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 #ifdef HAVE_UIDGID_HEADER
  37 # include <linux/uidgid.h>
  38 #endif
  39 #include <uapi/linux/lustre/lustre_param.h>
  40 #include <lprocfs_status.h>
  41 #include <obd_support.h>
  42
  43 #include "llite_internal.h"
  44 #include "vvp_internal.h"
  45
  46 static struct kobject *llite_kobj;
  47 static struct dentry *llite_root;
  48
  49 static void llite_kobj_release(struct kobject *kobj)
  50 {
  51         if (!IS_ERR_OR_NULL(llite_root)) {
  52                 debugfs_remove(llite_root);
  53                 llite_root = NULL;
  54         }
  55
  56         kfree(kobj);
  57 }
  58
  59 static struct kobj_type llite_kobj_ktype = {
  60         .release        = llite_kobj_release,
  61         .sysfs_ops      = &lustre_sysfs_ops,
  62 };
  63
  64 int llite_tunables_register(void)
  65 {
  66         int rc;
  67
  68         llite_kobj = kzalloc(sizeof(*llite_kobj), GFP_KERNEL);
  69         if (!llite_kobj)
  70                 return -ENOMEM;
  71
  72         llite_kobj->kset = lustre_kset;
  73         rc = kobject_init_and_add(llite_kobj, &llite_kobj_ktype,
  74                                   &lustre_kset->kobj, "%s", "llite");
  75         if (rc)
  76                 goto free_kobj;
  77
  78         llite_root = debugfs_create_dir("llite", debugfs_lustre_root);
  79         if (IS_ERR_OR_NULL(llite_root)) {
  80                 rc = llite_root ? PTR_ERR(llite_root) : -ENOMEM;
  81                 llite_root = NULL;
  82 free_kobj:
  83                 kobject_put(llite_kobj);
  84                 llite_kobj = NULL;
  85         }
  86
  87         return rc;
  88 }
  89
  90 void llite_tunables_unregister(void)
  91 {
  92         kobject_put(llite_kobj);
  93         llite_kobj = NULL;
  94 }
  95
  96 /* <debugfs>/lustre/llite mount point registration */
  97 static const struct file_operations ll_rw_extents_stats_fops;
  98 static const struct file_operations ll_rw_extents_stats_pp_fops;
  99 static const struct file_operations ll_rw_offset_stats_fops;
 100
 101 /**
 102  * ll_stats_pid_write() - Determine if stats collection should be enabled
 103  * @buf: Buffer containing the data written
 104  * @len: Number of bytes in the buffer
 105  *
 106  * Several proc files begin collecting stats when a value is written, and stop
 107  * collecting when either '0' or 'disable' is written. This function checks the
 108  * written value to see if collection should be enabled or disabled.
 109  *
 110  * Return: If '0' or 'disable' is provided, 0 is returned. If the text
 111  * equivalent of a number is written, that number is returned. Otherwise,
 112  * 1 is returned. Non-zero return values indicate collection should be enabled.
 113  */
 114 static s64 ll_stats_pid_write(const char __user *buf, size_t len)
 115 {
 116         unsigned long long value = 1;
 117         char kernbuf[16];
 118         int rc;
 119
 120         rc = kstrtoull_from_user(buf, len, 0, &value);
 121         if (rc < 0 && len < sizeof(kernbuf)) {
 122                 if (copy_from_user(kernbuf, buf, len))
 123                         return -EFAULT;
 124                 kernbuf[len] = 0;
 125
 126                 if (kernbuf[len - 1] == '\n')
 127                         kernbuf[len - 1] = 0;
 128
 129                 if (strncasecmp(kernbuf, "disable", 7) == 0)
 130                         value = 0;
 131         }
 132
 133         return value;
 134 }
 135
 136 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 137                               char *buf)
 138 {
 139         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 140                                               ll_kset.kobj);
 141         struct obd_statfs osfs;
 142         int rc;
 143
 144         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 145         if (rc)
 146                 return rc;
 147
 148         return sprintf(buf, "%u\n", osfs.os_bsize);
 149 }
 150 LUSTRE_RO_ATTR(blocksize);
 151
 152 static ssize_t stat_blocksize_show(struct kobject *kobj, struct attribute *attr,
 153                                    char *buf)
 154 {
 155         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 156                                               ll_kset.kobj);
 157
 158         return sprintf(buf, "%u\n", sbi->ll_stat_blksize);
 159 }
 160
 161 static ssize_t stat_blocksize_store(struct kobject *kobj,
 162                                     struct attribute *attr,
 163                                     const char *buffer,
 164                                     size_t count)
 165 {
 166         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 167                                               ll_kset.kobj);
 168         unsigned int val;
 169         int rc;
 170
 171         rc = kstrtouint(buffer, 10, &val);
 172         if (rc)
 173                 return rc;
 174
 175         if (val != 0 && (val < PAGE_SIZE || (val & (val - 1))) != 0)
 176                 return -ERANGE;
 177
 178         sbi->ll_stat_blksize = val;
 179
 180         return count;
 181 }
 182 LUSTRE_RW_ATTR(stat_blocksize);
 183
 184 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 185                                 char *buf)
 186 {
 187         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 188                                               ll_kset.kobj);
 189         struct obd_statfs osfs;
 190         u32 blk_size;
 191         u64 result;
 192         int rc;
 193
 194         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 195         if (rc)
 196                 return rc;
 197
 198         blk_size = osfs.os_bsize >> 10;
 199         result = osfs.os_blocks;
 200
 201         while (blk_size >>= 1)
 202                 result <<= 1;
 203
 204         return sprintf(buf, "%llu\n", result);
 205 }
 206 LUSTRE_RO_ATTR(kbytestotal);
 207
 208 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 209                                char *buf)
 210 {
 211         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 212                                               ll_kset.kobj);
 213         struct obd_statfs osfs;
 214         u32 blk_size;
 215         u64 result;
 216         int rc;
 217
 218         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 219         if (rc)
 220                 return rc;
 221
 222         blk_size = osfs.os_bsize >> 10;
 223         result = osfs.os_bfree;
 224
 225         while (blk_size >>= 1)
 226                 result <<= 1;
 227
 228         return sprintf(buf, "%llu\n", result);
 229 }
 230 LUSTRE_RO_ATTR(kbytesfree);
 231
 232 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 233                                 char *buf)
 234 {
 235         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 236                                               ll_kset.kobj);
 237         struct obd_statfs osfs;
 238         u32 blk_size;
 239         u64 result;
 240         int rc;
 241
 242         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 243         if (rc)
 244                 return rc;
 245
 246         blk_size = osfs.os_bsize >> 10;
 247         result = osfs.os_bavail;
 248
 249         while (blk_size >>= 1)
 250                 result <<= 1;
 251
 252         return sprintf(buf, "%llu\n", result);
 253 }
 254 LUSTRE_RO_ATTR(kbytesavail);
 255
 256 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 257                                char *buf)
 258 {
 259         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 260                                               ll_kset.kobj);
 261         struct obd_statfs osfs;
 262         int rc;
 263
 264         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 265         if (rc)
 266                 return rc;
 267
 268         return sprintf(buf, "%llu\n", osfs.os_files);
 269 }
 270 LUSTRE_RO_ATTR(filestotal);
 271
 272 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 273                               char *buf)
 274 {
 275         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 276                                               ll_kset.kobj);
 277         struct obd_statfs osfs;
 278         int rc;
 279
 280         rc = ll_statfs_internal(sbi, &osfs, OBD_STATFS_NODELAY);
 281         if (rc)
 282                 return rc;
 283
 284         return sprintf(buf, "%llu\n", osfs.os_ffree);
 285 }
 286 LUSTRE_RO_ATTR(filesfree);
 287
 288 static ssize_t client_type_show(struct kobject *kobj, struct attribute *attr,
 289                                 char *buf)
 290 {
 291         return sprintf(buf, "local client\n");
 292 }
 293 LUSTRE_RO_ATTR(client_type);
 294
 295 static ssize_t fstype_show(struct kobject *kobj, struct attribute *attr,
 296                            char *buf)
 297 {
 298         return sprintf(buf, "lustre\n");
 299 }
 300 LUSTRE_RO_ATTR(fstype);
 301
 302 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 303                          char *buf)
 304 {
 305         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 306                                               ll_kset.kobj);
 307
 308         return sprintf(buf, "%s\n", sbi->ll_sb_uuid.uuid);
 309 }
 310 LUSTRE_RO_ATTR(uuid);
 311
 312 static int ll_site_stats_seq_show(struct seq_file *m, void *v)
 313 {
 314         struct super_block *sb = m->private;
 315
 316         /*
 317          * See description of statistical counters in struct cl_site, and
 318          * struct lu_site.
