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 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 293                            char *buf)
 294 {
 295         return sprintf(buf, "lustre\n");
 296 }
 297 LUSTRE_RO_ATTR(fstype);
 298
 299 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 300                          char *buf)
 301 {
 302         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 303                                               ll_kset.kobj);
 304
 305         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 306 }
 307 LUSTRE_RO_ATTR(uuid);
 308
 309 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 310 {
 311         struct super_block *sb = m->private;
 312
 313         /*
 314          * See description of statistical counters in struct cl_site, and
 315          * struct lu_site.
 316          */
 317         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 318 }
 319
 320 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
 321
 322 static ssize_t max_read_ahead_mb_show(struct kobject *kobj,
 323                                       struct attribute *attr, char *buf)
 324 {
 325         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 326                                               ll_kset.kobj);
 327
 328         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 329                         PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 330 }
 331
 332 static ssize_t max_read_ahead_mb_store(struct kobject *kobj,
 333                                        struct attribute *attr,
 334                                        const char *buffer, size_t count)
 335 {
 336         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 337                                               ll_kset.kobj);
 338         u64 ra_max_mb, pages_number;
 339         int rc;
 340
 341         rc = sysfs_memparse(buffer, count, &ra_max_mb, "MiB");
 342         if (rc)
 343                 return rc;
 344
 345         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
 346         CDEBUG(D_INFO, "%s: set max_read_ahead_mb=%llu (%llu pages)\n",
 347                sbi->ll_fsname, PAGES_TO_MiB(pages_number), pages_number);
 348         if (pages_number > cfs_totalram_pages() / 2) {
 349                 /* 1/2 of RAM */
 350                 CERROR("%s: cannot set max_read_ahead_mb=%llu > totalram/2=%luMB\n",
 351                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 352                        PAGES_TO_MiB(cfs_totalram_pages() / 2));
 353                 return -ERANGE;
 354         }
 355
 356         spin_lock(&sbi->ll_lock);
 357         sbi->ll_ra_info.ra_max_pages = pages_number;
 358         spin_unlock(&sbi->ll_lock);
 359
 360         return count;
 361 }
 362 LUSTRE_RW_ATTR(max_read_ahead_mb);
 363
 364 static ssize_t max_read_ahead_per_file_mb_show(struct kobject *kobj,
 365                                                struct attribute *attr,
 366                                                char *buf)
 367 {
 368         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 369                                               ll_kset.kobj);
 370
 371         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 372                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 373 }
 374
 375 static ssize_t max_read_ahead_per_file_mb_store(struct kobject *kobj,
 376                                                 struct attribute *attr,
 377                                                 const char *buffer,
 378                                                 size_t count)
 379 {
 380         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 381                                               ll_kset.kobj);
 382         u64 ra_max_file_mb, pages_number;
 383         int rc;
 384
 385         rc = sysfs_memparse(buffer, count, &ra_max_file_mb, "MiB");
 386         if (rc)
 387                 return rc;
 388
 389         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
 390         if (pages_number > sbi->ll_ra_info.ra_max_pages) {
 391                 CERROR("%s: cannot set max_read_ahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
 392                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 393                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 394                 return -ERANGE;
 395         }
 396
 397         spin_lock(&sbi->ll_lock);
 398         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 399         spin_unlock(&sbi->ll_lock);
 400
 401         return count;
 402 }
 403 LUSTRE_RW_ATTR(max_read_ahead_per_file_mb);
 404
 405 static ssize_t max_read_ahead_whole_mb_show(struct kobject *kobj,
 406                                             struct attribute *attr, char *buf)
 407 {
 408         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 409                                               ll_kset.kobj);
 410
 411         return scnprintf(buf, PAGE_SIZE, "%lu\n",
 412                          PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages));
 413 }
 414
 415 static ssize_t max_read_ahead_whole_mb_store(struct kobject *kobj,
 416                                              struct attribute *attr,
 417                                              const char *buffer, size_t count)
 418 {
 419         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 420                                               ll_kset.kobj);
 421         u64 ra_max_whole_mb, pages_number;
 422         int rc;
 423
 424         rc = sysfs_memparse(buffer, count, &ra_max_whole_mb, "MiB");
 425         if (rc)
 426                 return rc;
 427
 428         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
 429         /* Cap this at the current max readahead window size, the readahead
 430          * algorithm does this anyway so it's pointless to set it larger.
 431          */
 432         if (pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 433                 CERROR("%s: cannot set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
 434                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 435                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 436
 437                 return -ERANGE;
 438         }
 439
 440         spin_lock(&sbi->ll_lock);
 441         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 442         spin_unlock(&sbi->ll_lock);
 443
 444         return count;
 445 }
 446 LUSTRE_RW_ATTR(max_read_ahead_whole_mb);
 447
 448 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 449 {
 450         struct super_block     *sb    = m->private;
 451         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 452         struct cl_client_cache *cache = sbi->ll_cache;
 453         long max_cached_mb;
 454         long unused_mb;
 455
 456         mutex_lock(&cache->ccc_max_cache_mb_lock);
 457         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
 458         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
 459         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 460         seq_printf(m, "users: %d\n"
 461                       "max_cached_mb: %ld\n"
 462                       "used_mb: %ld\n"
 463                       "unused_mb: %ld\n"
 464                       "reclaim_count: %u\n",
 465                    atomic_read(&cache->ccc_users),
 466                    max_cached_mb,
 467                    max_cached_mb - unused_mb,
 468                    unused_mb,
 469                    cache->ccc_lru_shrinkers);
 470         return 0;
 471 }
 472
 473 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
 474                                           const char __user *buffer,
 475                                           size_t count, loff_t *off)
 476 {
 477         struct seq_file *m = file->private_data;
 478         struct super_block *sb = m->private;
 479         struct ll_sb_info *sbi = ll_s2sbi(sb);
 480         struct cl_client_cache *cache = sbi->ll_cache;
 481         struct lu_env *env;
 482         long diff = 0;
 483         long nrpages = 0;
 484         __u16 refcheck;
 485         u64 pages_number;
 486         int rc;
 487         char kernbuf[128], *ptr;
 488
 489         ENTRY;
 490         if (count >= sizeof(kernbuf))
 491                 RETURN(-EINVAL);
 492
 493         if (copy_from_user(kernbuf, buffer, count))
 494                 RETURN(-EFAULT);
 495         kernbuf[count] = '\0';
 496
 497         ptr = lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count);
 498         rc = sysfs_memparse(ptr, count, &pages_number, "MiB");
 499         if (rc)
 500                 RETURN(rc);
 501
 502         pages_number >>= PAGE_SHIFT;
 503
 504         if (pages_number < 0 || pages_number > cfs_totalram_pages()) {
 505                 CERROR("%s: can't set max cache more than %lu MB\n",
 506                        sbi->ll_fsname,
 507                        PAGES_TO_MiB(cfs_totalram_pages()));
 508                 RETURN(-ERANGE);
 509         }
 510         /* Allow enough cache so clients can make well-formed RPCs */
 511         pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
 512
 513         mutex_lock(&cache->ccc_max_cache_mb_lock);
 514         diff = pages_number - cache->ccc_lru_max;
 515
 516         /* easy - add more LRU slots. */
 517         if (diff >= 0) {
 518                 atomic_long_add(diff, &cache->ccc_lru_left);
 519                 GOTO(out, rc = 0);
 520         }
 521
 522         env = cl_env_get(&refcheck);
 523         if (IS_ERR(env))
 524                 GOTO(out_unlock, rc = PTR_ERR(env));
 525
 526         diff = -diff;
 527         while (diff > 0) {
 528                 long tmp;
 529
 530                 /* reduce LRU budget from free slots. */
 531                 do {
 532                         long lru_left_old, lru_left_new, lru_left_ret;
 533
 534                         lru_left_old = atomic_long_read(&cache->ccc_lru_left);
 535                         if (lru_left_old == 0)
 536                                 break;
 537
 538                         lru_left_new = lru_left_old > diff ?
