241         spin_unlock(&sbi->ll_lock);
1242
1243         return count;
1244 }
1245 LUSTRE_RW_ATTR(file_heat);
1246
1247 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1248                                           struct attribute *attr,
1249                                           char *buf)
1250 {
1251         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1252                                               ll_kset.kobj);
1253
1254         return snprintf(buf, PAGE_SIZE, "%u\n",
1255                        (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1256 }
1257
1258 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1259                                            struct attribute *attr,
1260                                            const char *buffer,
1261                                            size_t count)
1262 {
1263         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1264                                               ll_kset.kobj);
1265         unsigned long val;
1266         int rc;
1267
1268         rc = kstrtoul(buffer, 10, &val);
1269         if (rc)
1270                 return rc;
1271
1272         if (val < 0 || val > 100)
1273                 return -ERANGE;
1274
1275         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1276
1277         return count;
1278 }
1279 LUSTRE_RW_ATTR(heat_decay_percentage);
1280
1281 static ssize_t heat_period_second_show(struct kobject *kobj,
1282                                        struct attribute *attr,
1283                                        char *buf)
1284 {
1285         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1286                                               ll_kset.kobj);
1287
1288         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1289 }
1290
1291 static ssize_t heat_period_second_store(struct kobject *kobj,
1292                                         struct attribute *attr,
1293                                         const char *buffer,
1294                                         size_t count)
1295 {
1296         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1297                                               ll_kset.kobj);
1298         unsigned long val;
1299         int rc;
1300
1301         rc = kstrtoul(buffer, 10, &val);
1302         if (rc)
1303                 return rc;
1304
1305         if (val <= 0)
1306                 return -ERANGE;
1307
1308         sbi->ll_heat_period_second = val;
1309
1310         return count;
1311 }
1312 LUSTRE_RW_ATTR(heat_period_second);
1313
1314 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1315 {
1316         struct super_block      *sb    = m->private;
1317         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1318         struct cl_client_cache  *cache = sbi->ll_cache;
1319         long pages;
1320         int mb;
1321
1322         pages = atomic_long_read(&cache->ccc_unstable_nr);
1323         mb    = (pages * PAGE_SIZE) >> 20;
1324
1325         seq_printf(m, "unstable_check:     %8d\n"
1326                       "unstable_pages: %12ld\n"
1327                       "unstable_mb:        %8d\n",
1328                    cache->ccc_unstable_check, pages, mb);
1329         return 0;
1330 }
1331
1332 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1333                                            const char __user *buffer,
1334                                            size_t count, loff_t *unused)
1335 {
1336         struct seq_file *seq = file->private_data;
1337         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1338         char kernbuf[128];
1339         bool val;
1340         int rc;
1341
1342         if (count == 0)
1343                 return 0;
1344         if (count >= sizeof(kernbuf))
1345                 return -EINVAL;
1346
1347         if (copy_from_user(kernbuf, buffer, count))
1348                 return -EFAULT;
1349         kernbuf[count] = 0;
1350
1351         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1352                   kernbuf;
1353         rc = kstrtobool_from_user(buffer, count, &val);
1354         if (rc < 0)
1355                 return rc;
1356
1357         /* borrow lru lock to set the value */
1358         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1359         sbi->ll_cache->ccc_unstable_check = val;
1360         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1361
1362         return count;
1363 }
1364
1365 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1366
1367 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1368 {
1369         struct super_block *sb = m->private;
1370         struct ll_sb_info *sbi = ll_s2sbi(sb);
1371         struct root_squash_info *squash = &sbi->ll_squash;
1372
1373         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1374         return 0;
1375 }
1376
1377 static ssize_t ll_root_squash_seq_write(struct file *file,
1378                                         const char __user *buffer,
1379                                         size_t count, loff_t *off)
1380 {
1381         struct seq_file *m = file->private_data;
1382         struct super_block *sb = m->private;
1383         struct ll_sb_info *sbi = ll_s2sbi(sb);
1384         struct root_squash_info *squash = &sbi->ll_squash;