 319          */
 320         return cl_site_stats_print(lu2cl_site(ll_s2sbi(sb)->ll_site), m);
 321 }
 322
 323 LDEBUGFS_SEQ_FOPS_RO(ll_site_stats);
 324
 325 static int ll_max_readahead_mb_seq_show(struct seq_file *m, void *v)
 326 {
 327         struct super_block *sb = m->private;
 328         struct ll_sb_info *sbi = ll_s2sbi(sb);
 329         unsigned long ra_max_mb;
 330
 331         spin_lock(&sbi->ll_lock);
 332         ra_max_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages);
 333         spin_unlock(&sbi->ll_lock);
 334
 335         seq_printf(m, "%lu\n", ra_max_mb);
 336         return 0;
 337 }
 338
 339 static ssize_t
 340 ll_max_readahead_mb_seq_write(struct file *file, const char __user *buffer,
 341                               size_t count, loff_t *off)
 342 {
 343         struct seq_file *m = file->private_data;
 344         struct super_block *sb = m->private;
 345         struct ll_sb_info *sbi = ll_s2sbi(sb);
 346         s64 ra_max_mb, pages_number;
 347         int rc;
 348
 349         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_mb, 'M');
 350         if (rc)
 351                 return rc;
 352
 353         pages_number = round_up(ra_max_mb, 1024 * 1024) >> PAGE_SHIFT;
 354         if (pages_number < 0 || pages_number > totalram_pages / 2) {
 355                 /* 1/2 of RAM */
 356                 CERROR("%s: can't set max_readahead_mb=%llu > %luMB\n",
 357                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 358                        PAGES_TO_MiB(totalram_pages));
 359                 return -ERANGE;
 360         }
 361
 362         spin_lock(&sbi->ll_lock);
 363         sbi->ll_ra_info.ra_max_pages = pages_number;
 364         spin_unlock(&sbi->ll_lock);
 365
 366         return count;
 367 }
 368
 369 LDEBUGFS_SEQ_FOPS(ll_max_readahead_mb);
 370
 371 static int ll_max_readahead_per_file_mb_seq_show(struct seq_file *m, void *v)
 372 {
 373         struct super_block *sb = m->private;
 374         struct ll_sb_info *sbi = ll_s2sbi(sb);
 375         unsigned long ra_max_file_mb;
 376
 377         spin_lock(&sbi->ll_lock);
 378         ra_max_file_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file);
 379         spin_unlock(&sbi->ll_lock);
 380
 381         seq_printf(m, "%lu\n", ra_max_file_mb);
 382         return 0;
 383 }
 384
 385 static ssize_t
 386 ll_max_readahead_per_file_mb_seq_write(struct file *file,
 387                                        const char __user *buffer,
 388                                        size_t count, loff_t *off)
 389 {
 390         struct seq_file *m = file->private_data;
 391         struct super_block *sb = m->private;
 392         struct ll_sb_info *sbi = ll_s2sbi(sb);
 393         s64 ra_max_file_mb, pages_number;
 394         int rc;
 395
 396         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_file_mb,
 397                                            'M');
 398         if (rc)
 399                 return rc;
 400
 401         pages_number = round_up(ra_max_file_mb, 1024 * 1024) >> PAGE_SHIFT;
 402         if (pages_number < 0 || pages_number > sbi->ll_ra_info.ra_max_pages) {
 403                 CERROR("%s: can't set max_readahead_per_file_mb=%llu > max_read_ahead_mb=%lu\n",
 404                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 405                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages));
 406                 return -ERANGE;
 407         }
 408
 409         spin_lock(&sbi->ll_lock);
 410         sbi->ll_ra_info.ra_max_pages_per_file = pages_number;
 411         spin_unlock(&sbi->ll_lock);
 412
 413         return count;
 414 }
 415
 416 LDEBUGFS_SEQ_FOPS(ll_max_readahead_per_file_mb);
 417
 418 static int ll_max_read_ahead_whole_mb_seq_show(struct seq_file *m, void *v)
 419 {
 420         struct super_block *sb = m->private;
 421         struct ll_sb_info *sbi = ll_s2sbi(sb);
 422         unsigned long ra_max_whole_mb;
 423
 424         spin_lock(&sbi->ll_lock);
 425         ra_max_whole_mb = PAGES_TO_MiB(sbi->ll_ra_info.ra_max_read_ahead_whole_pages);
 426         spin_unlock(&sbi->ll_lock);
 427
 428         seq_printf(m, "%lu\n", ra_max_whole_mb);
 429         return 0;
 430 }
 431
 432 static ssize_t
 433 ll_max_read_ahead_whole_mb_seq_write(struct file *file,
 434                                      const char __user *buffer,
 435                                      size_t count, loff_t *off)
 436 {
 437         struct seq_file *m = file->private_data;
 438         struct super_block *sb = m->private;
 439         struct ll_sb_info *sbi = ll_s2sbi(sb);
 440         s64 ra_max_whole_mb, pages_number;
 441         int rc;
 442
 443         rc = lprocfs_str_with_units_to_s64(buffer, count, &ra_max_whole_mb,
 444                                            'M');
 445         if (rc)
 446                 return rc;
 447
 448         pages_number = round_up(ra_max_whole_mb, 1024 * 1024) >> PAGE_SHIFT;
 449         /* Cap this at the current max readahead window size, the readahead
 450          * algorithm does this anyway so it's pointless to set it larger.
