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         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                    atomic_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 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 scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_sa_running_max);
 733 }
 734
 735 static ssize_t statahead_running_max_store(struct kobject *kobj,
 736                                            struct attribute *attr,
 737                                            const char *buffer,
 738                                            size_t count)
 739 {
 740         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 741                                               ll_kset.kobj);
 742         unsigned long val;
 743         int rc;
 744
 745         rc = kstrtoul(buffer, 0, &val);
 746         if (rc)
 747                 return rc;
 748
 749         if (val <= LL_SA_RUNNING_MAX) {
 750                 sbi->ll_sa_running_max = val;
 751                 return count;
 752         }
 753
 754         CERROR("Bad statahead_running_max value %lu. Valid values "
 755                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 756
 757         return -ERANGE;
 758 }
 759 LUSTRE_RW_ATTR(statahead_running_max);
 760
 761 static ssize_t statahead_max_show(struct kobject *kobj,
 762                                   struct attribute *attr,
 763                                   char *buf)
 764 {
 765         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 766                                               ll_kset.kobj);
 767
 768         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 769 }
 770
 771 static ssize_t statahead_max_store(struct kobject *kobj,
 772                                    struct attribute *attr,
 773                                    const char *buffer,
 774                                    size_t count)
 775 {
 776         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 777                                               ll_kset.kobj);
 778         unsigned long val;
 779         int rc;
 780
 781         rc = kstrtoul(buffer, 0, &val);
 782         if (rc)
 783                 return rc;
 784
 785         if (val <= LL_SA_RPC_MAX)
 786                 sbi->ll_sa_max = val;
 787         else
 788                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 789                        val, LL_SA_RPC_MAX);
 790
 791         return count;
 792 }
 793 LUSTRE_RW_ATTR(statahead_max);
 794
 795 static ssize_t statahead_agl_show(struct kobject *kobj,
 796                                   struct attribute *attr,
 797                                   char *buf)
 798 {
 799         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 800                                               ll_kset.kobj);
 801
 802         return 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 scnprintf(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 scnprintf(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 scnprintf(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 *const 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 scnprintf(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 scnprintf(buf, PAGE_SIZE, "%lu\n", PAGES_TO_MiB(
1143                          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 read_ahead_range_kb_show(struct kobject *kobj,
1177                                         struct attribute *attr,char *buf)
1178 {
1179         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1180                                               ll_kset.kobj);
1181
1182         return snprintf(buf, PAGE_SIZE, "%lu\n",
1183                         sbi->ll_ra_info.ra_range_pages << (PAGE_SHIFT - 10));
1184 }
1185
1186 static ssize_t
1187 read_ahead_range_kb_store(struct kobject *kobj,
1188                                struct attribute *attr,
1189                                const char *buffer, size_t count)
1190 {
1191         unsigned long pages_number;
1192         unsigned long max_ra_per_file;
1193         u64 val;
1194         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1195                                               ll_kset.kobj);
1196         int rc;
1197
1198         rc = sysfs_memparse(buffer, count, &val, "KiB");
1199         if (rc < 0)
1200                 return rc;
1201
1202         pages_number = val >> PAGE_SHIFT;
1203         /* Disable mmap range read */
1204         if (pages_number == 0)
1205                 goto out;
1206
1207         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1208         if (pages_number > max_ra_per_file ||
1209             pages_number < RA_MIN_MMAP_RANGE_PAGES)
1210                 return -ERANGE;
1211
1212 out:
1213         spin_lock(&sbi->ll_lock);
1214         sbi->ll_ra_info.ra_range_pages = pages_number;
1215         spin_unlock(&sbi->ll_lock);
1216
1217         return count;
1218 }
1219 LUSTRE_RW_ATTR(read_ahead_range_kb);
1220
1221 static ssize_t fast_read_show(struct kobject *kobj,
1222                               struct attribute *attr,
1223                               char *buf)
1224 {
1225         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1226                                               ll_kset.kobj);
1227
1228         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1229 }
1230
1231 static ssize_t fast_read_store(struct kobject *kobj,
1232                                struct attribute *attr,
1233                                const char *buffer,