 539                                         lru_left_old - diff : 0;
 540                         lru_left_ret =
 541                                 atomic_long_cmpxchg(&cache->ccc_lru_left,
 542                                                     lru_left_old,
 543                                                     lru_left_new);
 544                         if (likely(lru_left_old == lru_left_ret)) {
 545                                 diff -= lru_left_old - lru_left_new;
 546                                 nrpages += lru_left_old - lru_left_new;
 547                                 break;
 548                         }
 549                 } while (1);
 550
 551                 if (diff <= 0)
 552                         break;
 553
 554                 if (sbi->ll_dt_exp == NULL) { /* being initialized */
 555                         rc = -ENODEV;
 556                         break;
 557                 }
 558
 559                 /* Request extra free slots to avoid them all being used
 560                  * by other processes before this can continue shrinking.
 561                  */
 562                 tmp = diff + min_t(long, diff, MiB_TO_PAGES(1024));
 563                 /* difficult - have to ask OSCs to drop LRU slots. */
 564                 rc = obd_set_info_async(env, sbi->ll_dt_exp,
 565                                 sizeof(KEY_CACHE_LRU_SHRINK),
 566                                 KEY_CACHE_LRU_SHRINK,
 567                                 sizeof(tmp), &tmp, NULL);
 568                 if (rc < 0)
 569                         break;
 570         }
 571         cl_env_put(env, &refcheck);
 572
 573 out:
 574         if (rc >= 0) {
 575                 cache->ccc_lru_max = pages_number;
 576                 rc = count;
 577         } else {
 578                 atomic_long_add(nrpages, &cache->ccc_lru_left);
 579         }
 580 out_unlock:
 581         mutex_unlock(&cache->ccc_max_cache_mb_lock);
 582         return rc;
 583 }
 584 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
 585
 586 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
 587                               char *buf)
 588 {
 589         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 590                                               ll_kset.kobj);
 591
 592         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 593 }
 594
 595 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
 596                                const char *buffer, size_t count)
 597 {
 598         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 599                                               ll_kset.kobj);
 600         bool val;
 601         int tmp;
 602         int rc;
 603
 604         if (!sbi->ll_dt_exp)
 605                 /* Not set up yet */
 606                 return -EAGAIN;
 607
 608         rc = kstrtobool(buffer, &val);
 609         if (rc)
 610                 return rc;
 611         if (val)
 612                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 613         else
 614                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 615         tmp = val;
 616
 617         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 618                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
 619         if (rc)
 620                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 621
 622         return count;
 623 }
 624 LUSTRE_RW_ATTR(checksums);
 625
 626 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
 627
 628 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 629                               enum stats_track_type type)
 630 {
 631         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 632                                               ll_kset.kobj);
 633
 634         if (sbi->ll_stats_track_type == type)
 635                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 636         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 637                 return sprintf(buf, "0 (all)\n");
 638
 639         return sprintf(buf, "untracked\n");
 640 }
 641
 642 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 643                               size_t count, enum stats_track_type type)
 644 {
 645         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 646                                               ll_kset.kobj);
 647         unsigned long pid;
 648         int rc;
 649
 650         rc = kstrtoul(buffer, 10, &pid);
 651         if (rc)
 652                 return rc;
 653
 654         sbi->ll_stats_track_id = pid;
 655         if (pid == 0)
 656                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 657         else
 658                 sbi->ll_stats_track_type = type;
 659         lprocfs_clear_stats(sbi->ll_stats);
 660         return count;
 661 }
 662
 663 static ssize_t stats_track_pid_show(struct kobject *kobj,
 664                                     struct attribute *attr,
 665                                     char *buf)
 666 {
 667         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 668 }
 669
 670 static ssize_t stats_track_pid_store(struct kobject *kobj,
 671                                      struct attribute *attr,
 672                                      const char *buffer,
 673                                      size_t count)
 674 {
 675         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 676 }
 677 LUSTRE_RW_ATTR(stats_track_pid);
 678
 679 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 680                                      struct attribute *attr,
 681                                      char *buf)
 682 {
 683         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 684 }
 685
 686 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 687                                       struct attribute *attr,
 688                                       const char *buffer,
 689                                       size_t count)
 690 {
 691         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 692 }
 693 LUSTRE_RW_ATTR(stats_track_ppid);
 694
 695 static ssize_t stats_track_gid_show(struct kobject *kobj,
 696                                     struct attribute *attr,
 697                                     char *buf)
 698 {
 699         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 700 }
 701
 702 static ssize_t stats_track_gid_store(struct kobject *kobj,
 703                                      struct attribute *attr,
 704                                      const char *buffer,
 705                                      size_t count)
 706 {
 707         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 708 }
 709 LUSTRE_RW_ATTR(stats_track_gid);
 710
 711 static ssize_t statahead_running_max_show(struct kobject *kobj,
 712                                           struct attribute *attr,
 713                                           char *buf)
 714 {
 715         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 716                                               ll_kset.kobj);
 717
 718         return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
 719 }
 720
 721 static ssize_t statahead_running_max_store(struct kobject *kobj,
 722                                            struct attribute *attr,
 723                                            const char *buffer,
 724                                            size_t count)
 725 {
 726         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 727                                               ll_kset.kobj);
 728         unsigned long val;
 729         int rc;
 730
 731         rc = kstrtoul(buffer, 0, &val);
 732         if (rc)
 733                 return rc;
 734
 735         if (val <= LL_SA_RUNNING_MAX) {
 736                 sbi->ll_sa_running_max = val;
 737                 return count;
 738         }
 739
 740         CERROR("Bad statahead_running_max value %lu. Valid values "
 741                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 742
 743         return -ERANGE;
 744 }
 745 LUSTRE_RW_ATTR(statahead_running_max);
 746
 747 static ssize_t statahead_max_show(struct kobject *kobj,
 748                                   struct attribute *attr,
 749                                   char *buf)
 750 {
 751         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 752                                               ll_kset.kobj);
 753
 754         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 755 }
 756
 757 static ssize_t statahead_max_store(struct kobject *kobj,
 758                                    struct attribute *attr,
 759                                    const char *buffer,
 760                                    size_t count)
 761 {
 762         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 763                                               ll_kset.kobj);
 764         unsigned long val;
 765         int rc;
 766
 767         rc = kstrtoul(buffer, 0, &val);
 768         if (rc)
 769                 return rc;
 770
 771         if (val <= LL_SA_RPC_MAX)
 772                 sbi->ll_sa_max = val;
 773         else
 774                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 775                        val, LL_SA_RPC_MAX);
 776
 777         return count;
 778 }
 779 LUSTRE_RW_ATTR(statahead_max);
 780
 781 static ssize_t statahead_agl_show(struct kobject *kobj,
 782                                   struct attribute *attr,
 783                                   char *buf)
 784 {
 785         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 786                                               ll_kset.kobj);
 787
 788         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
 789 }
 790
 791 static ssize_t statahead_agl_store(struct kobject *kobj,
 792                                    struct attribute *attr,
 793                                    const char *buffer,
 794                                    size_t count)
 795 {
 796         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 797                                               ll_kset.kobj);
 798         bool val;
 799         int rc;
 800
 801         rc = kstrtobool(buffer, &val);
 802         if (rc)
 803                 return rc;
 804
 805         if (val)
 806                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 807         else
 808                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
 809
 810         return count;
 811 }
 812 LUSTRE_RW_ATTR(statahead_agl);