 451          */
 452         if (pages_number < 0 ||
 453             pages_number > sbi->ll_ra_info.ra_max_pages_per_file) {
 454                 CERROR("%s: can't set max_read_ahead_whole_mb=%llu > max_read_ahead_per_file_mb=%lu\n",
 455                        sbi->ll_fsname, PAGES_TO_MiB(pages_number),
 456                        PAGES_TO_MiB(sbi->ll_ra_info.ra_max_pages_per_file));
 457                 return -ERANGE;
 458         }
 459
 460         spin_lock(&sbi->ll_lock);
 461         sbi->ll_ra_info.ra_max_read_ahead_whole_pages = pages_number;
 462         spin_unlock(&sbi->ll_lock);
 463
 464         return count;
 465 }
 466
 467 LDEBUGFS_SEQ_FOPS(ll_max_read_ahead_whole_mb);
 468
 469 static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
 470 {
 471         struct super_block     *sb    = m->private;
 472         struct ll_sb_info      *sbi   = ll_s2sbi(sb);
 473         struct cl_client_cache *cache = sbi->ll_cache;
 474         long max_cached_mb;
 475         long unused_mb;
 476
 477         max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
 478         unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
 479         seq_printf(m, "users: %d\n"
 480                       "max_cached_mb: %ld\n"
 481                       "used_mb: %ld\n"
 482                       "unused_mb: %ld\n"
 483                       "reclaim_count: %u\n",
 484                    atomic_read(&cache->ccc_users),
 485                    max_cached_mb,
 486                    max_cached_mb - unused_mb,
 487                    unused_mb,
 488                    cache->ccc_lru_shrinkers);
 489         return 0;
 490 }
 491
 492 static ssize_t ll_max_cached_mb_seq_write(struct file *file,
 493                                           const char __user *buffer,
 494                                           size_t count, loff_t *off)
 495 {
 496         struct seq_file *m = file->private_data;
 497         struct super_block *sb = m->private;
 498         struct ll_sb_info *sbi = ll_s2sbi(sb);
 499         struct cl_client_cache *cache = sbi->ll_cache;
 500         struct lu_env *env;
 501         long diff = 0;
 502         long nrpages = 0;
 503         __u16 refcheck;
 504         __s64 pages_number;
 505         int rc;
 506         char kernbuf[128];
 507
 508         ENTRY;
 509         if (count >= sizeof(kernbuf))
 510                 RETURN(-EINVAL);
 511
 512         if (copy_from_user(kernbuf, buffer, count))
 513                 RETURN(-EFAULT);
 514         kernbuf[count] = 0;
 515
 516         buffer += lprocfs_find_named_value(kernbuf, "max_cached_mb:", &count) -
 517                   kernbuf;
 518         rc = lprocfs_str_with_units_to_s64(buffer, count, &pages_number, 'M');
 519         if (rc)
 520                 RETURN(rc);
 521
 522         pages_number >>= PAGE_SHIFT;
 523
 524         if (pages_number < 0 || pages_number > totalram_pages) {
 525                 CERROR("%s: can't set max cache more than %lu MB\n",
 526                        sbi->ll_fsname,
 527                        PAGES_TO_MiB(totalram_pages));
 528                 RETURN(-ERANGE);
 529         }
 530         /* Allow enough cache so clients can make well-formed RPCs */
 531         pages_number = max_t(long, pages_number, PTLRPC_MAX_BRW_PAGES);
 532
 533         spin_lock(&sbi->ll_lock);
 534         diff = pages_number - cache->ccc_lru_max;
 535         spin_unlock(&sbi->ll_lock);
 536
 537         /* easy - add more LRU slots. */
 538         if (diff >= 0) {
 539                 atomic_long_add(diff, &cache->ccc_lru_left);
 540                 GOTO(out, rc = 0);
 541         }
 542
 543         env = cl_env_get(&refcheck);
 544         if (IS_ERR(env))
 545                 RETURN(PTR_ERR(env));
 546
 547         diff = -diff;
 548         while (diff > 0) {
 549                 long tmp;
 550
 551                 /* reduce LRU budget from free slots. */
 552                 do {
 553                         long ov, nv;
 554
 555                         ov = atomic_long_read(&cache->ccc_lru_left);
 556                         if (ov == 0)
 557                                 break;
 558
 559                         nv = ov > diff ? ov - diff : 0;
 560                         rc = atomic_long_cmpxchg(&cache->ccc_lru_left, ov, nv);
 561                         if (likely(ov == rc)) {
 562                                 diff -= ov - nv;
 563                                 nrpages += ov - nv;
 564                                 break;
 565                         }
 566                 } while (1);
 567
 568                 if (diff <= 0)
 569                         break;
 570
 571                 if (sbi->ll_dt_exp == NULL) { /* being initialized */
 572                         rc = -ENODEV;
 573                         break;
 574                 }
 575
 576                 /* difficult - have to ask OSCs to drop LRU slots. */
 577                 tmp = diff << 1;
 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                 spin_lock(&sbi->ll_lock);
 590                 cache->ccc_lru_max = pages_number;
 591                 spin_unlock(&sbi->ll_lock);
 592                 rc = count;
 593         } else {
 594                 atomic_long_add(nrpages, &cache->ccc_lru_left);
 595         }
 596         return rc;
 597 }
 598
 599 LDEBUGFS_SEQ_FOPS(ll_max_cached_mb);
 600
 601 static ssize_t checksums_show(struct kobject *kobj, struct attribute *attr,
 602                               char *buf)
 603 {
 604         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 605                                               ll_kset.kobj);
 606
 607         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_CHECKSUM) ? 1 : 0);
 608 }
 609
 610 static ssize_t checksums_store(struct kobject *kobj, struct attribute *attr,
 611                                const char *buffer, size_t count)
 612 {
 613         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 614                                               ll_kset.kobj);
 615         bool val;
 616         int tmp;
 617         int rc;
 618
 619         if (!sbi->ll_dt_exp)
 620                 /* Not set up yet */
 621                 return -EAGAIN;
 622
 623         rc = kstrtobool(buffer, &val);
 624         if (rc)
 625                 return rc;
 626         if (val)
 627                 sbi->ll_flags |= LL_SBI_CHECKSUM;
 628         else
 629                 sbi->ll_flags &= ~LL_SBI_CHECKSUM;
 630         tmp = val;
 631
 632         rc = obd_set_info_async(NULL, sbi->ll_dt_exp, sizeof(KEY_CHECKSUM),
 633                                 KEY_CHECKSUM, sizeof(tmp), &tmp, NULL);
 634         if (rc)
 635                 CWARN("Failed to set OSC checksum flags: %d\n", rc);
 636
 637         return count;
 638 }
 639 LUSTRE_RW_ATTR(checksums);
 640
 641 LUSTRE_ATTR(checksum_pages, 0644, checksums_show, checksums_store);
 642
 643 static ssize_t ll_rd_track_id(struct kobject *kobj, char *buf,
 644                               enum stats_track_type type)
 645 {
 646         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 647                                               ll_kset.kobj);
 648
 649         if (sbi->ll_stats_track_type == type)
 650                 return sprintf(buf, "%d\n", sbi->ll_stats_track_id);
 651         else if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
 652                 return sprintf(buf, "0 (all)\n");
 653
 654         return sprintf(buf, "untracked\n");
 655 }
 656
 657 static ssize_t ll_wr_track_id(struct kobject *kobj, const char *buffer,
 658                               size_t count, enum stats_track_type type)
 659 {
 660         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 661                                               ll_kset.kobj);
 662         unsigned long pid;
 663         int rc;
 664
 665         rc = kstrtoul(buffer, 10, &pid);
 666         if (rc)
 667                 return rc;
 668
 669         sbi->ll_stats_track_id = pid;
 670         if (pid == 0)
 671                 sbi->ll_stats_track_type = STATS_TRACK_ALL;
 672         else
 673                 sbi->ll_stats_track_type = type;
 674         lprocfs_clear_stats(sbi->ll_stats);
 675         return count;
 676 }
 677
 678 static ssize_t stats_track_pid_show(struct kobject *kobj,
 679                                     struct attribute *attr,
 680                                     char *buf)
 681 {
 682         return ll_rd_track_id(kobj, buf, STATS_TRACK_PID);
 683 }
 684
 685 static ssize_t stats_track_pid_store(struct kobject *kobj,
 686                                      struct attribute *attr,