1234                                size_t count)
1235 {
1236         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1237                                               ll_kset.kobj);
1238         bool val;
1239         int rc;
1240
1241         rc = kstrtobool(buffer, &val);
1242         if (rc)
1243                 return rc;
1244
1245         spin_lock(&sbi->ll_lock);
1246         if (val)
1247                 sbi->ll_flags |= LL_SBI_FAST_READ;
1248         else
1249                 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1250         spin_unlock(&sbi->ll_lock);
1251
1252         return count;
1253 }
1254 LUSTRE_RW_ATTR(fast_read);
1255
1256 static ssize_t file_heat_show(struct kobject *kobj,
1257                               struct attribute *attr,
1258                               char *buf)
1259 {
1260         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1261                                               ll_kset.kobj);
1262
1263         return scnprintf(buf, PAGE_SIZE, "%u\n",
1264                          !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1265 }
1266
1267 static ssize_t file_heat_store(struct kobject *kobj,
1268                                struct attribute *attr,
1269                                const char *buffer,
1270                                size_t count)
1271 {
1272         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1273                                               ll_kset.kobj);
1274         bool val;
1275         int rc;
1276
1277         rc = kstrtobool(buffer, &val);
1278         if (rc)
1279                 return rc;
1280
1281         spin_lock(&sbi->ll_lock);
1282         if (val)
1283                 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1284         else
1285                 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1286         spin_unlock(&sbi->ll_lock);
1287
1288         return count;
1289 }
1290 LUSTRE_RW_ATTR(file_heat);
1291
1292 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1293                                           struct attribute *attr,
1294                                           char *buf)
1295 {
1296         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1297                                               ll_kset.kobj);
1298
1299         return scnprintf(buf, PAGE_SIZE, "%u\n",
1300                          (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1301 }
1302
1303 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1304                                            struct attribute *attr,
1305                                            const char *buffer,
1306                                            size_t count)
1307 {
1308         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1309                                               ll_kset.kobj);
1310         unsigned long val;
1311         int rc;
1312
1313         rc = kstrtoul(buffer, 10, &val);
1314         if (rc)
1315                 return rc;
1316
1317         if (val < 0 || val > 100)
1318                 return -ERANGE;
1319
1320         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1321
1322         return count;
1323 }
1324 LUSTRE_RW_ATTR(heat_decay_percentage);
1325
1326 static ssize_t heat_period_second_show(struct kobject *kobj,
1327                                        struct attribute *attr,
1328                                        char *buf)
1329 {
1330         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1331                                               ll_kset.kobj);
1332
1333         return scnprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1334 }
1335
1336 static ssize_t heat_period_second_store(struct kobject *kobj,
1337                                         struct attribute *attr,
1338                                         const char *buffer,
1339                                         size_t count)
1340 {
1341         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1342                                               ll_kset.kobj);
1343         unsigned long val;
1344         int rc;
1345
1346         rc = kstrtoul(buffer, 10, &val);
1347         if (rc)
1348                 return rc;
1349
1350         if (val <= 0)
1351                 return -ERANGE;
1352
1353         sbi->ll_heat_period_second = val;
1354
1355         return count;
1356 }
1357 LUSTRE_RW_ATTR(heat_period_second);
1358
1359 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1360 {
1361         struct super_block      *sb    = m->private;
1362         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1363         struct cl_client_cache  *cache = sbi->ll_cache;
1364         long pages;
1365         int mb;
1366
1367         pages = atomic_long_read(&cache->ccc_unstable_nr);
1368         mb    = (pages * PAGE_SIZE) >> 20;
1369
1370         seq_printf(m, "unstable_check:     %8d\n"
1371                       "unstable_pages: %12ld\n"
1372                       "unstable_mb:        %8d\n",
1373                    cache->ccc_unstable_check, pages, mb);
1374         return 0;
1375 }
1376
1377 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1378                                            const char __user *buffer,
1379                                            size_t count, loff_t *unused)
1380 {
1381         struct seq_file *seq = file->private_data;
1382         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1383         char kernbuf[128];
1384         bool val;
1385         int rc;