 813
 814 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 815 {
 816         struct super_block *sb = m->private;
 817         struct ll_sb_info *sbi = ll_s2sbi(sb);
 818
 819         seq_printf(m, "statahead total: %u\n"
 820                       "statahead wrong: %u\n"
 821                       "agl total: %u\n",
 822                    atomic_read(&sbi->ll_sa_total),
 823                    atomic_read(&sbi->ll_sa_wrong),
 824                    atomic_read(&sbi->ll_agl_total));
 825         return 0;
 826 }
 827
 828 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
 829
 830 static ssize_t lazystatfs_show(struct kobject *kobj,
 831                                struct attribute *attr,
 832                                char *buf)
 833 {
 834         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 835                                               ll_kset.kobj);
 836
 837         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 838 }
 839
 840 static ssize_t lazystatfs_store(struct kobject *kobj,
 841                                 struct attribute *attr,
 842                                 const char *buffer,
 843                                 size_t count)
 844 {
 845         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 846                                               ll_kset.kobj);
 847         bool val;
 848         int rc;
 849
 850         rc = kstrtobool(buffer, &val);
 851         if (rc)
 852                 return rc;
 853
 854         if (val)
 855                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 856         else
 857                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 858
 859         return count;
 860 }
 861 LUSTRE_RW_ATTR(lazystatfs);
 862
 863 static ssize_t statfs_max_age_show(struct kobject *kobj, struct attribute *attr,
 864                                    char *buf)
 865 {
 866         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 867                                               ll_kset.kobj);
 868
 869         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_statfs_max_age);
 870 }
 871
 872 static ssize_t statfs_max_age_store(struct kobject *kobj,
 873                                     struct attribute *attr, const char *buffer,
 874                                     size_t count)
 875 {
 876         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 877                                               ll_kset.kobj);
 878         unsigned int val;
 879         int rc;
 880
 881         rc = kstrtouint(buffer, 10, &val);
 882         if (rc)
 883                 return rc;
 884         if (val > OBD_STATFS_CACHE_MAX_AGE)
 885                 return -EINVAL;
 886
 887         sbi->ll_statfs_max_age = val;
 888
 889         return count;
 890 }
 891 LUSTRE_RW_ATTR(statfs_max_age);
 892
 893 static ssize_t max_easize_show(struct kobject *kobj,
 894                                struct attribute *attr,
 895                                char *buf)
 896 {
 897         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 898                                               ll_kset.kobj);
 899         unsigned int ealen;
 900         int rc;
 901
 902         rc = ll_get_max_mdsize(sbi, &ealen);
 903         if (rc)
 904                 return rc;
 905
 906         /* Limit xattr size returned to userspace based on kernel maximum */
 907         return snprintf(buf, PAGE_SIZE, "%u\n",
 908                         ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 909 }
 910 LUSTRE_RO_ATTR(max_easize);
 911
 912 /**
 913  * Get default_easize.
 914  *
 915  * \see client_obd::cl_default_mds_easize
 916  *
 917  * \param[in] m         seq_file handle
 918  * \param[in] v         unused for single entry
 919  *
 920  * \retval 0            on success
 921  * \retval negative     negated errno on failure
 922  */
 923 static ssize_t default_easize_show(struct kobject *kobj,
 924                                    struct attribute *attr,
 925                                    char *buf)
 926 {
 927         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 928                                               ll_kset.kobj);
 929         unsigned int ealen;
 930         int rc;
 931
 932         rc = ll_get_default_mdsize(sbi, &ealen);
 933         if (rc)
 934                 return rc;
 935
 936         /* Limit xattr size returned to userspace based on kernel maximum */
 937         return snprintf(buf, PAGE_SIZE, "%u\n",
 938                         ealen > XATTR_SIZE_MAX ? XATTR_SIZE_MAX : ealen);
 939 }
 940
 941 /**
 942  * Set default_easize.
 943  *
 944  * Range checking on the passed value is handled by
 945  * ll_set_default_mdsize().
 946  *
 947  * \see client_obd::cl_default_mds_easize
 948  *
 949  * \param[in] file      proc file
 950  * \param[in] buffer    string passed from user space
 951  * \param[in] count     \a buffer length
 952  * \param[in] off       unused for single entry
 953  *
 954  * \retval positive     \a count on success
 955  * \retval negative     negated errno on failure
 956  */
 957 static ssize_t default_easize_store(struct kobject *kobj,
 958                                     struct attribute *attr,
 959                                     const char *buffer,
 960                                     size_t count)
 961 {
 962         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 963                                               ll_kset.kobj);
 964         unsigned int val;
 965         int rc;
 966
 967         if (count == 0)
 968                 return 0;
 969
 970         rc = kstrtouint(buffer, 10, &val);
 971         if (rc)
 972                 return rc;
 973
 974         rc = ll_set_default_mdsize(sbi, val);
 975         if (rc)
 976                 return rc;
 977
 978         return count;
 979 }
 980 LUSTRE_RW_ATTR(default_easize);
 981
 982 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 983 {
 984         const char *str[] = LL_SBI_FLAGS;
 985         struct super_block *sb = m->private;
 986         int flags = ll_s2sbi(sb)->ll_flags;
 987         int i = 0;
 988
 989         while (flags != 0) {
 990                 if (ARRAY_SIZE(str) <= i) {
 991                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
 992                                 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
 993                         return -EINVAL;
 994                 }
 995
 996                 if (flags & 0x1)
 997                         seq_printf(m, "%s ", str[i]);
 998                 flags >>= 1;
 999                 ++i;
1000         }
1001         seq_printf(m, "\b\n");
1002         return 0;
1003 }
1004
1005 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
1006
1007 static ssize_t xattr_cache_show(struct kobject *kobj,
1008                                 struct attribute *attr,
1009                                 char *buf)
1010 {
1011         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1012                                               ll_kset.kobj);
1013
1014         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
1015 }
1016
1017 static ssize_t xattr_cache_store(struct kobject *kobj,
1018                                  struct attribute *attr,
1019                                  const char *buffer,
1020                                  size_t count)
1021 {
1022         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1023                                               ll_kset.kobj);
1024         bool val;
1025         int rc;
1026
1027         rc = kstrtobool(buffer, &val);
1028         if (rc)
1029                 return rc;
1030
1031         if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1032                 return -ENOTSUPP;
1033
1034         sbi->ll_xattr_cache_enabled = val;
1035         sbi->ll_xattr_cache_set = 1;
1036
1037         return count;
1038 }
1039 LUSTRE_RW_ATTR(xattr_cache);
1040
1041 static ssize_t tiny_write_show(struct kobject *kobj,
1042                                struct attribute *attr,
1043                                char *buf)
1044 {
1045         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1046                                               ll_kset.kobj);
1047
1048         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1049 }
1050
1051 static ssize_t tiny_write_store(struct kobject *kobj,
1052                                 struct attribute *attr,
1053                                 const char *buffer,
1054                                 size_t count)
1055 {
1056         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1057                                               ll_kset.kobj);
1058         bool val;
1059         int rc;
1060
1061         rc = kstrtobool(buffer, &val);
1062         if (rc)
1063                 return rc;
1064
1065         spin_lock(&sbi->ll_lock);
1066         if (val)
1067                 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1068         else
1069                 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1070         spin_unlock(&sbi->ll_lock);
1071
1072         return count;
1073 }
1074 LUSTRE_RW_ATTR(tiny_write);
1075
1076 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1077                                                struct attribute *attr,
1078                                                char *buf)
1079 {
1080         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1081                                               ll_kset.kobj);
1082
1083         return snprintf(buf, PAGE_SIZE, "%u\n",
1084                         sbi->ll_ra_info.ra_async_max_active);
1085 }
1086
1087 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1088                                                  struct attribute *attr,
1089                                                  const char *buffer,
1090                                                  size_t count)
1091 {
1092         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1093                                               ll_kset.kobj);
1094         unsigned int val;
1095         int rc;
1096
1097         rc = kstrtouint(buffer, 10, &val);
1098         if (rc)
1099                 return rc;
1100
1101         /**
1102          * It doesn't make any sense to make it exceed what
1103          * workqueue could acutally support. This can easily
1104          * over subscripe the cores but Lustre internally
1105          * throttles to avoid those impacts.