1385
1386         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1387 }
1388
1389 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1390
1391 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1392 {
1393         struct super_block *sb = m->private;
1394         struct ll_sb_info *sbi = ll_s2sbi(sb);
1395         struct root_squash_info *squash = &sbi->ll_squash;
1396         int len;
1397
1398         spin_lock(&squash->rsi_lock);
1399         if (!list_empty(&squash->rsi_nosquash_nids)) {
1400                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1401                                         &squash->rsi_nosquash_nids);
1402                 m->count += len;
1403                 seq_putc(m, '\n');
1404         } else {
1405                 seq_puts(m, "NONE\n");
1406         }
1407         spin_unlock(&squash->rsi_lock);
1408
1409         return 0;
1410 }
1411
1412 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1413                                           const char __user *buffer,
1414                                           size_t count, loff_t *off)
1415 {
1416         struct seq_file *m = file->private_data;
1417         struct super_block *sb = m->private;
1418         struct ll_sb_info *sbi = ll_s2sbi(sb);
1419         struct root_squash_info *squash = &sbi->ll_squash;
1420         int rc;
1421
1422         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1423         if (rc < 0)
1424                 return rc;
1425
1426         ll_compute_rootsquash_state(sbi);
1427
1428         return rc;
1429 }
1430
1431 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1432
1433 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1434 {
1435         struct super_block *sb = m->private;
1436         struct ll_sb_info *sbi = ll_s2sbi(sb);
1437
1438         return pcc_super_dump(&sbi->ll_pcc_super, m);
1439 }
1440
1441 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1442                                 size_t count, loff_t *off)
1443 {
1444         struct seq_file *m = file->private_data;
1445         struct super_block *sb = m->private;
1446         struct ll_sb_info *sbi = ll_s2sbi(sb);
1447         int rc;
1448         char *kernbuf;
1449
1450         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1451                 return -EINVAL;
1452
1453         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1454                 return -EOPNOTSUPP;
1455
1456         OBD_ALLOC(kernbuf, count + 1);
1457         if (kernbuf == NULL)
1458                 return -ENOMEM;
1459
1460         if (copy_from_user(kernbuf, buffer, count))
1461                 GOTO(out_free_kernbuff, rc = -EFAULT);
1462
1463         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1464 out_free_kernbuff:
1465         OBD_FREE(kernbuf, count + 1);
1466         return rc ? rc : count;
1467 }
1468 LPROC_SEQ_FOPS(ll_pcc);
1469
1470 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1471         { .name =       "site",
1472           .fops =       &ll_site_stats_fops                     },
1473         { .name =       "max_cached_mb",
1474           .fops =       &ll_max_cached_mb_fops                  },
1475         { .name =       "statahead_stats",
1476           .fops =       &ll_statahead_stats_fops                },
1477         { .name =       "unstable_stats",
1478           .fops =       &ll_unstable_stats_fops                 },
1479         { .name =       "sbi_flags",
1480           .fops =       &ll_sbi_flags_fops                      },
1481         { .name =       "root_squash",
1482           .fops =       &ll_root_squash_fops                    },
1483         { .name =       "nosquash_nids",
1484           .fops =       &ll_nosquash_nids_fops                  },
1485         { .name =       "pcc",
1486           .fops =       &ll_pcc_fops,                           },
1487         { NULL }
1488 };
1489
1490 #define MAX_STRING_SIZE 128
1491
1492 static struct attribute *llite_attrs[] = {
1493         &lustre_attr_blocksize.attr,
1494         &lustre_attr_stat_blocksize.attr,
1495         &lustre_attr_kbytestotal.attr,
1496         &lustre_attr_kbytesfree.attr,
1497         &lustre_attr_kbytesavail.attr,
1498         &lustre_attr_filestotal.attr,
1499         &lustre_attr_filesfree.attr,
1500         &lustre_attr_client_type.attr,
1501         &lustre_attr_fstype.attr,
1502         &lustre_attr_uuid.attr,
1503         &lustre_attr_checksums.attr,
1504         &lustre_attr_checksum_pages.attr,
1505         &lustre_attr_max_read_ahead_mb.attr,
1506         &lustre_attr_max_read_ahead_per_file_mb.attr,
1507         &lustre_attr_max_read_ahead_whole_mb.attr,
1508         &lustre_attr_max_read_ahead_async_active.attr,
1509         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1510         &lustre_attr_stats_track_pid.attr,
1511         &lustre_attr_stats_track_ppid.attr,
1512         &lustre_attr_stats_track_gid.attr,
1513         &lustre_attr_statahead_running_max.attr,
1514         &lustre_attr_statahead_max.attr,
1515         &lustre_attr_statahead_agl.attr,
1516         &lustre_attr_lazystatfs.attr,
1517         &lustre_attr_statfs_max_age.attr,
1518         &lustre_attr_max_easize.attr,
1519         &lustre_attr_default_easize.attr,