1386
1387         if (count == 0)
1388                 return 0;
1389         if (count >= sizeof(kernbuf))
1390                 return -EINVAL;
1391
1392         if (copy_from_user(kernbuf, buffer, count))
1393                 return -EFAULT;
1394         kernbuf[count] = 0;
1395
1396         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1397                   kernbuf;
1398         rc = kstrtobool_from_user(buffer, count, &val);
1399         if (rc < 0)
1400                 return rc;
1401
1402         /* borrow lru lock to set the value */
1403         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1404         sbi->ll_cache->ccc_unstable_check = val;
1405         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1406
1407         return count;
1408 }
1409
1410 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1411
1412 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1413 {
1414         struct super_block *sb = m->private;
1415         struct ll_sb_info *sbi = ll_s2sbi(sb);
1416         struct root_squash_info *squash = &sbi->ll_squash;
1417
1418         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1419         return 0;
1420 }
1421
1422 static ssize_t ll_root_squash_seq_write(struct file *file,
1423                                         const char __user *buffer,
1424                                         size_t count, loff_t *off)
1425 {
1426         struct seq_file *m = file->private_data;
1427         struct super_block *sb = m->private;
1428         struct ll_sb_info *sbi = ll_s2sbi(sb);
1429         struct root_squash_info *squash = &sbi->ll_squash;
1430
1431         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1432 }
1433
1434 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1435
1436 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1437 {
1438         struct super_block *sb = m->private;
1439         struct ll_sb_info *sbi = ll_s2sbi(sb);
1440         struct root_squash_info *squash = &sbi->ll_squash;
1441         int len;
1442
1443         spin_lock(&squash->rsi_lock);
1444         if (!list_empty(&squash->rsi_nosquash_nids)) {
1445                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1446                                         &squash->rsi_nosquash_nids);
1447                 m->count += len;
1448                 seq_putc(m, '\n');
1449         } else {
1450                 seq_puts(m, "NONE\n");
1451         }
1452         spin_unlock(&squash->rsi_lock);
1453
1454         return 0;
1455 }
1456
1457 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1458                                           const char __user *buffer,
1459                                           size_t count, loff_t *off)
1460 {
1461         struct seq_file *m = file->private_data;
1462         struct super_block *sb = m->private;
1463         struct ll_sb_info *sbi = ll_s2sbi(sb);
1464         struct root_squash_info *squash = &sbi->ll_squash;
1465         int rc;
1466
1467         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1468         if (rc < 0)
1469                 return rc;
1470
1471         ll_compute_rootsquash_state(sbi);
1472
1473         return rc;
1474 }
1475
1476 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1477
1478 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1479 {
1480         struct super_block *sb = m->private;
1481         struct ll_sb_info *sbi = ll_s2sbi(sb);
1482
1483         return pcc_super_dump(&sbi->ll_pcc_super, m);
1484 }
1485
1486 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1487                                 size_t count, loff_t *off)
1488 {
1489         struct seq_file *m = file->private_data;
1490         struct super_block *sb = m->private;
1491         struct ll_sb_info *sbi = ll_s2sbi(sb);
1492         int rc;
1493         char *kernbuf;
1494
1495         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1496                 return -EINVAL;
1497
1498         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1499                 return -EOPNOTSUPP;
1500
1501         OBD_ALLOC(kernbuf, count + 1);
1502         if (kernbuf == NULL)
1503                 return -ENOMEM;
1504
1505         if (copy_from_user(kernbuf, buffer, count))
1506                 GOTO(out_free_kernbuff, rc = -EFAULT);
1507
1508         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1509 out_free_kernbuff:
1510         OBD_FREE(kernbuf, count + 1);
1511         return rc ? rc : count;
1512 }
1513 LDEBUGFS_SEQ_FOPS(ll_pcc);
1514
1515 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1516         { .name =       "site",
1517           .fops =       &ll_site_stats_fops                     },
1518         { .name =       "max_cached_mb",
1519           .fops =       &ll_max_cached_mb_fops                  },
1520         { .name =       "statahead_stats",
1521           .fops =       &ll_statahead_stats_fops                },
1522         { .name =       "unstable_stats",
1523           .fops =       &ll_unstable_stats_fops                 },
1524         { .name =       "sbi_flags",
1525           .fops =       &ll_sbi_flags_fops                      },
1526         { .name =       "root_squash",
1527           .fops =       &ll_root_squash_fops                    },
1528         { .name =       "nosquash_nids",
1529           .fops =       &ll_nosquash_nids_fops                  },
1530         { .name =       "pcc",
1531           .fops =       &ll_pcc_fops,                           },
1532         { NULL }
1533 };
1534