 687                                      const char *buffer,
 688                                      size_t count)
 689 {
 690         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PID);
 691 }
 692 LUSTRE_RW_ATTR(stats_track_pid);
 693
 694 static ssize_t stats_track_ppid_show(struct kobject *kobj,
 695                                      struct attribute *attr,
 696                                      char *buf)
 697 {
 698         return ll_rd_track_id(kobj, buf, STATS_TRACK_PPID);
 699 }
 700
 701 static ssize_t stats_track_ppid_store(struct kobject *kobj,
 702                                       struct attribute *attr,
 703                                       const char *buffer,
 704                                       size_t count)
 705 {
 706         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_PPID);
 707 }
 708 LUSTRE_RW_ATTR(stats_track_ppid);
 709
 710 static ssize_t stats_track_gid_show(struct kobject *kobj,
 711                                     struct attribute *attr,
 712                                     char *buf)
 713 {
 714         return ll_rd_track_id(kobj, buf, STATS_TRACK_GID);
 715 }
 716
 717 static ssize_t stats_track_gid_store(struct kobject *kobj,
 718                                      struct attribute *attr,
 719                                      const char *buffer,
 720                                      size_t count)
 721 {
 722         return ll_wr_track_id(kobj, buffer, count, STATS_TRACK_GID);
 723 }
 724 LUSTRE_RW_ATTR(stats_track_gid);
 725
 726 static ssize_t statahead_running_max_show(struct kobject *kobj,
 727                                           struct attribute *attr,
 728                                           char *buf)
 729 {
 730         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 731                                               ll_kset.kobj);
 732
 733         return snprintf(buf, 16, "%u\n", sbi->ll_sa_running_max);
 734 }
 735
 736 static ssize_t statahead_running_max_store(struct kobject *kobj,
 737                                            struct attribute *attr,
 738                                            const char *buffer,
 739                                            size_t count)
 740 {
 741         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 742                                               ll_kset.kobj);
 743         unsigned long val;
 744         int rc;
 745
 746         rc = kstrtoul(buffer, 0, &val);
 747         if (rc)
 748                 return rc;
 749
 750         if (val <= LL_SA_RUNNING_MAX) {
 751                 sbi->ll_sa_running_max = val;
 752                 return count;
 753         }
 754
 755         CERROR("Bad statahead_running_max value %lu. Valid values "
 756                "are in the range [0, %d]\n", val, LL_SA_RUNNING_MAX);
 757
 758         return -ERANGE;
 759 }
 760 LUSTRE_RW_ATTR(statahead_running_max);
 761
 762 static ssize_t statahead_max_show(struct kobject *kobj,
 763                                   struct attribute *attr,
 764                                   char *buf)
 765 {
 766         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 767                                               ll_kset.kobj);
 768
 769         return sprintf(buf, "%u\n", sbi->ll_sa_max);
 770 }
 771
 772 static ssize_t statahead_max_store(struct kobject *kobj,
 773                                    struct attribute *attr,
 774                                    const char *buffer,
 775                                    size_t count)
 776 {
 777         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 778                                               ll_kset.kobj);
 779         unsigned long val;
 780         int rc;
 781
 782         rc = kstrtoul(buffer, 0, &val);
 783         if (rc)
 784                 return rc;
 785
 786         if (val <= LL_SA_RPC_MAX)
 787                 sbi->ll_sa_max = val;
 788         else
 789                 CERROR("Bad statahead_max value %lu. Valid values are in the range [0, %d]\n",
 790                        val, LL_SA_RPC_MAX);
 791
 792         return count;
 793 }
 794 LUSTRE_RW_ATTR(statahead_max);
 795
 796 static ssize_t statahead_agl_show(struct kobject *kobj,
 797                                   struct attribute *attr,
 798                                   char *buf)
 799 {
 800         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 801                                               ll_kset.kobj);
 802
 803         return sprintf(buf, "%u\n", sbi->ll_flags & LL_SBI_AGL_ENABLED ? 1 : 0);
 804 }
 805
 806 static ssize_t statahead_agl_store(struct kobject *kobj,
 807                                    struct attribute *attr,
 808                                    const char *buffer,
 809                                    size_t count)
 810 {
 811         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 812                                               ll_kset.kobj);
 813         bool val;
 814         int rc;
 815
 816         rc = kstrtobool(buffer, &val);
 817         if (rc)
 818                 return rc;
 819
 820         if (val)
 821                 sbi->ll_flags |= LL_SBI_AGL_ENABLED;
 822         else
 823                 sbi->ll_flags &= ~LL_SBI_AGL_ENABLED;
 824
 825         return count;
 826 }
 827 LUSTRE_RW_ATTR(statahead_agl);
 828
 829 static int ll_statahead_stats_seq_show(struct seq_file *m, void *v)
 830 {
 831         struct super_block *sb = m->private;
 832         struct ll_sb_info *sbi = ll_s2sbi(sb);
 833
 834         seq_printf(m, "statahead total: %u\n"
 835                       "statahead wrong: %u\n"
 836                       "agl total: %u\n",
 837                    atomic_read(&sbi->ll_sa_total),
 838                    atomic_read(&sbi->ll_sa_wrong),
 839                    atomic_read(&sbi->ll_agl_total));
 840         return 0;
 841 }
 842
 843 LDEBUGFS_SEQ_FOPS_RO(ll_statahead_stats);
 844
 845 static ssize_t lazystatfs_show(struct kobject *kobj,
 846                                struct attribute *attr,
 847                                char *buf)
 848 {
 849         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 850                                               ll_kset.kobj);
 851
 852         return sprintf(buf, "%u\n", (sbi->ll_flags & LL_SBI_LAZYSTATFS) ? 1 : 0);
 853 }
 854
 855 static ssize_t lazystatfs_store(struct kobject *kobj,
 856                                 struct attribute *attr,
 857                                 const char *buffer,
 858                                 size_t count)
 859 {
 860         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 861                                               ll_kset.kobj);
 862         bool val;
 863         int rc;
 864
 865         rc = kstrtobool(buffer, &val);
 866         if (rc)
 867                 return rc;
 868
 869         if (val)
 870                 sbi->ll_flags |= LL_SBI_LAZYSTATFS;
 871         else
 872                 sbi->ll_flags &= ~LL_SBI_LAZYSTATFS;
 873
 874         return count;
 875 }
 876 LUSTRE_RW_ATTR(lazystatfs);
 877
 878 static ssize_t max_easize_show(struct kobject *kobj,
 879                                struct attribute *attr,
 880                                char *buf)
 881 {
 882         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 883                                               ll_kset.kobj);
 884         unsigned int ealen;
 885         int rc;
 886
 887         rc = ll_get_max_mdsize(sbi, &ealen);
 888         if (rc)
 889                 return rc;
 890
 891         return sprintf(buf, "%u\n", ealen);
 892 }
 893 LUSTRE_RO_ATTR(max_easize);
 894
 895 /**
 896  * Get default_easize.
 897  *
 898  * \see client_obd::cl_default_mds_easize
 899  *
 900  * \param[in] m         seq_file handle
 901  * \param[in] v         unused for single entry
 902  *
 903  * \retval 0            on success
 904  * \retval negative     negated errno on failure
 905  */
 906 static ssize_t default_easize_show(struct kobject *kobj,
 907                                    struct attribute *attr,
 908                                    char *buf)
 909 {
 910         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 911                                               ll_kset.kobj);
 912         unsigned int ealen;
 913         int rc;
 914
 915         rc = ll_get_default_mdsize(sbi, &ealen);
 916         if (rc)
 917                 return rc;
 918
 919         return sprintf(buf, "%u\n", ealen);
 920 }
 921
 922 /**
 923  * Set default_easize.
 924  *
 925  * Range checking on the passed value is handled by
 926  * ll_set_default_mdsize().