1520         &lustre_attr_xattr_cache.attr,
1521         &lustre_attr_fast_read.attr,
1522         &lustre_attr_tiny_write.attr,
1523         &lustre_attr_file_heat.attr,
1524         &lustre_attr_heat_decay_percentage.attr,
1525         &lustre_attr_heat_period_second.attr,
1526         NULL,
1527 };
1528
1529 static void sbi_kobj_release(struct kobject *kobj)
1530 {
1531         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1532                                               ll_kset.kobj);
1533         complete(&sbi->ll_kobj_unregister);
1534 }
1535
1536 static struct kobj_type sbi_ktype = {
1537         .default_attrs  = llite_attrs,
1538         .sysfs_ops      = &lustre_sysfs_ops,
1539         .release        = sbi_kobj_release,
1540 };
1541
1542 #define LPROCFS_TYPE_LATENCY \
1543         (LPROCFS_TYPE_USEC | LPROCFS_CNTR_AVGMINMAX | LPROCFS_CNTR_STDDEV)
1544 static const struct llite_file_opcode {
1545         __u32           opcode;
1546         __u32           type;
1547         const char      *opname;
1548 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1549         /* file operation */
1550         { LPROC_LL_READ_BYTES,  LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1551                 "read_bytes" },
1552         { LPROC_LL_WRITE_BYTES, LPROCFS_CNTR_AVGMINMAX | LPROCFS_TYPE_BYTES,
1553                 "write_bytes" },
1554         { LPROC_LL_READ,        LPROCFS_TYPE_LATENCY,   "read" },
1555         { LPROC_LL_WRITE,       LPROCFS_TYPE_LATENCY,   "write" },
1556         { LPROC_LL_IOCTL,       LPROCFS_TYPE_REQS,      "ioctl" },
1557         { LPROC_LL_OPEN,        LPROCFS_TYPE_LATENCY,   "open" },
1558         { LPROC_LL_RELEASE,     LPROCFS_TYPE_LATENCY,   "close" },
1559         { LPROC_LL_MMAP,        LPROCFS_TYPE_LATENCY,   "mmap" },
1560         { LPROC_LL_FAULT,       LPROCFS_TYPE_LATENCY,   "page_fault" },
1561         { LPROC_LL_MKWRITE,     LPROCFS_TYPE_LATENCY,   "page_mkwrite" },
1562         { LPROC_LL_LLSEEK,      LPROCFS_TYPE_LATENCY,   "seek" },
1563         { LPROC_LL_FSYNC,       LPROCFS_TYPE_LATENCY,   "fsync" },
1564         { LPROC_LL_READDIR,     LPROCFS_TYPE_LATENCY,   "readdir" },
1565         /* inode operation */
1566         { LPROC_LL_SETATTR,     LPROCFS_TYPE_LATENCY,   "setattr" },
1567         { LPROC_LL_TRUNC,       LPROCFS_TYPE_LATENCY,   "truncate" },
1568         { LPROC_LL_FLOCK,       LPROCFS_TYPE_LATENCY,   "flock" },
1569         { LPROC_LL_GETATTR,     LPROCFS_TYPE_LATENCY,   "getattr" },
1570         /* dir inode operation */
1571         { LPROC_LL_CREATE,      LPROCFS_TYPE_LATENCY,   "create" },
1572         { LPROC_LL_LINK,        LPROCFS_TYPE_LATENCY,   "link" },
1573         { LPROC_LL_UNLINK,      LPROCFS_TYPE_LATENCY,   "unlink" },
1574         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_LATENCY,   "symlink" },
1575         { LPROC_LL_MKDIR,       LPROCFS_TYPE_LATENCY,   "mkdir" },
1576         { LPROC_LL_RMDIR,       LPROCFS_TYPE_LATENCY,   "rmdir" },
1577         { LPROC_LL_MKNOD,       LPROCFS_TYPE_LATENCY,   "mknod" },
1578         { LPROC_LL_RENAME,      LPROCFS_TYPE_LATENCY,   "rename" },
1579         /* special inode operation */
1580         { LPROC_LL_STATFS,      LPROCFS_TYPE_LATENCY,   "statfs" },
1581         { LPROC_LL_SETXATTR,    LPROCFS_TYPE_LATENCY,   "setxattr" },
1582         { LPROC_LL_GETXATTR,    LPROCFS_TYPE_LATENCY,   "getxattr" },
1583         { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS,    "getxattr_hits" },
1584         { LPROC_LL_LISTXATTR,   LPROCFS_TYPE_LATENCY,   "listxattr" },
1585         { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY,   "removexattr" },
1586         { LPROC_LL_INODE_PERM,  LPROCFS_TYPE_LATENCY,   "inode_permission" },
1587 };
1588
1589 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1590 {
1591         if (!sbi->ll_stats)
1592                 return;
1593
1594         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1595                 lprocfs_counter_add(sbi->ll_stats, op, count);
1596         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1597                  sbi->ll_stats_track_id == current->pid)
1598                 lprocfs_counter_add(sbi->ll_stats, op, count);
1599         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1600                  sbi->ll_stats_track_id == current->parent->pid)
1601                 lprocfs_counter_add(sbi->ll_stats, op, count);
1602         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1603                  sbi->ll_stats_track_id ==
1604                         from_kgid(&init_user_ns, current_gid()))
1605                 lprocfs_counter_add(sbi->ll_stats, op, count);
1606 }
1607 EXPORT_SYMBOL(ll_stats_ops_tally);
1608
1609 static const char *ra_stat_string[] = {
1610         [RA_STAT_HIT] = "hits",
1611         [RA_STAT_MISS] = "misses",
1612         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1613         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1614         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1615         [RA_STAT_FAILED_MATCH] = "failed lock match",
1616         [RA_STAT_DISCARDED] = "read but discarded",
1617         [RA_STAT_ZERO_LEN] = "zero length file",
1618         [RA_STAT_ZERO_WINDOW] = "zero size window",