1535 #define MAX_STRING_SIZE 128
1536
1537 static struct attribute *llite_attrs[] = {
1538         &lustre_attr_blocksize.attr,
1539         &lustre_attr_stat_blocksize.attr,
1540         &lustre_attr_kbytestotal.attr,
1541         &lustre_attr_kbytesfree.attr,
1542         &lustre_attr_kbytesavail.attr,
1543         &lustre_attr_filestotal.attr,
1544         &lustre_attr_filesfree.attr,
1545         &lustre_attr_client_type.attr,
1546         &lustre_attr_foreign_symlink_enable.attr,
1547         &lustre_attr_foreign_symlink_prefix.attr,
1548         &lustre_attr_foreign_symlink_upcall.attr,
1549         &lustre_attr_foreign_symlink_upcall_info.attr,
1550         &lustre_attr_fstype.attr,
1551         &lustre_attr_uuid.attr,
1552         &lustre_attr_checksums.attr,
1553         &lustre_attr_checksum_pages.attr,
1554         &lustre_attr_max_read_ahead_mb.attr,
1555         &lustre_attr_max_read_ahead_per_file_mb.attr,
1556         &lustre_attr_max_read_ahead_whole_mb.attr,
1557         &lustre_attr_max_read_ahead_async_active.attr,
1558         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1559         &lustre_attr_read_ahead_range_kb.attr,
1560         &lustre_attr_stats_track_pid.attr,
1561         &lustre_attr_stats_track_ppid.attr,
1562         &lustre_attr_stats_track_gid.attr,
1563         &lustre_attr_statahead_running_max.attr,
1564         &lustre_attr_statahead_max.attr,
1565         &lustre_attr_statahead_agl.attr,
1566         &lustre_attr_lazystatfs.attr,
1567         &lustre_attr_statfs_max_age.attr,
1568         &lustre_attr_max_easize.attr,
1569         &lustre_attr_default_easize.attr,
1570         &lustre_attr_xattr_cache.attr,
1571         &lustre_attr_fast_read.attr,
1572         &lustre_attr_tiny_write.attr,
1573         &lustre_attr_file_heat.attr,
1574         &lustre_attr_heat_decay_percentage.attr,
1575         &lustre_attr_heat_period_second.attr,
1576         NULL,
1577 };
1578
1579 static void sbi_kobj_release(struct kobject *kobj)
1580 {
1581         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1582                                               ll_kset.kobj);
1583         complete(&sbi->ll_kobj_unregister);
1584 }
1585
1586 static struct kobj_type sbi_ktype = {
1587         .default_attrs  = llite_attrs,
1588         .sysfs_ops      = &lustre_sysfs_ops,
1589         .release        = sbi_kobj_release,
1590 };
1591
1592 static const struct llite_file_opcode {
1593         __u32           opcode;
1594         __u32           type;
1595         const char      *opname;
1596 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1597         /* file operation */
1598         { LPROC_LL_READ_BYTES,  LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1599         { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1600         { LPROC_LL_READ,        LPROCFS_TYPE_LATENCY,   "read" },
1601         { LPROC_LL_WRITE,       LPROCFS_TYPE_LATENCY,   "write" },
1602         { LPROC_LL_IOCTL,       LPROCFS_TYPE_REQS,      "ioctl" },
1603         { LPROC_LL_OPEN,        LPROCFS_TYPE_LATENCY,   "open" },
1604         { LPROC_LL_RELEASE,     LPROCFS_TYPE_LATENCY,   "close" },
1605         { LPROC_LL_MMAP,        LPROCFS_TYPE_LATENCY,   "mmap" },
1606         { LPROC_LL_FAULT,       LPROCFS_TYPE_LATENCY,   "page_fault" },
1607         { LPROC_LL_MKWRITE,     LPROCFS_TYPE_LATENCY,   "page_mkwrite" },
1608         { LPROC_LL_LLSEEK,      LPROCFS_TYPE_LATENCY,   "seek" },
1609         { LPROC_LL_FSYNC,       LPROCFS_TYPE_LATENCY,   "fsync" },
1610         { LPROC_LL_READDIR,     LPROCFS_TYPE_LATENCY,   "readdir" },
1611         /* inode operation */
1612         { LPROC_LL_SETATTR,     LPROCFS_TYPE_LATENCY,   "setattr" },
1613         { LPROC_LL_TRUNC,       LPROCFS_TYPE_LATENCY,   "truncate" },
1614         { LPROC_LL_FLOCK,       LPROCFS_TYPE_LATENCY,   "flock" },
1615         { LPROC_LL_GETATTR,     LPROCFS_TYPE_LATENCY,   "getattr" },
1616         { LPROC_LL_FALLOCATE,   LPROCFS_TYPE_LATENCY, "fallocate"},
1617         /* dir inode operation */
1618         { LPROC_LL_CREATE,      LPROCFS_TYPE_LATENCY,   "create" },
1619         { LPROC_LL_LINK,        LPROCFS_TYPE_LATENCY,   "link" },
1620         { LPROC_LL_UNLINK,      LPROCFS_TYPE_LATENCY,   "unlink" },
1621         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_LATENCY,   "symlink" },
1622         { LPROC_LL_MKDIR,       LPROCFS_TYPE_LATENCY,   "mkdir" },
1623         { LPROC_LL_RMDIR,       LPROCFS_TYPE_LATENCY,   "rmdir" },
1624         { LPROC_LL_MKNOD,       LPROCFS_TYPE_LATENCY,   "mknod" },
1625         { LPROC_LL_RENAME,      LPROCFS_TYPE_LATENCY,   "rename" },
1626         /* special inode operation */
1627         { LPROC_LL_STATFS,      LPROCFS_TYPE_LATENCY,   "statfs" },
1628         { LPROC_LL_SETXATTR,    LPROCFS_TYPE_LATENCY,   "setxattr" },
1629         { LPROC_LL_GETXATTR,    LPROCFS_TYPE_LATENCY,   "getxattr" },
1630         { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS,    "getxattr_hits" },
1631         { LPROC_LL_LISTXATTR,   LPROCFS_TYPE_LATENCY,   "listxattr" },
1632         { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY,   "removexattr" },
1633         { LPROC_LL_INODE_PERM,  LPROCFS_TYPE_LATENCY,   "inode_permission" },
1634 };
1635
1636 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1637 {
1638         if (!sbi->ll_stats)
1639                 return;
1640
1641         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1642                 lprocfs_counter_add(sbi->ll_stats, op, count);