1106          */
1107         if (val > WQ_UNBOUND_MAX_ACTIVE) {
1108                 CERROR("%s: cannot set max_read_ahead_async_active=%u larger than %u\n",
1109                        sbi->ll_fsname, val, WQ_UNBOUND_MAX_ACTIVE);
1110                 return -ERANGE;
1111         }
1112
1113         spin_lock(&sbi->ll_lock);
1114         sbi->ll_ra_info.ra_async_max_active = val;
1115         spin_unlock(&sbi->ll_lock);
1116
1117         return count;
1118 }
1119 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1120
1121 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1122                                                        struct attribute *attr,
1123                                                        char *buf)
1124 {
1125         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1126                                               ll_kset.kobj);
1127
1128         return snprintf(buf, PAGE_SIZE, "%lu\n",
1129              PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1130 }
1131
1132 static ssize_t
1133 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1134                                          struct attribute *attr,
1135                                          const char *buffer, size_t count)
1136 {
1137         unsigned long pages_number;
1138         unsigned long max_ra_per_file;
1139         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1140                                               ll_kset.kobj);
1141         int rc;
1142
1143         rc = kstrtoul(buffer, 10, &pages_number);
1144         if (rc)
1145                 return rc;
1146
1147         pages_number = MiB_TO_PAGES(pages_number);
1148         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1149         if (pages_number < 0 || pages_number > max_ra_per_file) {
1150                 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1151                        "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1152                        PAGES_TO_MiB(pages_number),
1153                        PAGES_TO_MiB(max_ra_per_file));
1154                 return -ERANGE;
1155         }
1156         sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1157
1158         return count;
1159 }
1160 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1161
1162 static ssize_t fast_read_show(struct kobject *kobj,
1163                               struct attribute *attr,
1164                               char *buf)
1165 {
1166         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1167                                               ll_kset.kobj);
1168
1169         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1170 }
1171
1172 static ssize_t fast_read_store(struct kobject *kobj,
1173                                struct attribute *attr,
1174                                const char *buffer,
1175                                size_t count)
1176 {
1177         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1178                                               ll_kset.kobj);
1179         bool val;
1180         int rc;
1181
1182         rc = kstrtobool(buffer, &val);
1183         if (rc)
1184                 return rc;
1185
1186         spin_lock(&sbi->ll_lock);
1187         if (val)
1188                 sbi->ll_flags |= LL_SBI_FAST_READ;
1189         else
1190                 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1191         spin_unlock(&sbi->ll_lock);
1192
1193         return count;
1194 }
1195 LUSTRE_RW_ATTR(fast_read);
1196
1197 static ssize_t file_heat_show(struct kobject *kobj,
1198                               struct attribute *attr,
1199                               char *buf)
1200 {
1201         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1202                                               ll_kset.kobj);
1203
1204         return snprintf(buf, PAGE_SIZE, "%u\n",
1205                         !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1206 }
1207
1208 static ssize_t file_heat_store(struct kobject *kobj,
1209                                struct attribute *attr,
1210                                const char *buffer,
1211                                size_t count)
1212 {
1213         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1214                                               ll_kset.kobj);
1215         bool val;
1216         int rc;
1217
1218         rc = kstrtobool(buffer, &val);
1219         if (rc)
1220                 return rc;
1221
1222         spin_lock(&sbi->ll_lock);
1223         if (val)
1224                 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1225         else
1226                 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1227         spin_unlock(&sbi->ll_lock);
1228
1229         return count;
1230 }
1231 LUSTRE_RW_ATTR(file_heat);
1232
1233 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1234                                           struct attribute *attr,
1235                                           char *buf)
1236 {
1237         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1238                                               ll_kset.kobj);
1239
1240         return snprintf(buf, PAGE_SIZE, "%u\n",
1241                        (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1242 }
1243
1244 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1245                                            struct attribute *attr,
1246                                            const char *buffer,
1247                                            size_t count)
1248 {
1249         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1250                                               ll_kset.kobj);
1251         unsigned long val;
1252         int rc;
1253
1254         rc = kstrtoul(buffer, 10, &val);
1255         if (rc)
1256                 return rc;
1257
1258         if (val < 0 || val > 100)
1259                 return -ERANGE;
1260
1261         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1262
1263         return count;
1264 }
1265 LUSTRE_RW_ATTR(heat_decay_percentage);
1266
1267 static ssize_t heat_period_second_show(struct kobject *kobj,
1268                                        struct attribute *attr,
1269                                        char *buf)
1270 {
1271         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1272                                               ll_kset.kobj);
1273
1274         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1275 }
1276
1277 static ssize_t heat_period_second_store(struct kobject *kobj,
1278                                         struct attribute *attr,
1279                                         const char *buffer,
1280                                         size_t count)
1281 {
1282         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1283                                               ll_kset.kobj);
1284         unsigned long val;
1285         int rc;
1286
1287         rc = kstrtoul(buffer, 10, &val);
1288         if (rc)
1289                 return rc;
1290
1291         if (val <= 0)
1292                 return -ERANGE;
1293
1294         sbi->ll_heat_period_second = val;
1295
1296         return count;
1297 }
1298 LUSTRE_RW_ATTR(heat_period_second);
1299
1300 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1301 {
1302         struct super_block      *sb    = m->private;
1303         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1304         struct cl_client_cache  *cache = sbi->ll_cache;
1305         long pages;
1306         int mb;
1307
1308         pages = atomic_long_read(&cache->ccc_unstable_nr);
1309         mb    = (pages * PAGE_SIZE) >> 20;
1310
1311         seq_printf(m, "unstable_check:     %8d\n"
1312                       "unstable_pages: %12ld\n"
1313                       "unstable_mb:        %8d\n",
1314                    cache->ccc_unstable_check, pages, mb);
1315         return 0;
1316 }
1317
1318 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1319                                            const char __user *buffer,
1320                                            size_t count, loff_t *unused)
1321 {
1322         struct seq_file *seq = file->private_data;