1619         [RA_STAT_EOF] = "read-ahead to EOF",
1620         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1621         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1622         [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1623         [RA_STAT_ASYNC] = "async readahead",
1624         [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1625 };
1626
1627 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1628 {
1629         struct lustre_sb_info *lsi = s2lsi(sb);
1630         struct ll_sb_info *sbi = ll_s2sbi(sb);
1631         int err, id;
1632
1633         ENTRY;
1634         LASSERT(sbi);
1635
1636         if (IS_ERR_OR_NULL(llite_root))
1637                 goto out_ll_kset;
1638
1639         sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1640         ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1641
1642         debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1643                             &vvp_dump_pgcache_file_ops);
1644
1645         debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1646                                  &ll_rw_extents_stats_fops);
1647
1648         debugfs_create_file("extents_stats_per_process", 0644,
1649                             sbi->ll_debugfs_entry, sbi,
1650                             &ll_rw_extents_stats_pp_fops);
1651
1652         debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1653                             &ll_rw_offset_stats_fops);
1654
1655         /* File operations stats */
1656         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1657                                             LPROCFS_STATS_FLAG_NONE);
1658         if (sbi->ll_stats == NULL)
1659                 GOTO(out_debugfs, err = -ENOMEM);
1660
1661         /* do counter init */
1662         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1663                 u32 type = llite_opcode_table[id].type;
1664                 void *ptr = NULL;
1665
1666                 if (type & LPROCFS_TYPE_REQS)
1667                         ptr = "reqs";
1668                 else if (type & LPROCFS_TYPE_BYTES)
1669                         ptr = "bytes";
1670                 else if (type & LPROCFS_TYPE_PAGES)
1671                         ptr = "pages";
1672                 else if (type & LPROCFS_TYPE_USEC)
1673                         ptr = "usec";
1674                 lprocfs_counter_init(sbi->ll_stats,
1675                                      llite_opcode_table[id].opcode,
1676                                      (type & LPROCFS_CNTR_AVGMINMAX),
1677                                      llite_opcode_table[id].opname, ptr);
1678         }
1679
1680         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1681                                       sbi->ll_stats);
1682         if (err)
1683                 GOTO(out_stats, err);
1684
1685         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1686                                                LPROCFS_STATS_FLAG_NONE);
1687         if (sbi->ll_ra_stats == NULL)
1688                 GOTO(out_stats, err = -ENOMEM);
1689
1690         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1691                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1692                                      ra_stat_string[id], "pages");
1693
1694         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1695                                       sbi->ll_ra_stats);
1696         if (err)
1697                 GOTO(out_ra_stats, err);
1698
1699 out_ll_kset:
1700         /* Yes we also register sysfs mount kset here as well */
1701         sbi->ll_kset.kobj.parent = llite_kobj;
1702         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1703         init_completion(&sbi->ll_kobj_unregister);
1704         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1705         if (err)
1706                 GOTO(out_ra_stats, err);
1707
1708         err = kset_register(&sbi->ll_kset);
1709         if (err)
1710                 GOTO(out_ra_stats, err);
1711
1712         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1713
1714         RETURN(0);
1715 out_ra_stats:
1716         lprocfs_free_stats(&sbi->ll_ra_stats);
1717 out_stats:
1718         lprocfs_free_stats(&sbi->ll_stats);
1719 out_debugfs:
1720         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1721
1722         RETURN(err);
1723 }
1724
1725 void ll_debugfs_unregister_super(struct super_block *sb)
1726 {
1727         struct lustre_sb_info *lsi = s2lsi(sb);
1728         struct ll_sb_info *sbi = ll_s2sbi(sb);
1729
1730         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1731
1732         if (sbi->ll_dt_obd)
1733                 sysfs_remove_link(&sbi->ll_kset.kobj,
1734                                   sbi->ll_dt_obd->obd_type->typ_name);
1735
1736         if (sbi->ll_md_obd)
1737                 sysfs_remove_link(&sbi->ll_kset.kobj,
1738                                   sbi->ll_md_obd->obd_type->typ_name);
1739
1740         kobject_put(lsi->lsi_kobj);
1741
1742         kset_unregister(&sbi->ll_kset);
1743         wait_for_completion(&sbi->ll_kobj_unregister);
1744
1745         lprocfs_free_stats(&sbi->ll_ra_stats);
1746         lprocfs_free_stats(&sbi->ll_stats);
1747 }
1748 #undef MAX_STRING_SIZE
1749