1643         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1644                  sbi->ll_stats_track_id == current->pid)
1645                 lprocfs_counter_add(sbi->ll_stats, op, count);
1646         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1647                  sbi->ll_stats_track_id == current->parent->pid)
1648                 lprocfs_counter_add(sbi->ll_stats, op, count);
1649         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1650                  sbi->ll_stats_track_id ==
1651                         from_kgid(&init_user_ns, current_gid()))
1652                 lprocfs_counter_add(sbi->ll_stats, op, count);
1653 }
1654 EXPORT_SYMBOL(ll_stats_ops_tally);
1655
1656 static const char *const ra_stat_string[] = {
1657         [RA_STAT_HIT] = "hits",
1658         [RA_STAT_MISS] = "misses",
1659         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1660         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1661         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1662         [RA_STAT_FAILED_MATCH] = "failed lock match",
1663         [RA_STAT_DISCARDED] = "read but discarded",
1664         [RA_STAT_ZERO_LEN] = "zero length file",
1665         [RA_STAT_ZERO_WINDOW] = "zero size window",
1666         [RA_STAT_EOF] = "read-ahead to EOF",
1667         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1668         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1669         [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1670         [RA_STAT_ASYNC] = "async readahead",
1671         [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1672         [RA_STAT_MMAP_RANGE_READ] = "mmap range read",
1673 };
1674
1675 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1676 {
1677         struct lustre_sb_info *lsi = s2lsi(sb);
1678         struct ll_sb_info *sbi = ll_s2sbi(sb);
1679         int err, id;
1680
1681         ENTRY;
1682         LASSERT(sbi);
1683
1684         if (IS_ERR_OR_NULL(llite_root))
1685                 goto out_ll_kset;
1686
1687         sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1688         ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1689
1690         debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1691                             &vvp_dump_pgcache_file_ops);
1692
1693         debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1694                                  &ll_rw_extents_stats_fops);
1695
1696         debugfs_create_file("extents_stats_per_process", 0644,
1697                             sbi->ll_debugfs_entry, sbi,
1698                             &ll_rw_extents_stats_pp_fops);
1699
1700         debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1701                             &ll_rw_offset_stats_fops);
1702
1703         /* File operations stats */
1704         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1705                                             LPROCFS_STATS_FLAG_NONE);
1706         if (sbi->ll_stats == NULL)
1707                 GOTO(out_debugfs, err = -ENOMEM);
1708
1709         /* do counter init */
1710         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1711                 u32 type = llite_opcode_table[id].type;
1712                 void *ptr = "unknown";
1713
1714                 if (type & LPROCFS_TYPE_REQS)
1715                         ptr = "reqs";
1716                 else if (type & LPROCFS_TYPE_BYTES)
1717                         ptr = "bytes";
1718                 else if (type & LPROCFS_TYPE_USEC)
1719                         ptr = "usec";
1720                 lprocfs_counter_init(sbi->ll_stats,
1721                                      llite_opcode_table[id].opcode, type,
1722                                      llite_opcode_table[id].opname, ptr);
1723         }
1724
1725         debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1726                             sbi->ll_stats, &ldebugfs_stats_seq_fops);
1727
1728         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1729                                                LPROCFS_STATS_FLAG_NONE);
1730         if (sbi->ll_ra_stats == NULL)
1731                 GOTO(out_stats, err = -ENOMEM);
1732
1733         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1734                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1735                                      ra_stat_string[id], "pages");
1736
1737         debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1738                             sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1739
1740 out_ll_kset:
1741         /* Yes we also register sysfs mount kset here as well */
1742         sbi->ll_kset.kobj.parent = llite_kobj;
1743         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1744         init_completion(&sbi->ll_kobj_unregister);
1745         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1746         if (err)
1747                 GOTO(out_ra_stats, err);
1748
1749         err = kset_register(&sbi->ll_kset);
1750         if (err)
1751                 GOTO(out_ra_stats, err);
1752
1753         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1754
1755         RETURN(0);
1756 out_ra_stats:
1757         lprocfs_free_stats(&sbi->ll_ra_stats);
1758 out_stats:
1759         lprocfs_free_stats(&sbi->ll_stats);
1760 out_debugfs:
1761         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1762
1763         RETURN(err);
1764 }
1765
1766 void ll_debugfs_unregister_super(struct super_block *sb)
1767 {
1768         struct lustre_sb_info *lsi = s2lsi(sb);