 927  *
 928  * \see client_obd::cl_default_mds_easize
 929  *
 930  * \param[in] file      proc file
 931  * \param[in] buffer    string passed from user space
 932  * \param[in] count     \a buffer length
 933  * \param[in] off       unused for single entry
 934  *
 935  * \retval positive     \a count on success
 936  * \retval negative     negated errno on failure
 937  */
 938 static ssize_t default_easize_store(struct kobject *kobj,
 939                                     struct attribute *attr,
 940                                     const char *buffer,
 941                                     size_t count)
 942 {
 943         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 944                                               ll_kset.kobj);
 945         unsigned int val;
 946         int rc;
 947
 948         if (count == 0)
 949                 return 0;
 950
 951         rc = kstrtouint(buffer, 10, &val);
 952         if (rc)
 953                 return rc;
 954
 955         rc = ll_set_default_mdsize(sbi, val);
 956         if (rc)
 957                 return rc;
 958
 959         return count;
 960 }
 961 LUSTRE_RW_ATTR(default_easize);
 962
 963 static int ll_sbi_flags_seq_show(struct seq_file *m, void *v)
 964 {
 965         const char *str[] = LL_SBI_FLAGS;
 966         struct super_block *sb = m->private;
 967         int flags = ll_s2sbi(sb)->ll_flags;
 968         int i = 0;
 969
 970         while (flags != 0) {
 971                 if (ARRAY_SIZE(str) <= i) {
 972                         CERROR("%s: Revise array LL_SBI_FLAGS to match sbi "
 973                                 "flags please.\n", ll_s2sbi(sb)->ll_fsname);
 974                         return -EINVAL;
 975                 }
 976
 977                 if (flags & 0x1)
 978                         seq_printf(m, "%s ", str[i]);
 979                 flags >>= 1;
 980                 ++i;
 981         }
 982         seq_printf(m, "\b\n");
 983         return 0;
 984 }
 985
 986 LDEBUGFS_SEQ_FOPS_RO(ll_sbi_flags);
 987
 988 static ssize_t xattr_cache_show(struct kobject *kobj,
 989                                 struct attribute *attr,
 990                                 char *buf)
 991 {
 992         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
 993                                               ll_kset.kobj);
 994
 995         return sprintf(buf, "%u\n", sbi->ll_xattr_cache_enabled);
 996 }
 997
 998 static ssize_t xattr_cache_store(struct kobject *kobj,
 999                                  struct attribute *attr,
1000                                  const char *buffer,
1001                                  size_t count)
1002 {
1003         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1004                                               ll_kset.kobj);
1005         bool val;
1006         int rc;
1007
1008         rc = kstrtobool(buffer, &val);
1009         if (rc)
1010                 return rc;
1011
1012         if (val && !(sbi->ll_flags & LL_SBI_XATTR_CACHE))
1013                 return -ENOTSUPP;
1014
1015         sbi->ll_xattr_cache_enabled = val;
1016         sbi->ll_xattr_cache_set = 1;
1017
1018         return count;
1019 }
1020 LUSTRE_RW_ATTR(xattr_cache);
1021
1022 static ssize_t tiny_write_show(struct kobject *kobj,
1023                                struct attribute *attr,
1024                                char *buf)
1025 {
1026         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1027                                               ll_kset.kobj);
1028
1029         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_TINY_WRITE));
1030 }
1031
1032 static ssize_t tiny_write_store(struct kobject *kobj,
1033                                 struct attribute *attr,
1034                                 const char *buffer,
1035                                 size_t count)
1036 {
1037         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1038                                               ll_kset.kobj);
1039         bool val;
1040         int rc;
1041
1042         rc = kstrtobool(buffer, &val);
1043         if (rc)
1044                 return rc;
1045
1046         spin_lock(&sbi->ll_lock);
1047         if (val)
1048                 sbi->ll_flags |= LL_SBI_TINY_WRITE;
1049         else
1050                 sbi->ll_flags &= ~LL_SBI_TINY_WRITE;
1051         spin_unlock(&sbi->ll_lock);
1052
1053         return count;
1054 }
1055 LUSTRE_RW_ATTR(tiny_write);
1056
1057 static ssize_t max_read_ahead_async_active_show(struct kobject *kobj,
1058                                                struct attribute *attr,
1059                                                char *buf)
1060 {
1061         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1062                                               ll_kset.kobj);
1063
1064         return snprintf(buf, PAGE_SIZE, "%u\n",
1065                         sbi->ll_ra_info.ra_async_max_active);
1066 }
1067
1068 static ssize_t max_read_ahead_async_active_store(struct kobject *kobj,
1069                                                 struct attribute *attr,
1070                                                 const char *buffer,
1071                                                 size_t count)
1072 {
1073         unsigned int val;
1074         int rc;
1075         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1076                                               ll_kset.kobj);
1077
1078         rc = kstrtouint(buffer, 10, &val);
1079         if (rc)
1080                 return rc;
1081
1082         if (val < 1 || val > WQ_UNBOUND_MAX_ACTIVE) {
1083                 CERROR("%s: cannot set max_read_ahead_async_active=%u %s than %u\n",
1084                        sbi->ll_fsname, val,
1085                        val < 1 ? "smaller" : "larger",
1086                        val < 1 ? 1 : WQ_UNBOUND_MAX_ACTIVE);
1087                 return -ERANGE;
1088         }
1089
1090         sbi->ll_ra_info.ra_async_max_active = val;
1091         workqueue_set_max_active(sbi->ll_ra_info.ll_readahead_wq, val);
1092
1093         return count;
1094 }
1095 LUSTRE_RW_ATTR(max_read_ahead_async_active);
1096
1097 static ssize_t read_ahead_async_file_threshold_mb_show(struct kobject *kobj,
1098                                                        struct attribute *attr,
1099                                                        char *buf)
1100 {
1101         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1102                                               ll_kset.kobj);
1103
1104         return snprintf(buf, PAGE_SIZE, "%lu\n",
1105              PAGES_TO_MiB(sbi->ll_ra_info.ra_async_pages_per_file_threshold));
1106 }
1107
1108 static ssize_t
1109 read_ahead_async_file_threshold_mb_store(struct kobject *kobj,
1110                                          struct attribute *attr,
1111                                          const char *buffer, size_t count)
1112 {
1113         unsigned long pages_number;
1114         unsigned long max_ra_per_file;
1115         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1116                                               ll_kset.kobj);
1117         int rc;
1118
1119         rc = kstrtoul(buffer, 10, &pages_number);
1120         if (rc)
1121                 return rc;
1122
1123         pages_number = MiB_TO_PAGES(pages_number);
1124         max_ra_per_file = sbi->ll_ra_info.ra_max_pages_per_file;