1323         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1324         char kernbuf[128];
1325         bool val;
1326         int rc;
1327
1328         if (count == 0)
1329                 return 0;
1330         if (count >= sizeof(kernbuf))
1331                 return -EINVAL;
1332
1333         if (copy_from_user(kernbuf, buffer, count))
1334                 return -EFAULT;
1335         kernbuf[count] = 0;
1336
1337         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1338                   kernbuf;
1339         rc = kstrtobool_from_user(buffer, count, &val);
1340         if (rc < 0)
1341                 return rc;
1342
1343         /* borrow lru lock to set the value */
1344         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1345         sbi->ll_cache->ccc_unstable_check = val;
1346         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1347
1348         return count;
1349 }
1350
1351 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1352
1353 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1354 {
1355         struct super_block *sb = m->private;
1356         struct ll_sb_info *sbi = ll_s2sbi(sb);
1357         struct root_squash_info *squash = &sbi->ll_squash;
1358
1359         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1360         return 0;
1361 }
1362
1363 static ssize_t ll_root_squash_seq_write(struct file *file,
1364                                         const char __user *buffer,
1365                                         size_t count, loff_t *off)
1366 {
1367         struct seq_file *m = file->private_data;
1368         struct super_block *sb = m->private;
1369         struct ll_sb_info *sbi = ll_s2sbi(sb);
1370         struct root_squash_info *squash = &sbi->ll_squash;
1371
1372         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1373 }
1374
1375 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1376
1377 static int ll_nosquash_nids_seq_show(struct seq_file *m, void *v)
1378 {
1379         struct super_block *sb = m->private;
1380         struct ll_sb_info *sbi = ll_s2sbi(sb);
1381         struct root_squash_info *squash = &sbi->ll_squash;
1382         int len;
1383
1384         spin_lock(&squash->rsi_lock);
1385         if (!list_empty(&squash->rsi_nosquash_nids)) {
1386                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1387                                         &squash->rsi_nosquash_nids);
1388                 m->count += len;
1389                 seq_putc(m, '\n');
1390         } else {
1391                 seq_puts(m, "NONE\n");
1392         }
1393         spin_unlock(&squash->rsi_lock);
1394
1395         return 0;
1396 }
1397
1398 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1399                                           const char __user *buffer,
1400                                           size_t count, loff_t *off)
1401 {
1402         struct seq_file *m = file->private_data;
1403         struct super_block *sb = m->private;
1404         struct ll_sb_info *sbi = ll_s2sbi(sb);
1405         struct root_squash_info *squash = &sbi->ll_squash;
1406         int rc;
1407
1408         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1409         if (rc < 0)
1410                 return rc;
1411
1412         ll_compute_rootsquash_state(sbi);
1413
1414         return rc;
1415 }
1416
1417 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1418
1419 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1420 {
1421         struct super_block *sb = m->private;
1422         struct ll_sb_info *sbi = ll_s2sbi(sb);
1423
1424         return pcc_super_dump(&sbi->ll_pcc_super, m);
1425 }
1426
1427 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1428                                 size_t count, loff_t *off)
1429 {
1430         struct seq_file *m = file->private_data;
1431         struct super_block *sb = m->private;
1432         struct ll_sb_info *sbi = ll_s2sbi(sb);
1433         int rc;
1434         char *kernbuf;
1435
1436         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1437                 return -EINVAL;
1438
1439         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1440                 return -EOPNOTSUPP;
1441
1442         OBD_ALLOC(kernbuf, count + 1);
1443         if (kernbuf == NULL)
1444                 return -ENOMEM;
1445
1446         if (copy_from_user(kernbuf, buffer, count))
1447                 GOTO(out_free_kernbuff, rc = -EFAULT);
1448
1449         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1450 out_free_kernbuff:
1451         OBD_FREE(kernbuf, count + 1);
1452         return rc ? rc : count;
1453 }
1454 LDEBUGFS_SEQ_FOPS(ll_pcc);
1455
1456 struct ldebugfs_vars lprocfs_llite_obd_vars[] = {
1457         { .name =       "site",
1458           .fops =       &ll_site_stats_fops                     },
1459         { .name =       "max_cached_mb",
1460           .fops =       &ll_max_cached_mb_fops                  },
1461         { .name =       "statahead_stats",
1462           .fops =       &ll_statahead_stats_fops                },
1463         { .name =       "unstable_stats",
1464           .fops =       &ll_unstable_stats_fops                 },
1465         { .name =       "sbi_flags",
1466           .fops =       &ll_sbi_flags_fops                      },
1467         { .name =       "root_squash",
1468           .fops =       &ll_root_squash_fops                    },
1469         { .name =       "nosquash_nids",
1470           .fops =       &ll_nosquash_nids_fops                  },
1471         { .name =       "pcc",
1472           .fops =       &ll_pcc_fops,                           },
1473         { NULL }
1474 };
1475
1476 #define MAX_STRING_SIZE 128
1477
1478 static struct attribute *llite_attrs[] = {
1479         &lustre_attr_blocksize.attr,
1480         &lustre_attr_stat_blocksize.attr,
1481         &lustre_attr_kbytestotal.attr,
1482         &lustre_attr_kbytesfree.attr,
1483         &lustre_attr_kbytesavail.attr,
1484         &lustre_attr_filestotal.attr,
1485         &lustre_attr_filesfree.attr,
1486         &lustre_attr_client_type.attr,
1487         &lustre_attr_fstype.attr,
1488         &lustre_attr_uuid.attr,
1489         &lustre_attr_checksums.attr,
1490         &lustre_attr_checksum_pages.attr,
1491         &lustre_attr_max_read_ahead_mb.attr,
1492         &lustre_attr_max_read_ahead_per_file_mb.attr,
1493         &lustre_attr_max_read_ahead_whole_mb.attr,
1494         &lustre_attr_max_read_ahead_async_active.attr,
1495         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1496         &lustre_attr_stats_track_pid.attr,
1497         &lustre_attr_stats_track_ppid.attr,
1498         &lustre_attr_stats_track_gid.attr,
1499         &lustre_attr_statahead_running_max.attr,
1500         &lustre_attr_statahead_max.attr,
1501         &lustre_attr_statahead_agl.attr,
1502         &lustre_attr_lazystatfs.attr,
1503         &lustre_attr_statfs_max_age.attr,
1504         &lustre_attr_max_easize.attr,
1505         &lustre_attr_default_easize.attr,
1506         &lustre_attr_xattr_cache.attr,
1507         &lustre_attr_fast_read.attr,
1508         &lustre_attr_tiny_write.attr,
1509         &lustre_attr_file_heat.attr,
1510         &lustre_attr_heat_decay_percentage.attr,
1511         &lustre_attr_heat_period_second.attr,
1512         NULL,
1513 };
1514
1515 static void sbi_kobj_release(struct kobject *kobj)
1516 {
1517         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1518                                               ll_kset.kobj);
1519         complete(&sbi->ll_kobj_unregister);
1520 }
1521
1522 static struct kobj_type sbi_ktype = {
1523         .default_attrs  = llite_attrs,
1524         .sysfs_ops      = &lustre_sysfs_ops,
1525         .release        = sbi_kobj_release,
1526 };
1527
1528 static const struct llite_file_opcode {
1529         __u32           opcode;
1530         __u32           type;
1531         const char      *opname;
1532 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1533         /* file operation */
1534         { LPROC_LL_READ_BYTES,  LPROCFS_TYPE_BYTES_FULL, "read_bytes" },
1535         { LPROC_LL_WRITE_BYTES, LPROCFS_TYPE_BYTES_FULL, "write_bytes" },
1536         { LPROC_LL_READ,        LPROCFS_TYPE_LATENCY,   "read" },