1750 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1751                                    struct seq_file *seq, int which)
1752 {
1753         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1754         unsigned long start, end, r, w;
1755         char *unitp = "KMGTPEZY";
1756         int i, units = 10;
1757         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1758
1759         read_cum = 0;
1760         write_cum = 0;
1761         start = 0;
1762
1763         for(i = 0; i < LL_HIST_MAX; i++) {
1764                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1765                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1766         }
1767
1768         for(i = 0; i < LL_HIST_MAX; i++) {
1769                 r = pp_info->pp_r_hist.oh_buckets[i];
1770                 w = pp_info->pp_w_hist.oh_buckets[i];
1771                 read_cum += r;
1772                 write_cum += w;
1773                 end = BIT(i + LL_HIST_START - units);
1774                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1775                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1776                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1777                            r, pct(r, read_tot), pct(read_cum, read_tot),
1778                            w, pct(w, write_tot), pct(write_cum, write_tot));
1779                 start = end;
1780                 if (start == BIT(10)) {
1781                         start = 1;
1782                         units += 10;
1783                         unitp++;
1784                 }
1785                 if (read_cum == read_tot && write_cum == write_tot)
1786                         break;
1787         }
1788 }
1789
1790 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1791 {
1792         struct timespec64 now;
1793         struct ll_sb_info *sbi = seq->private;
1794         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1795         int k;
1796
1797         ktime_get_real_ts64(&now);
1798
1799         if (!sbi->ll_rw_stats_on) {
1800                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1801                 return 0;
1802         }
1803         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1804                    (s64)now.tv_sec, now.tv_nsec);
1805         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1806         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1807                    "extents", "calls", "%", "cum%",
1808                    "calls", "%", "cum%");
1809         spin_lock(&sbi->ll_pp_extent_lock);
1810         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1811                 if (io_extents->pp_extents[k].pid != 0) {
1812                         seq_printf(seq, "\nPID: %d\n",
1813                                    io_extents->pp_extents[k].pid);
1814                         ll_display_extents_info(io_extents, seq, k);
1815                 }
1816         }
1817         spin_unlock(&sbi->ll_pp_extent_lock);
1818         return 0;
1819 }
1820
1821 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1822                                                 const char __user *buf,
1823                                                 size_t len,
1824                                                 loff_t *off)
1825 {
1826         struct seq_file *seq = file->private_data;
1827         struct ll_sb_info *sbi = seq->private;
1828         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1829         int i;
1830         __s64 value;
1831
1832         if (len == 0)
1833                 return -EINVAL;
1834
1835         value = ll_stats_pid_write(buf, len);
1836
1837         if (value == 0)
1838                 sbi->ll_rw_stats_on = 0;
1839         else
1840                 sbi->ll_rw_stats_on = 1;
1841
1842         spin_lock(&sbi->ll_pp_extent_lock);
1843         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1844                 io_extents->pp_extents[i].pid = 0;
1845                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1846                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1847         }
1848         spin_unlock(&sbi->ll_pp_extent_lock);
1849         return len;
1850 }
1851
1852 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1853
1854 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1855 {
1856         struct timespec64 now;
1857         struct ll_sb_info *sbi = seq->private;
1858         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1859
1860         ktime_get_real_ts64(&now);
1861
1862         if (!sbi->ll_rw_stats_on) {
1863                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1864                 return 0;
1865         }
1866         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1867                    (s64)now.tv_sec, now.tv_nsec);
1868
1869         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1870         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1871                    "extents", "calls", "%", "cum%",
1872                    "calls", "%", "cum%");
1873         spin_lock(&sbi->ll_lock);
1874         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1875         spin_unlock(&sbi->ll_lock);
1876
1877         return 0;
1878 }
1879