1125         if (pages_number < 0 || pages_number > max_ra_per_file) {
1126                 CERROR("%s: can't set read_ahead_async_file_threshold_mb=%lu > "
1127                        "max_read_readahead_per_file_mb=%lu\n", sbi->ll_fsname,
1128                        PAGES_TO_MiB(pages_number),
1129                        PAGES_TO_MiB(max_ra_per_file));
1130                 return -ERANGE;
1131         }
1132         sbi->ll_ra_info.ra_async_pages_per_file_threshold = pages_number;
1133
1134         return count;
1135 }
1136 LUSTRE_RW_ATTR(read_ahead_async_file_threshold_mb);
1137
1138 static ssize_t fast_read_show(struct kobject *kobj,
1139                               struct attribute *attr,
1140                               char *buf)
1141 {
1142         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1143                                               ll_kset.kobj);
1144
1145         return sprintf(buf, "%u\n", !!(sbi->ll_flags & LL_SBI_FAST_READ));
1146 }
1147
1148 static ssize_t fast_read_store(struct kobject *kobj,
1149                                struct attribute *attr,
1150                                const char *buffer,
1151                                size_t count)
1152 {
1153         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1154                                               ll_kset.kobj);
1155         bool val;
1156         int rc;
1157
1158         rc = kstrtobool(buffer, &val);
1159         if (rc)
1160                 return rc;
1161
1162         spin_lock(&sbi->ll_lock);
1163         if (val)
1164                 sbi->ll_flags |= LL_SBI_FAST_READ;
1165         else
1166                 sbi->ll_flags &= ~LL_SBI_FAST_READ;
1167         spin_unlock(&sbi->ll_lock);
1168
1169         return count;
1170 }
1171 LUSTRE_RW_ATTR(fast_read);
1172
1173 static ssize_t file_heat_show(struct kobject *kobj,
1174                               struct attribute *attr,
1175                               char *buf)
1176 {
1177         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1178                                               ll_kset.kobj);
1179
1180         return snprintf(buf, PAGE_SIZE, "%u\n",
1181                         !!(sbi->ll_flags & LL_SBI_FILE_HEAT));
1182 }
1183
1184 static ssize_t file_heat_store(struct kobject *kobj,
1185                                struct attribute *attr,
1186                                const char *buffer,
1187                                size_t count)
1188 {
1189         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1190                                               ll_kset.kobj);
1191         bool val;
1192         int rc;
1193
1194         rc = kstrtobool(buffer, &val);
1195         if (rc)
1196                 return rc;
1197
1198         spin_lock(&sbi->ll_lock);
1199         if (val)
1200                 sbi->ll_flags |= LL_SBI_FILE_HEAT;
1201         else
1202                 sbi->ll_flags &= ~LL_SBI_FILE_HEAT;
1203         spin_unlock(&sbi->ll_lock);
1204
1205         return count;
1206 }
1207 LUSTRE_RW_ATTR(file_heat);
1208
1209 static ssize_t heat_decay_percentage_show(struct kobject *kobj,
1210                                           struct attribute *attr,
1211                                           char *buf)
1212 {
1213         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1214                                               ll_kset.kobj);
1215
1216         return snprintf(buf, PAGE_SIZE, "%u\n",
1217                        (sbi->ll_heat_decay_weight * 100 + 128) / 256);
1218 }
1219
1220 static ssize_t heat_decay_percentage_store(struct kobject *kobj,
1221                                            struct attribute *attr,
1222                                            const char *buffer,
1223                                            size_t count)
1224 {
1225         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1226                                               ll_kset.kobj);
1227         unsigned long val;
1228         int rc;
1229
1230         rc = kstrtoul(buffer, 10, &val);
1231         if (rc)
1232                 return rc;
1233
1234         if (val < 0 || val > 100)
1235                 return -ERANGE;
1236
1237         sbi->ll_heat_decay_weight = (val * 256 + 50) / 100;
1238
1239         return count;
1240 }
1241 LUSTRE_RW_ATTR(heat_decay_percentage);
1242
1243 static ssize_t heat_period_second_show(struct kobject *kobj,
1244                                        struct attribute *attr,
1245                                        char *buf)
1246 {
1247         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1248                                               ll_kset.kobj);
1249
1250         return snprintf(buf, PAGE_SIZE, "%u\n", sbi->ll_heat_period_second);
1251 }
1252
1253 static ssize_t heat_period_second_store(struct kobject *kobj,
1254                                         struct attribute *attr,
1255                                         const char *buffer,
1256                                         size_t count)
1257 {
1258         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1259                                               ll_kset.kobj);
1260         unsigned long val;
1261         int rc;
1262
1263         rc = kstrtoul(buffer, 10, &val);
1264         if (rc)
1265                 return rc;
1266
1267         if (val <= 0)
1268                 return -ERANGE;
1269
1270         sbi->ll_heat_period_second = val;
1271
1272         return count;
1273 }
1274 LUSTRE_RW_ATTR(heat_period_second);
1275
1276 static int ll_unstable_stats_seq_show(struct seq_file *m, void *v)
1277 {
1278         struct super_block      *sb    = m->private;
1279         struct ll_sb_info       *sbi   = ll_s2sbi(sb);
1280         struct cl_client_cache  *cache = sbi->ll_cache;
1281         long pages;
1282         int mb;
1283
1284         pages = atomic_long_read(&cache->ccc_unstable_nr);
1285         mb    = (pages * PAGE_SIZE) >> 20;
1286
1287         seq_printf(m, "unstable_check:     %8d\n"
1288                       "unstable_pages: %12ld\n"
1289                       "unstable_mb:        %8d\n",
1290                    cache->ccc_unstable_check, pages, mb);
1291         return 0;
1292 }
1293
1294 static ssize_t ll_unstable_stats_seq_write(struct file *file,
1295                                            const char __user *buffer,
1296                                            size_t count, loff_t *unused)
1297 {
1298         struct seq_file *seq = file->private_data;
1299         struct ll_sb_info *sbi = ll_s2sbi((struct super_block *)seq->private);
1300         char kernbuf[128];
1301         bool val;
1302         int rc;
1303
1304         if (count == 0)
1305                 return 0;
1306         if (count >= sizeof(kernbuf))
1307                 return -EINVAL;
1308
1309         if (copy_from_user(kernbuf, buffer, count))
1310                 return -EFAULT;
1311         kernbuf[count] = 0;
1312
1313         buffer += lprocfs_find_named_value(kernbuf, "unstable_check:", &count) -
1314                   kernbuf;
1315         rc = kstrtobool_from_user(buffer, count, &val);
1316         if (rc < 0)
1317                 return rc;
1318
1319         /* borrow lru lock to set the value */
1320         spin_lock(&sbi->ll_cache->ccc_lru_lock);
1321         sbi->ll_cache->ccc_unstable_check = val;
1322         spin_unlock(&sbi->ll_cache->ccc_lru_lock);