1769         struct ll_sb_info *sbi = ll_s2sbi(sb);
1770
1771         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1772
1773         if (sbi->ll_dt_obd)
1774                 sysfs_remove_link(&sbi->ll_kset.kobj,
1775                                   sbi->ll_dt_obd->obd_type->typ_name);
1776
1777         if (sbi->ll_md_obd)
1778                 sysfs_remove_link(&sbi->ll_kset.kobj,
1779                                   sbi->ll_md_obd->obd_type->typ_name);
1780
1781         kobject_put(lsi->lsi_kobj);
1782
1783         kset_unregister(&sbi->ll_kset);
1784         wait_for_completion(&sbi->ll_kobj_unregister);
1785
1786         lprocfs_free_stats(&sbi->ll_ra_stats);
1787         lprocfs_free_stats(&sbi->ll_stats);
1788 }
1789 #undef MAX_STRING_SIZE
1790
1791 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1792                                    struct seq_file *seq, int which)
1793 {
1794         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1795         unsigned long start, end, r, w;
1796         char *unitp = "KMGTPEZY";
1797         int i, units = 10;
1798         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1799
1800         read_cum = 0;
1801         write_cum = 0;
1802         start = 0;
1803
1804         for(i = 0; i < LL_HIST_MAX; i++) {
1805                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1806                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1807         }
1808
1809         for(i = 0; i < LL_HIST_MAX; i++) {
1810                 r = pp_info->pp_r_hist.oh_buckets[i];
1811                 w = pp_info->pp_w_hist.oh_buckets[i];
1812                 read_cum += r;
1813                 write_cum += w;
1814                 end = 1 << (i + LL_HIST_START - units);
1815                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1816                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1817                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1818                            r, pct(r, read_tot), pct(read_cum, read_tot),
1819                            w, pct(w, write_tot), pct(write_cum, write_tot));
1820                 start = end;
1821                 if (start == (1 << 10)) {
1822                         start = 1;
1823                         units += 10;
1824                         unitp++;
1825                 }
1826                 if (read_cum == read_tot && write_cum == write_tot)
1827                         break;
1828         }
1829 }
1830
1831 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1832 {
1833         struct timespec64 now;
1834         struct ll_sb_info *sbi = seq->private;
1835         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1836         int k;
1837
1838         ktime_get_real_ts64(&now);
1839
1840         if (!sbi->ll_rw_stats_on) {
1841                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1842                 return 0;
1843         }
1844         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1845                    (s64)now.tv_sec, now.tv_nsec);
1846         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1847         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1848                    "extents", "calls", "%", "cum%",
1849                    "calls", "%", "cum%");
1850         spin_lock(&sbi->ll_pp_extent_lock);
1851         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1852                 if (io_extents->pp_extents[k].pid != 0) {
1853                         seq_printf(seq, "\nPID: %d\n",
1854                                    io_extents->pp_extents[k].pid);
1855                         ll_display_extents_info(io_extents, seq, k);
1856                 }
1857         }
1858         spin_unlock(&sbi->ll_pp_extent_lock);
1859         return 0;
1860 }
1861
1862 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1863                                                 const char __user *buf,
1864                                                 size_t len,
1865                                                 loff_t *off)
1866 {
1867         struct seq_file *seq = file->private_data;
1868         struct ll_sb_info *sbi = seq->private;
1869         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1870         int i;
1871         __s64 value;
1872
1873         if (len == 0)
1874                 return -EINVAL;
1875
1876         value = ll_stats_pid_write(buf, len);
1877
1878         if (value == 0)
1879                 sbi->ll_rw_stats_on = 0;
1880         else
1881                 sbi->ll_rw_stats_on = 1;
1882
1883         spin_lock(&sbi->ll_pp_extent_lock);
1884         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1885                 io_extents->pp_extents[i].pid = 0;
1886                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1887                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1888         }
1889         spin_unlock(&sbi->ll_pp_extent_lock);
1890         return len;
1891 }
1892
1893 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1894
1895 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1896 {
1897         struct timespec64 now;
1898         struct ll_sb_info *sbi = seq->private;
1899         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1900
1901         ktime_get_real_ts64(&now);
1902
1903         if (!sbi->ll_rw_stats_on) {
1904                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1905                 return 0;
1906         }