1880 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1881                                              const char __user *buf,
1882                                              size_t len, loff_t *off)
1883 {
1884         struct seq_file *seq = file->private_data;
1885         struct ll_sb_info *sbi = seq->private;
1886         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1887         int i;
1888         __s64 value;
1889
1890         if (len == 0)
1891                 return -EINVAL;
1892
1893         value = ll_stats_pid_write(buf, len);
1894
1895         if (value == 0)
1896                 sbi->ll_rw_stats_on = 0;
1897         else
1898                 sbi->ll_rw_stats_on = 1;
1899
1900         spin_lock(&sbi->ll_pp_extent_lock);
1901         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1902                 io_extents->pp_extents[i].pid = 0;
1903                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1904                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1905         }
1906         spin_unlock(&sbi->ll_pp_extent_lock);
1907
1908         return len;
1909 }
1910
1911 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1912
1913 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1914                        struct ll_file_data *file, loff_t pos,
1915                        size_t count, int rw)
1916 {
1917         int i, cur = -1;
1918         struct ll_rw_process_info *process;
1919         struct ll_rw_process_info *offset;
1920         int *off_count = &sbi->ll_rw_offset_entry_count;
1921         int *process_count = &sbi->ll_offset_process_count;
1922         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1923
1924         if(!sbi->ll_rw_stats_on)
1925                 return;
1926         process = sbi->ll_rw_process_info;
1927         offset = sbi->ll_rw_offset_info;
1928
1929         spin_lock(&sbi->ll_pp_extent_lock);
1930         /* Extent statistics */
1931         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1932                 if(io_extents->pp_extents[i].pid == pid) {
1933                         cur = i;
1934                         break;
1935                 }
1936         }
1937
1938         if (cur == -1) {
1939                 /* new process */
1940                 sbi->ll_extent_process_count =
1941                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1942                 cur = sbi->ll_extent_process_count;
1943                 io_extents->pp_extents[cur].pid = pid;
1944                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1945                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1946         }
1947
1948         for (i = 0; (count >= BIT(LL_HIST_START + i)) &&
1949              (i < (LL_HIST_MAX - 1)); i++);
1950         if (rw == 0) {
1951                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1952                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1953         } else {
1954                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1955                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1956         }
1957         spin_unlock(&sbi->ll_pp_extent_lock);
1958
1959         spin_lock(&sbi->ll_process_lock);
1960         /* Offset statistics */
1961         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1962                 if (process[i].rw_pid == pid) {
1963                         if (process[i].rw_last_file != file) {
1964                                 process[i].rw_range_start = pos;
1965                                 process[i].rw_last_file_pos = pos + count;
1966                                 process[i].rw_smallest_extent = count;
1967                                 process[i].rw_largest_extent = count;
1968                                 process[i].rw_offset = 0;
1969                                 process[i].rw_last_file = file;
1970                                 spin_unlock(&sbi->ll_process_lock);
1971                                 return;
1972                         }
1973                         if (process[i].rw_last_file_pos != pos) {
1974                                 *off_count =
1975                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1976                                 offset[*off_count].rw_op = process[i].rw_op;
1977                                 offset[*off_count].rw_pid = pid;
1978                                 offset[*off_count].rw_range_start =
1979                                         process[i].rw_range_start;
1980                                 offset[*off_count].rw_range_end =
1981                                         process[i].rw_last_file_pos;
1982                                 offset[*off_count].rw_smallest_extent =
1983                                         process[i].rw_smallest_extent;
1984                                 offset[*off_count].rw_largest_extent =
1985                                         process[i].rw_largest_extent;
1986                                 offset[*off_count].rw_offset =
1987                                         process[i].rw_offset;
1988                                 process[i].rw_op = rw;
1989                                 process[i].rw_range_start = pos;