1323
1324         return count;
1325 }
1326
1327 LDEBUGFS_SEQ_FOPS(ll_unstable_stats);
1328
1329 static int ll_root_squash_seq_show(struct seq_file *m, void *v)
1330 {
1331         struct super_block *sb = m->private;
1332         struct ll_sb_info *sbi = ll_s2sbi(sb);
1333         struct root_squash_info *squash = &sbi->ll_squash;
1334
1335         seq_printf(m, "%u:%u\n", squash->rsi_uid, squash->rsi_gid);
1336         return 0;
1337 }
1338
1339 static ssize_t ll_root_squash_seq_write(struct file *file,
1340                                         const char __user *buffer,
1341                                         size_t count, loff_t *off)
1342 {
1343         struct seq_file *m = file->private_data;
1344         struct super_block *sb = m->private;
1345         struct ll_sb_info *sbi = ll_s2sbi(sb);
1346         struct root_squash_info *squash = &sbi->ll_squash;
1347
1348         return lprocfs_wr_root_squash(buffer, count, squash, sbi->ll_fsname);
1349 }
1350
1351 LDEBUGFS_SEQ_FOPS(ll_root_squash);
1352
1353 static int ll_nosquash_nids_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         int len;
1359
1360         down_read(&squash->rsi_sem);
1361         if (!list_empty(&squash->rsi_nosquash_nids)) {
1362                 len = cfs_print_nidlist(m->buf + m->count, m->size - m->count,
1363                                         &squash->rsi_nosquash_nids);
1364                 m->count += len;
1365                 seq_putc(m, '\n');
1366         } else {
1367                 seq_puts(m, "NONE\n");
1368         }
1369         up_read(&squash->rsi_sem);
1370
1371         return 0;
1372 }
1373
1374 static ssize_t ll_nosquash_nids_seq_write(struct file *file,
1375                                           const char __user *buffer,
1376                                           size_t count, loff_t *off)
1377 {
1378         struct seq_file *m = file->private_data;
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 rc;
1383
1384         rc = lprocfs_wr_nosquash_nids(buffer, count, squash, sbi->ll_fsname);
1385         if (rc < 0)
1386                 return rc;
1387
1388         ll_compute_rootsquash_state(sbi);
1389
1390         return rc;
1391 }
1392
1393 LDEBUGFS_SEQ_FOPS(ll_nosquash_nids);
1394
1395 static int ll_pcc_seq_show(struct seq_file *m, void *v)
1396 {
1397         struct super_block *sb = m->private;
1398         struct ll_sb_info *sbi = ll_s2sbi(sb);
1399
1400         return pcc_super_dump(&sbi->ll_pcc_super, m);
1401 }
1402
1403 static ssize_t ll_pcc_seq_write(struct file *file, const char __user *buffer,
1404                                 size_t count, loff_t *off)
1405 {
1406         struct seq_file *m = file->private_data;
1407         struct super_block *sb = m->private;
1408         struct ll_sb_info *sbi = ll_s2sbi(sb);
1409         int rc;
1410         char *kernbuf;
1411
1412         if (count >= LPROCFS_WR_PCC_MAX_CMD)
1413                 return -EINVAL;
1414
1415         if (!(exp_connect_flags2(sbi->ll_md_exp) & OBD_CONNECT2_PCC))
1416                 return -EOPNOTSUPP;
1417
1418         OBD_ALLOC(kernbuf, count + 1);
1419         if (kernbuf == NULL)
1420                 return -ENOMEM;
1421
1422         if (copy_from_user(kernbuf, buffer, count))
1423                 GOTO(out_free_kernbuff, rc = -EFAULT);
1424
1425         rc = pcc_cmd_handle(kernbuf, count, &sbi->ll_pcc_super);
1426 out_free_kernbuff:
1427         OBD_FREE(kernbuf, count + 1);
1428         return rc ? rc : count;
1429 }
1430 LPROC_SEQ_FOPS(ll_pcc);
1431
1432 struct lprocfs_vars lprocfs_llite_obd_vars[] = {
1433         { .name =       "site",
1434           .fops =       &ll_site_stats_fops                     },
1435         { .name =       "max_read_ahead_mb",
1436           .fops =       &ll_max_readahead_mb_fops               },
1437         { .name =       "max_read_ahead_per_file_mb",
1438           .fops =       &ll_max_readahead_per_file_mb_fops      },
1439         { .name =       "max_read_ahead_whole_mb",
1440           .fops =       &ll_max_read_ahead_whole_mb_fops        },
1441         { .name =       "max_cached_mb",
1442           .fops =       &ll_max_cached_mb_fops                  },
1443         { .name =       "statahead_stats",
1444           .fops =       &ll_statahead_stats_fops                },
1445         { .name =       "unstable_stats",
1446           .fops =       &ll_unstable_stats_fops                 },
1447         { .name =       "sbi_flags",
1448           .fops =       &ll_sbi_flags_fops                      },
1449         { .name =       "root_squash",
1450           .fops =       &ll_root_squash_fops                    },
1451         { .name =       "nosquash_nids",
1452           .fops =       &ll_nosquash_nids_fops                  },
1453         { .name =       "pcc",
1454           .fops =       &ll_pcc_fops,                           },
1455         { NULL }
1456 };
1457
1458 #define MAX_STRING_SIZE 128
1459
1460 static struct attribute *llite_attrs[] = {
1461         &lustre_attr_blocksize.attr,
1462         &lustre_attr_stat_blocksize.attr,
1463         &lustre_attr_kbytestotal.attr,
1464         &lustre_attr_kbytesfree.attr,
1465         &lustre_attr_kbytesavail.attr,
1466         &lustre_attr_filestotal.attr,
1467         &lustre_attr_filesfree.attr,
1468         &lustre_attr_client_type.attr,
1469         &lustre_attr_fstype.attr,
1470         &lustre_attr_uuid.attr,
1471         &lustre_attr_checksums.attr,
1472         &lustre_attr_checksum_pages.attr,
1473         &lustre_attr_stats_track_pid.attr,
1474         &lustre_attr_stats_track_ppid.attr,
1475         &lustre_attr_stats_track_gid.attr,
1476         &lustre_attr_statahead_running_max.attr,
1477         &lustre_attr_statahead_max.attr,
1478         &lustre_attr_statahead_agl.attr,
1479         &lustre_attr_lazystatfs.attr,
1480         &lustre_attr_max_easize.attr,
1481         &lustre_attr_default_easize.attr,
1482         &lustre_attr_xattr_cache.attr,
1483         &lustre_attr_fast_read.attr,
1484         &lustre_attr_tiny_write.attr,
1485         &lustre_attr_file_heat.attr,
1486         &lustre_attr_heat_decay_percentage.attr,
1487         &lustre_attr_heat_period_second.attr,
1488         &lustre_attr_max_read_ahead_async_active.attr,
1489         &lustre_attr_read_ahead_async_file_threshold_mb.attr,
1490         NULL,
1491 };
1492
1493 static void sbi_kobj_release(struct kobject *kobj)
1494 {
1495         struct ll_sb_info *sbi = container_of(kobj, struct ll_sb_info,
1496                                               ll_kset.kobj);
1497         complete(&sbi->ll_kobj_unregister);
1498 }
1499
1500 static struct kobj_type sbi_ktype = {
1501         .default_attrs  = llite_attrs,
1502         .sysfs_ops      = &lustre_sysfs_ops,
1503         .release        = sbi_kobj_release,
1504 };
1505
1506 static const struct llite_file_opcode {
1507         __u32       opcode;
1508         __u32       type;
1509         const char *opname;
1510 } llite_opcode_table[LPROC_LL_FILE_OPCODES] = {
1511         /* file operation */
1512         { LPROC_LL_DIRTY_HITS,     LPROCFS_TYPE_REGS, "dirty_pages_hits" },
1513         { LPROC_LL_DIRTY_MISSES,   LPROCFS_TYPE_REGS, "dirty_pages_misses" },
1514         { LPROC_LL_READ_BYTES,     LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1515                                    "read_bytes" },