1907         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1908                    (s64)now.tv_sec, now.tv_nsec);
1909
1910         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1911         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1912                    "extents", "calls", "%", "cum%",
1913                    "calls", "%", "cum%");
1914         spin_lock(&sbi->ll_lock);
1915         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1916         spin_unlock(&sbi->ll_lock);
1917
1918         return 0;
1919 }
1920
1921 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1922                                              const char __user *buf,
1923                                              size_t len, loff_t *off)
1924 {
1925         struct seq_file *seq = file->private_data;
1926         struct ll_sb_info *sbi = seq->private;
1927         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1928         int i;
1929         __s64 value;
1930
1931         if (len == 0)
1932                 return -EINVAL;
1933
1934         value = ll_stats_pid_write(buf, len);
1935
1936         if (value == 0)
1937                 sbi->ll_rw_stats_on = 0;
1938         else
1939                 sbi->ll_rw_stats_on = 1;
1940
1941         spin_lock(&sbi->ll_pp_extent_lock);
1942         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1943                 io_extents->pp_extents[i].pid = 0;
1944                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1945                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1946         }
1947         spin_unlock(&sbi->ll_pp_extent_lock);
1948
1949         return len;
1950 }
1951
1952 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1953
1954 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1955                        struct ll_file_data *file, loff_t pos,
1956                        size_t count, int rw)
1957 {
1958         int i, cur = -1;
1959         struct ll_rw_process_info *process;
1960         struct ll_rw_process_info *offset;
1961         int *off_count = &sbi->ll_rw_offset_entry_count;
1962         int *process_count = &sbi->ll_offset_process_count;
1963         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1964
1965         if(!sbi->ll_rw_stats_on)
1966                 return;
1967         process = sbi->ll_rw_process_info;
1968         offset = sbi->ll_rw_offset_info;
1969
1970         spin_lock(&sbi->ll_pp_extent_lock);
1971         /* Extent statistics */
1972         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1973                 if(io_extents->pp_extents[i].pid == pid) {
1974                         cur = i;
1975                         break;
1976                 }
1977         }
1978
1979         if (cur == -1) {
1980                 /* new process */
1981                 sbi->ll_extent_process_count =
1982                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1983                 cur = sbi->ll_extent_process_count;
1984                 io_extents->pp_extents[cur].pid = pid;
1985                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1986                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1987         }
1988
1989         for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1990              (i < (LL_HIST_MAX - 1)); i++);
1991         if (rw == 0) {
1992                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1993                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1994         } else {
1995                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1996                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1997         }
1998         spin_unlock(&sbi->ll_pp_extent_lock);
1999
2000         spin_lock(&sbi->ll_process_lock);
2001         /* Offset statistics */
2002         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2003                 if (process[i].rw_pid == pid) {
2004                         if (process[i].rw_last_file != file) {
2005                                 process[i].rw_range_start = pos;
2006                                 process[i].rw_last_file_pos = pos + count;
2007                                 process[i].rw_smallest_extent = count;
2008                                 process[i].rw_largest_extent = count;
2009                                 process[i].rw_offset = 0;
2010                                 process[i].rw_last_file = file;
2011                                 spin_unlock(&sbi->ll_process_lock);
2012                                 return;
2013                         }
2014                         if (process[i].rw_last_file_pos != pos) {
2015                                 *off_count =
2016                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
2017                                 offset[*off_count].rw_op = process[i].rw_op;
2018                                 offset[*off_count].rw_pid = pid;
2019                                 offset[*off_count].rw_range_start =
2020                                         process[i].rw_range_start;
2021                                 offset[*off_count].rw_range_end =
2022                                         process[i].rw_last_file_pos;
2023                                 offset[*off_count].rw_smallest_extent =
2024                                         process[i].rw_smallest_extent;
2025                                 offset[*off_count].rw_largest_extent =
2026                                         process[i].rw_largest_extent;
2027                                 offset[*off_count].rw_offset =