1990                                 process[i].rw_smallest_extent = count;
1991                                 process[i].rw_largest_extent = count;
1992                                 process[i].rw_offset = pos -
1993                                         process[i].rw_last_file_pos;
1994                         }
1995                         if(process[i].rw_smallest_extent > count)
1996                                 process[i].rw_smallest_extent = count;
1997                         if(process[i].rw_largest_extent < count)
1998                                 process[i].rw_largest_extent = count;
1999                         process[i].rw_last_file_pos = pos + count;
2000                         spin_unlock(&sbi->ll_process_lock);
2001                         return;
2002                 }
2003         }
2004         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2005         process[*process_count].rw_pid = pid;
2006         process[*process_count].rw_op = rw;
2007         process[*process_count].rw_range_start = pos;
2008         process[*process_count].rw_last_file_pos = pos + count;
2009         process[*process_count].rw_smallest_extent = count;
2010         process[*process_count].rw_largest_extent = count;
2011         process[*process_count].rw_offset = 0;
2012         process[*process_count].rw_last_file = file;
2013         spin_unlock(&sbi->ll_process_lock);
2014 }
2015
2016 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2017 {
2018         struct timespec64 now;
2019         struct ll_sb_info *sbi = seq->private;
2020         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2021         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2022         int i;
2023
2024         ktime_get_real_ts64(&now);
2025
2026         if (!sbi->ll_rw_stats_on) {
2027                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2028                 return 0;
2029         }
2030         spin_lock(&sbi->ll_process_lock);
2031
2032         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
2033                    (s64)now.tv_sec, now.tv_nsec);
2034         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2035                    "R/W", "PID", "RANGE START", "RANGE END",
2036                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2037
2038         /* We stored the discontiguous offsets here; print them first */
2039         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2040                 if (offset[i].rw_pid != 0)
2041                         seq_printf(seq,
2042                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2043                                    offset[i].rw_op == READ ? 'R' : 'W',
2044                                    offset[i].rw_pid,
2045                                    offset[i].rw_range_start,
2046                                    offset[i].rw_range_end,
2047                                    (unsigned long)offset[i].rw_smallest_extent,
2048                                    (unsigned long)offset[i].rw_largest_extent,
2049                                    offset[i].rw_offset);
2050         }
2051
2052         /* Then print the current offsets for each process */
2053         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2054                 if (process[i].rw_pid != 0)
2055                         seq_printf(seq,
2056                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2057                                    process[i].rw_op == READ ? 'R' : 'W',
2058                                    process[i].rw_pid,
2059                                    process[i].rw_range_start,
2060                                    process[i].rw_last_file_pos,
2061                                    (unsigned long)process[i].rw_smallest_extent,
2062                                    (unsigned long)process[i].rw_largest_extent,
2063                                    process[i].rw_offset);
2064         }
2065         spin_unlock(&sbi->ll_process_lock);
2066
2067         return 0;
2068 }
2069
2070 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2071                                             const char __user *buf,
2072                                             size_t len, loff_t *off)
2073 {
2074         struct seq_file *seq = file->private_data;
2075         struct ll_sb_info *sbi = seq->private;
2076         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2077         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2078         __s64 value;
2079
2080         if (len == 0)
2081                 return -EINVAL;
2082
2083         value = ll_stats_pid_write(buf, len);
2084
2085         if (value == 0)
2086                 sbi->ll_rw_stats_on = 0;
2087         else
2088                 sbi->ll_rw_stats_on = 1;
2089
2090         spin_lock(&sbi->ll_process_lock);
2091         sbi->ll_offset_process_count = 0;
2092         sbi->ll_rw_offset_entry_count = 0;
2093         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2094                LL_PROCESS_HIST_MAX);
2095         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2096                LL_OFFSET_HIST_MAX);
2097         spin_unlock(&sbi->ll_process_lock);
2098
2099         return len;
2100 }
2101
2102 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);