1537         { LPROC_LL_WRITE,       LPROCFS_TYPE_LATENCY,   "write" },
1538         { LPROC_LL_IOCTL,       LPROCFS_TYPE_REQS,      "ioctl" },
1539         { LPROC_LL_OPEN,        LPROCFS_TYPE_LATENCY,   "open" },
1540         { LPROC_LL_RELEASE,     LPROCFS_TYPE_LATENCY,   "close" },
1541         { LPROC_LL_MMAP,        LPROCFS_TYPE_LATENCY,   "mmap" },
1542         { LPROC_LL_FAULT,       LPROCFS_TYPE_LATENCY,   "page_fault" },
1543         { LPROC_LL_MKWRITE,     LPROCFS_TYPE_LATENCY,   "page_mkwrite" },
1544         { LPROC_LL_LLSEEK,      LPROCFS_TYPE_LATENCY,   "seek" },
1545         { LPROC_LL_FSYNC,       LPROCFS_TYPE_LATENCY,   "fsync" },
1546         { LPROC_LL_READDIR,     LPROCFS_TYPE_LATENCY,   "readdir" },
1547         /* inode operation */
1548         { LPROC_LL_SETATTR,     LPROCFS_TYPE_LATENCY,   "setattr" },
1549         { LPROC_LL_TRUNC,       LPROCFS_TYPE_LATENCY,   "truncate" },
1550         { LPROC_LL_FLOCK,       LPROCFS_TYPE_LATENCY,   "flock" },
1551         { LPROC_LL_GETATTR,     LPROCFS_TYPE_LATENCY,   "getattr" },
1552         { LPROC_LL_FALLOCATE,   LPROCFS_TYPE_LATENCY, "fallocate"},
1553         /* dir inode operation */
1554         { LPROC_LL_CREATE,      LPROCFS_TYPE_LATENCY,   "create" },
1555         { LPROC_LL_LINK,        LPROCFS_TYPE_LATENCY,   "link" },
1556         { LPROC_LL_UNLINK,      LPROCFS_TYPE_LATENCY,   "unlink" },
1557         { LPROC_LL_SYMLINK,     LPROCFS_TYPE_LATENCY,   "symlink" },
1558         { LPROC_LL_MKDIR,       LPROCFS_TYPE_LATENCY,   "mkdir" },
1559         { LPROC_LL_RMDIR,       LPROCFS_TYPE_LATENCY,   "rmdir" },
1560         { LPROC_LL_MKNOD,       LPROCFS_TYPE_LATENCY,   "mknod" },
1561         { LPROC_LL_RENAME,      LPROCFS_TYPE_LATENCY,   "rename" },
1562         /* special inode operation */
1563         { LPROC_LL_STATFS,      LPROCFS_TYPE_LATENCY,   "statfs" },
1564         { LPROC_LL_SETXATTR,    LPROCFS_TYPE_LATENCY,   "setxattr" },
1565         { LPROC_LL_GETXATTR,    LPROCFS_TYPE_LATENCY,   "getxattr" },
1566         { LPROC_LL_GETXATTR_HITS, LPROCFS_TYPE_REQS,    "getxattr_hits" },
1567         { LPROC_LL_LISTXATTR,   LPROCFS_TYPE_LATENCY,   "listxattr" },
1568         { LPROC_LL_REMOVEXATTR, LPROCFS_TYPE_LATENCY,   "removexattr" },
1569         { LPROC_LL_INODE_PERM,  LPROCFS_TYPE_LATENCY,   "inode_permission" },
1570 };
1571
1572 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, long count)
1573 {
1574         if (!sbi->ll_stats)
1575                 return;
1576
1577         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1578                 lprocfs_counter_add(sbi->ll_stats, op, count);
1579         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1580                  sbi->ll_stats_track_id == current->pid)
1581                 lprocfs_counter_add(sbi->ll_stats, op, count);
1582         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1583                  sbi->ll_stats_track_id == current->parent->pid)
1584                 lprocfs_counter_add(sbi->ll_stats, op, count);
1585         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1586                  sbi->ll_stats_track_id ==
1587                         from_kgid(&init_user_ns, current_gid()))
1588                 lprocfs_counter_add(sbi->ll_stats, op, count);
1589 }
1590 EXPORT_SYMBOL(ll_stats_ops_tally);
1591
1592 static const char *ra_stat_string[] = {
1593         [RA_STAT_HIT] = "hits",
1594         [RA_STAT_MISS] = "misses",
1595         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1596         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1597         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1598         [RA_STAT_FAILED_MATCH] = "failed lock match",
1599         [RA_STAT_DISCARDED] = "read but discarded",
1600         [RA_STAT_ZERO_LEN] = "zero length file",
1601         [RA_STAT_ZERO_WINDOW] = "zero size window",
1602         [RA_STAT_EOF] = "read-ahead to EOF",
1603         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1604         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1605         [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1606         [RA_STAT_ASYNC] = "async readahead",
1607         [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1608 };
1609
1610 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1611 {
1612         struct lustre_sb_info *lsi = s2lsi(sb);
1613         struct ll_sb_info *sbi = ll_s2sbi(sb);
1614         int err, id;
1615
1616         ENTRY;
1617         LASSERT(sbi);
1618
1619         if (IS_ERR_OR_NULL(llite_root))
1620                 goto out_ll_kset;
1621
1622         sbi->ll_debugfs_entry = debugfs_create_dir(name, llite_root);
1623         ldebugfs_add_vars(sbi->ll_debugfs_entry, lprocfs_llite_obd_vars, sb);
1624
1625         debugfs_create_file("dump_page_cache", 0444, sbi->ll_debugfs_entry, sbi,
1626                             &vvp_dump_pgcache_file_ops);
1627
1628         debugfs_create_file("extents_stats", 0644, sbi->ll_debugfs_entry, sbi,
1629                                  &ll_rw_extents_stats_fops);
1630
1631         debugfs_create_file("extents_stats_per_process", 0644,
1632                             sbi->ll_debugfs_entry, sbi,
1633                             &ll_rw_extents_stats_pp_fops);
1634
1635         debugfs_create_file("offset_stats", 0644, sbi->ll_debugfs_entry, sbi,
1636                             &ll_rw_offset_stats_fops);
1637
1638         /* File operations stats */
1639         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1640                                             LPROCFS_STATS_FLAG_NONE);
1641         if (sbi->ll_stats == NULL)
1642                 GOTO(out_debugfs, err = -ENOMEM);
1643
1644         /* do counter init */
1645         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1646                 u32 type = llite_opcode_table[id].type;
1647                 void *ptr = "unknown";
1648
1649                 if (type & LPROCFS_TYPE_REQS)
1650                         ptr = "reqs";
1651                 else if (type & LPROCFS_TYPE_BYTES)
1652                         ptr = "bytes";
1653                 else if (type & LPROCFS_TYPE_USEC)
1654                         ptr = "usec";
1655                 lprocfs_counter_init(sbi->ll_stats,
1656                                      llite_opcode_table[id].opcode, type,
1657                                      llite_opcode_table[id].opname, ptr);
1658         }
1659
1660         debugfs_create_file("stats", 0644, sbi->ll_debugfs_entry,
1661                             sbi->ll_stats, &ldebugfs_stats_seq_fops);
1662
1663         sbi->ll_ra_stats = lprocfs_alloc_stats(ARRAY_SIZE(ra_stat_string),
1664                                                LPROCFS_STATS_FLAG_NONE);
1665         if (sbi->ll_ra_stats == NULL)
1666                 GOTO(out_stats, err = -ENOMEM);
1667
1668         for (id = 0; id < ARRAY_SIZE(ra_stat_string); id++)
1669                 lprocfs_counter_init(sbi->ll_ra_stats, id, 0,
1670                                      ra_stat_string[id], "pages");
1671
1672         debugfs_create_file("read_ahead_stats", 0644, sbi->ll_debugfs_entry,
1673                             sbi->ll_ra_stats, &ldebugfs_stats_seq_fops);
1674
1675 out_ll_kset:
1676         /* Yes we also register sysfs mount kset here as well */
1677         sbi->ll_kset.kobj.parent = llite_kobj;
1678         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1679         init_completion(&sbi->ll_kobj_unregister);
1680         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1681         if (err)
1682                 GOTO(out_ra_stats, err);
1683
1684         err = kset_register(&sbi->ll_kset);
1685         if (err)
1686                 GOTO(out_ra_stats, err);
1687
1688         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1689
1690         RETURN(0);
1691 out_ra_stats:
1692         lprocfs_free_stats(&sbi->ll_ra_stats);
1693 out_stats:
1694         lprocfs_free_stats(&sbi->ll_stats);
1695 out_debugfs:
1696         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1697
1698         RETURN(err);
1699 }
1700
1701 void ll_debugfs_unregister_super(struct super_block *sb)
1702 {
1703         struct lustre_sb_info *lsi = s2lsi(sb);
1704         struct ll_sb_info *sbi = ll_s2sbi(sb);
1705