2028                                         process[i].rw_offset;
2029                                 process[i].rw_op = rw;
2030                                 process[i].rw_range_start = pos;
2031                                 process[i].rw_smallest_extent = count;
2032                                 process[i].rw_largest_extent = count;
2033                                 process[i].rw_offset = pos -
2034                                         process[i].rw_last_file_pos;
2035                         }
2036                         if(process[i].rw_smallest_extent > count)
2037                                 process[i].rw_smallest_extent = count;
2038                         if(process[i].rw_largest_extent < count)
2039                                 process[i].rw_largest_extent = count;
2040                         process[i].rw_last_file_pos = pos + count;
2041                         spin_unlock(&sbi->ll_process_lock);
2042                         return;
2043                 }
2044         }
2045         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
2046         process[*process_count].rw_pid = pid;
2047         process[*process_count].rw_op = rw;
2048         process[*process_count].rw_range_start = pos;
2049         process[*process_count].rw_last_file_pos = pos + count;
2050         process[*process_count].rw_smallest_extent = count;
2051         process[*process_count].rw_largest_extent = count;
2052         process[*process_count].rw_offset = 0;
2053         process[*process_count].rw_last_file = file;
2054         spin_unlock(&sbi->ll_process_lock);
2055 }
2056
2057 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
2058 {
2059         struct timespec64 now;
2060         struct ll_sb_info *sbi = seq->private;
2061         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2062         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2063         int i;
2064
2065         ktime_get_real_ts64(&now);
2066
2067         if (!sbi->ll_rw_stats_on) {
2068                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2069                 return 0;
2070         }
2071         spin_lock(&sbi->ll_process_lock);
2072
2073         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
2074                    (s64)now.tv_sec, now.tv_nsec);
2075         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2076                    "R/W", "PID", "RANGE START", "RANGE END",
2077                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2078
2079         /* We stored the discontiguous offsets here; print them first */
2080         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2081                 if (offset[i].rw_pid != 0)
2082                         seq_printf(seq,
2083                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2084                                    offset[i].rw_op == READ ? 'R' : 'W',
2085                                    offset[i].rw_pid,
2086                                    offset[i].rw_range_start,
2087                                    offset[i].rw_range_end,
2088                                    (unsigned long)offset[i].rw_smallest_extent,
2089                                    (unsigned long)offset[i].rw_largest_extent,
2090                                    offset[i].rw_offset);
2091         }
2092
2093         /* Then print the current offsets for each process */
2094         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2095                 if (process[i].rw_pid != 0)
2096                         seq_printf(seq,
2097                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2098                                    process[i].rw_op == READ ? 'R' : 'W',
2099                                    process[i].rw_pid,
2100                                    process[i].rw_range_start,
2101                                    process[i].rw_last_file_pos,
2102                                    (unsigned long)process[i].rw_smallest_extent,
2103                                    (unsigned long)process[i].rw_largest_extent,
2104                                    process[i].rw_offset);
2105         }
2106         spin_unlock(&sbi->ll_process_lock);
2107
2108         return 0;
2109 }
2110
2111 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2112                                             const char __user *buf,
2113                                             size_t len, loff_t *off)
2114 {
2115         struct seq_file *seq = file->private_data;
2116         struct ll_sb_info *sbi = seq->private;
2117         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2118         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2119         __s64 value;
2120
2121         if (len == 0)
2122                 return -EINVAL;
2123
2124         value = ll_stats_pid_write(buf, len);
2125
2126         if (value == 0)
2127                 sbi->ll_rw_stats_on = 0;
2128         else
2129                 sbi->ll_rw_stats_on = 1;
2130
2131         spin_lock(&sbi->ll_process_lock);
2132         sbi->ll_offset_process_count = 0;
2133         sbi->ll_rw_offset_entry_count = 0;
2134         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2135                LL_PROCESS_HIST_MAX);
2136         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2137                LL_OFFSET_HIST_MAX);
2138         spin_unlock(&sbi->ll_process_lock);
2139
2140         return len;
2141 }
2142
2143 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);