1706         debugfs_remove_recursive(sbi->ll_debugfs_entry);
1707
1708         if (sbi->ll_dt_obd)
1709                 sysfs_remove_link(&sbi->ll_kset.kobj,
1710                                   sbi->ll_dt_obd->obd_type->typ_name);
1711
1712         if (sbi->ll_md_obd)
1713                 sysfs_remove_link(&sbi->ll_kset.kobj,
1714                                   sbi->ll_md_obd->obd_type->typ_name);
1715
1716         kobject_put(lsi->lsi_kobj);
1717
1718         kset_unregister(&sbi->ll_kset);
1719         wait_for_completion(&sbi->ll_kobj_unregister);
1720
1721         lprocfs_free_stats(&sbi->ll_ra_stats);
1722         lprocfs_free_stats(&sbi->ll_stats);
1723 }
1724 #undef MAX_STRING_SIZE
1725
1726 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1727                                    struct seq_file *seq, int which)
1728 {
1729         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1730         unsigned long start, end, r, w;
1731         char *unitp = "KMGTPEZY";
1732         int i, units = 10;
1733         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1734
1735         read_cum = 0;
1736         write_cum = 0;
1737         start = 0;
1738
1739         for(i = 0; i < LL_HIST_MAX; i++) {
1740                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1741                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1742         }
1743
1744         for(i = 0; i < LL_HIST_MAX; i++) {
1745                 r = pp_info->pp_r_hist.oh_buckets[i];
1746                 w = pp_info->pp_w_hist.oh_buckets[i];
1747                 read_cum += r;
1748                 write_cum += w;
1749                 end = 1 << (i + LL_HIST_START - units);
1750                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1751                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1752                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1753                            r, pct(r, read_tot), pct(read_cum, read_tot),
1754                            w, pct(w, write_tot), pct(write_cum, write_tot));
1755                 start = end;
1756                 if (start == (1 << 10)) {
1757                         start = 1;
1758                         units += 10;
1759                         unitp++;
1760                 }
1761                 if (read_cum == read_tot && write_cum == write_tot)
1762                         break;
1763         }
1764 }
1765
1766 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1767 {
1768         struct timespec64 now;
1769         struct ll_sb_info *sbi = seq->private;
1770         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1771         int k;
1772
1773         ktime_get_real_ts64(&now);
1774
1775         if (!sbi->ll_rw_stats_on) {
1776                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1777                 return 0;
1778         }
1779         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1780                    (s64)now.tv_sec, now.tv_nsec);
1781         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1782         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1783                    "extents", "calls", "%", "cum%",
1784                    "calls", "%", "cum%");
1785         spin_lock(&sbi->ll_pp_extent_lock);
1786         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1787                 if (io_extents->pp_extents[k].pid != 0) {
1788                         seq_printf(seq, "\nPID: %d\n",
1789                                    io_extents->pp_extents[k].pid);
1790                         ll_display_extents_info(io_extents, seq, k);
1791                 }
1792         }
1793         spin_unlock(&sbi->ll_pp_extent_lock);
1794         return 0;
1795 }
1796
1797 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1798                                                 const char __user *buf,
1799                                                 size_t len,
1800                                                 loff_t *off)
1801 {
1802         struct seq_file *seq = file->private_data;
1803         struct ll_sb_info *sbi = seq->private;
1804         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1805         int i;
1806         __s64 value;
1807
1808         if (len == 0)
1809                 return -EINVAL;
1810
1811         value = ll_stats_pid_write(buf, len);
1812
1813         if (value == 0)
1814                 sbi->ll_rw_stats_on = 0;
1815         else
1816                 sbi->ll_rw_stats_on = 1;
1817
1818         spin_lock(&sbi->ll_pp_extent_lock);
1819         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1820                 io_extents->pp_extents[i].pid = 0;
1821                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1822                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1823         }
1824         spin_unlock(&sbi->ll_pp_extent_lock);
1825         return len;
1826 }
1827
1828 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1829
1830 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1831 {
1832         struct timespec64 now;
1833         struct ll_sb_info *sbi = seq->private;
1834         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1835
1836         ktime_get_real_ts64(&now);
1837
1838         if (!sbi->ll_rw_stats_on) {
1839                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1840                 return 0;
1841         }
1842         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1843                    (s64)now.tv_sec, now.tv_nsec);
1844
1845         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1846         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1847                    "extents", "calls", "%", "cum%",
1848                    "calls", "%", "cum%");
1849         spin_lock(&sbi->ll_lock);
1850         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1851         spin_unlock(&sbi->ll_lock);
1852
1853         return 0;
1854 }
1855
1856 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1857                                              const char __user *buf,
1858                                              size_t len, loff_t *off)
1859 {
1860         struct seq_file *seq = file->private_data;
1861         struct ll_sb_info *sbi = seq->private;
1862         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1863         int i;
1864         __s64 value;
1865
1866         if (len == 0)
1867                 return -EINVAL;
1868
1869         value = ll_stats_pid_write(buf, len);
1870
1871         if (value == 0)
1872                 sbi->ll_rw_stats_on = 0;
1873         else
1874                 sbi->ll_rw_stats_on = 1;
1875
1876         spin_lock(&sbi->ll_pp_extent_lock);
1877         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1878                 io_extents->pp_extents[i].pid = 0;
1879                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1880                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1881         }
1882         spin_unlock(&sbi->ll_pp_extent_lock);
1883
1884         return len;
1885 }
1886
1887 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1888
1889 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1890                        struct ll_file_data *file, loff_t pos,
1891                        size_t count, int rw)
1892 {
1893         int i, cur = -1;
1894         struct ll_rw_process_info *process;
1895         struct ll_rw_process_info *offset;
1896         int *off_count = &sbi->ll_rw_offset_entry_count;
1897         int *process_count = &sbi->ll_offset_process_count;
1898         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1899
1900         if(!sbi->ll_rw_stats_on)
1901                 return;
1902         process = sbi->ll_rw_process_info;
1903         offset = sbi->ll_rw_offset_info;
1904
1905         spin_lock(&sbi->ll_pp_extent_lock);
1906         /* Extent statistics */
1907         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1908                 if(io_extents->pp_extents[i].pid == pid) {
1909                         cur = i;
1910                         break;
1911                 }
1912         }
1913
1914         if (cur == -1) {
1915                 /* new process */
1916                 sbi->ll_extent_process_count =
1917                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1918                 cur = sbi->ll_extent_process_count;
1919                 io_extents->pp_extents[cur].pid = pid;
1920                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1921                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1922         }
1923
1924         for (i = 0; (count >= 1 << (LL_HIST_START + i)) &&
1925              (i < (LL_HIST_MAX - 1)); i++);
1926         if (rw == 0) {
1927                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1928                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1929         } else {
1930                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1931                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1932         }
1933         spin_unlock(&sbi->ll_pp_extent_lock);
1934
1935         spin_lock(&sbi->ll_process_lock);
1936         /* Offset statistics */
1937         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1938                 if (process[i].rw_pid == pid) {
1939                         if (process[i].rw_last_file != file) {
1940                                 process[i].rw_range_start = pos;
1941                                 process[i].rw_last_file_pos = pos + count;
1942                                 process[i].rw_smallest_extent = count;
1943                                 process[i].rw_largest_extent = count;
1944                                 process[i].rw_offset = 0;
1945                                 process[i].rw_last_file = file;
1946                                 spin_unlock(&sbi->ll_process_lock);
1947                                 return;
1948                         }
1949                         if (process[i].rw_last_file_pos != pos) {
1950                                 *off_count =
1951                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1952                                 offset[*off_count].rw_op = process[i].rw_op;
1953                                 offset[*off_count].rw_pid = pid;
1954                                 offset[*off_count].rw_range_start =
1955                                         process[i].rw_range_start;
1956                                 offset[*off_count].rw_range_end =
1957                                         process[i].rw_last_file_pos;
1958                                 offset[*off_count].rw_smallest_extent =
1959                                         process[i].rw_smallest_extent;
1960                                 offset[*off_count].rw_largest_extent =
1961                                         process[i].rw_largest_extent;
1962                                 offset[*off_count].rw_offset =
1963                                         process[i].rw_offset;
1964                                 process[i].rw_op = rw;
1965                                 process[i].rw_range_start = pos;
1966                                 process[i].rw_smallest_extent = count;
1967                                 process[i].rw_largest_extent = count;
1968                                 process[i].rw_offset = pos -
1969                                         process[i].rw_last_file_pos;
1970                         }
1971                         if(process[i].rw_smallest_extent > count)
1972                                 process[i].rw_smallest_extent = count;
1973                         if(process[i].rw_largest_extent < count)
1974                                 process[i].rw_largest_extent = count;
1975                         process[i].rw_last_file_pos = pos + count;
1976                         spin_unlock(&sbi->ll_process_lock);
1977                         return;
1978                 }
1979         }
1980         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1981         process[*process_count].rw_pid = pid;
1982         process[*process_count].rw_op = rw;
1983         process[*process_count].rw_range_start = pos;
1984         process[*process_count].rw_last_file_pos = pos + count;
1985         process[*process_count].rw_smallest_extent = count;
1986         process[*process_count].rw_largest_extent = count;
1987         process[*process_count].rw_offset = 0;
1988         process[*process_count].rw_last_file = file;
1989         spin_unlock(&sbi->ll_process_lock);
1990 }
1991
1992 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1993 {
1994         struct timespec64 now;
1995         struct ll_sb_info *sbi = seq->private;
1996         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
1997         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
1998         int i;
1999
2000         ktime_get_real_ts64(&now);
2001
2002         if (!sbi->ll_rw_stats_on) {
2003                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2004                 return 0;
2005         }
2006         spin_lock(&sbi->ll_process_lock);
2007
2008         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
2009                    (s64)now.tv_sec, now.tv_nsec);
2010         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2011                    "R/W", "PID", "RANGE START", "RANGE END",
2012                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2013
2014         /* We stored the discontiguous offsets here; print them first */
2015         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2016                 if (offset[i].rw_pid != 0)
2017                         seq_printf(seq,
2018                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2019                                    offset[i].rw_op == READ ? 'R' : 'W',
2020                                    offset[i].rw_pid,
2021                                    offset[i].rw_range_start,
2022                                    offset[i].rw_range_end,
2023                                    (unsigned long)offset[i].rw_smallest_extent,
2024                                    (unsigned long)offset[i].rw_largest_extent,
2025                                    offset[i].rw_offset);
2026         }
2027
2028         /* Then print the current offsets for each process */
2029         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2030                 if (process[i].rw_pid != 0)
2031                         seq_printf(seq,
2032                                   "%3c %10d %14llu %14llu %17lu %17lu %14lld\n",
2033                                    process[i].rw_op == READ ? 'R' : 'W',
2034                                    process[i].rw_pid,
2035                                    process[i].rw_range_start,
2036                                    process[i].rw_last_file_pos,
2037                                    (unsigned long)process[i].rw_smallest_extent,
2038                                    (unsigned long)process[i].rw_largest_extent,
2039                                    process[i].rw_offset);
2040         }
2041         spin_unlock(&sbi->ll_process_lock);
2042
2043         return 0;
2044 }
2045
2046 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2047                                             const char __user *buf,
2048                                             size_t len, loff_t *off)
2049 {
2050         struct seq_file *seq = file->private_data;
2051         struct ll_sb_info *sbi = seq->private;
2052         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2053         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2054         __s64 value;
2055
2056         if (len == 0)
2057                 return -EINVAL;
2058
2059         value = ll_stats_pid_write(buf, len);
2060
2061         if (value == 0)
2062                 sbi->ll_rw_stats_on = 0;
2063         else
2064                 sbi->ll_rw_stats_on = 1;
2065
2066         spin_lock(&sbi->ll_process_lock);
2067         sbi->ll_offset_process_count = 0;
2068         sbi->ll_rw_offset_entry_count = 0;
2069         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2070                LL_PROCESS_HIST_MAX);
2071         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2072                LL_OFFSET_HIST_MAX);
2073         spin_unlock(&sbi->ll_process_lock);
2074
2075         return len;
2076 }
2077
2078 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);