1516         { LPROC_LL_WRITE_BYTES,    LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_BYTES,
1517                                    "write_bytes" },
1518         { LPROC_LL_BRW_READ,       LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1519                                    "brw_read" },
1520         { LPROC_LL_BRW_WRITE,      LPROCFS_CNTR_AVGMINMAX|LPROCFS_TYPE_PAGES,
1521                                    "brw_write" },
1522         { LPROC_LL_IOCTL,          LPROCFS_TYPE_REGS, "ioctl" },
1523         { LPROC_LL_OPEN,           LPROCFS_TYPE_REGS, "open" },
1524         { LPROC_LL_RELEASE,        LPROCFS_TYPE_REGS, "close" },
1525         { LPROC_LL_MAP,            LPROCFS_TYPE_REGS, "mmap" },
1526         { LPROC_LL_FAULT,          LPROCFS_TYPE_REGS, "page_fault" },
1527         { LPROC_LL_MKWRITE,        LPROCFS_TYPE_REGS, "page_mkwrite" },
1528         { LPROC_LL_LLSEEK,         LPROCFS_TYPE_REGS, "seek" },
1529         { LPROC_LL_FSYNC,          LPROCFS_TYPE_REGS, "fsync" },
1530         { LPROC_LL_READDIR,        LPROCFS_TYPE_REGS, "readdir" },
1531         /* inode operation */
1532         { LPROC_LL_SETATTR,        LPROCFS_TYPE_REGS, "setattr" },
1533         { LPROC_LL_TRUNC,          LPROCFS_TYPE_REGS, "truncate" },
1534         { LPROC_LL_FLOCK,          LPROCFS_TYPE_REGS, "flock" },
1535         { LPROC_LL_GETATTR,        LPROCFS_TYPE_REGS, "getattr" },
1536         /* dir inode operation */
1537         { LPROC_LL_CREATE,         LPROCFS_TYPE_REGS, "create" },
1538         { LPROC_LL_LINK,           LPROCFS_TYPE_REGS, "link" },
1539         { LPROC_LL_UNLINK,         LPROCFS_TYPE_REGS, "unlink" },
1540         { LPROC_LL_SYMLINK,        LPROCFS_TYPE_REGS, "symlink" },
1541         { LPROC_LL_MKDIR,          LPROCFS_TYPE_REGS, "mkdir" },
1542         { LPROC_LL_RMDIR,          LPROCFS_TYPE_REGS, "rmdir" },
1543         { LPROC_LL_MKNOD,          LPROCFS_TYPE_REGS, "mknod" },
1544         { LPROC_LL_RENAME,         LPROCFS_TYPE_REGS, "rename" },
1545         /* special inode operation */
1546         { LPROC_LL_STAFS,          LPROCFS_TYPE_REGS, "statfs" },
1547         { LPROC_LL_ALLOC_INODE,    LPROCFS_TYPE_REGS, "alloc_inode" },
1548         { LPROC_LL_SETXATTR,       LPROCFS_TYPE_REGS, "setxattr" },
1549         { LPROC_LL_GETXATTR,       LPROCFS_TYPE_REGS, "getxattr" },
1550         { LPROC_LL_GETXATTR_HITS,  LPROCFS_TYPE_REGS, "getxattr_hits" },
1551         { LPROC_LL_LISTXATTR,      LPROCFS_TYPE_REGS, "listxattr" },
1552         { LPROC_LL_REMOVEXATTR,    LPROCFS_TYPE_REGS, "removexattr" },
1553         { LPROC_LL_INODE_PERM,     LPROCFS_TYPE_REGS, "inode_permission" },
1554 };
1555
1556 void ll_stats_ops_tally(struct ll_sb_info *sbi, int op, int count)
1557 {
1558         if (!sbi->ll_stats)
1559                 return;
1560         if (sbi->ll_stats_track_type == STATS_TRACK_ALL)
1561                 lprocfs_counter_add(sbi->ll_stats, op, count);
1562         else if (sbi->ll_stats_track_type == STATS_TRACK_PID &&
1563                  sbi->ll_stats_track_id == current->pid)
1564                 lprocfs_counter_add(sbi->ll_stats, op, count);
1565         else if (sbi->ll_stats_track_type == STATS_TRACK_PPID &&
1566                  sbi->ll_stats_track_id == current->parent->pid)
1567                 lprocfs_counter_add(sbi->ll_stats, op, count);
1568         else if (sbi->ll_stats_track_type == STATS_TRACK_GID &&
1569                  sbi->ll_stats_track_id ==
1570                         from_kgid(&init_user_ns, current_gid()))
1571                 lprocfs_counter_add(sbi->ll_stats, op, count);
1572 }
1573 EXPORT_SYMBOL(ll_stats_ops_tally);
1574
1575 static const char *ra_stat_string[] = {
1576         [RA_STAT_HIT] = "hits",
1577         [RA_STAT_MISS] = "misses",
1578         [RA_STAT_DISTANT_READPAGE] = "readpage not consecutive",
1579         [RA_STAT_MISS_IN_WINDOW] = "miss inside window",
1580         [RA_STAT_FAILED_GRAB_PAGE] = "failed grab_cache_page",
1581         [RA_STAT_FAILED_MATCH] = "failed lock match",
1582         [RA_STAT_DISCARDED] = "read but discarded",
1583         [RA_STAT_ZERO_LEN] = "zero length file",
1584         [RA_STAT_ZERO_WINDOW] = "zero size window",
1585         [RA_STAT_EOF] = "read-ahead to EOF",
1586         [RA_STAT_MAX_IN_FLIGHT] = "hit max r-a issue",
1587         [RA_STAT_WRONG_GRAB_PAGE] = "wrong page from grab_cache_page",
1588         [RA_STAT_FAILED_REACH_END] = "failed to reach end",
1589         [RA_STAT_ASYNC] = "async readahead",
1590         [RA_STAT_FAILED_FAST_READ] = "failed to fast read",
1591 };
1592
1593 int ll_debugfs_register_super(struct super_block *sb, const char *name)
1594 {
1595         struct lustre_sb_info *lsi = s2lsi(sb);
1596         struct ll_sb_info *sbi = ll_s2sbi(sb);
1597         int err, id, rc;
1598
1599         ENTRY;
1600         LASSERT(sbi);
1601
1602         if (IS_ERR_OR_NULL(llite_root))
1603                 goto out_ll_kset;
1604
1605         sbi->ll_debugfs_entry = ldebugfs_register(name, llite_root,
1606                                                   lprocfs_llite_obd_vars, sb);
1607         if (IS_ERR_OR_NULL(sbi->ll_debugfs_entry)) {
1608                 err = sbi->ll_debugfs_entry ? PTR_ERR(sbi->ll_debugfs_entry) :
1609                                               -ENOMEM;
1610                 sbi->ll_debugfs_entry = NULL;
1611                 RETURN(err);
1612         }
1613
1614         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "dump_page_cache",0444,
1615                                  &vvp_dump_pgcache_file_ops, sbi);
1616         if (rc)
1617                 CWARN("Error adding the dump_page_cache file\n");
1618
1619         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "extents_stats", 0644,
1620                                  &ll_rw_extents_stats_fops, sbi);
1621         if (rc)
1622                 CWARN("Error adding the extent_stats file\n");
1623
1624         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry,
1625                                  "extents_stats_per_process", 0644,
1626                                  &ll_rw_extents_stats_pp_fops, sbi);
1627         if (rc)
1628                 CWARN("Error adding the extents_stats_per_process file\n");
1629
1630         rc = ldebugfs_seq_create(sbi->ll_debugfs_entry, "offset_stats", 0644,
1631                                  &ll_rw_offset_stats_fops, sbi);
1632         if (rc)
1633                 CWARN("Error adding the offset_stats file\n");
1634
1635         /* File operations stats */
1636         sbi->ll_stats = lprocfs_alloc_stats(LPROC_LL_FILE_OPCODES,
1637                                             LPROCFS_STATS_FLAG_NONE);
1638         if (sbi->ll_stats == NULL)
1639                 GOTO(out_debugfs, err = -ENOMEM);
1640
1641         /* do counter init */
1642         for (id = 0; id < LPROC_LL_FILE_OPCODES; id++) {
1643                 u32 type = llite_opcode_table[id].type;
1644                 void *ptr = NULL;
1645
1646                 if (type & LPROCFS_TYPE_REGS)
1647                         ptr = "regs";
1648                 else if (type & LPROCFS_TYPE_BYTES)
1649                         ptr = "bytes";
1650                 else if (type & LPROCFS_TYPE_PAGES)
1651                         ptr = "pages";
1652                 lprocfs_counter_init(sbi->ll_stats,
1653                                      llite_opcode_table[id].opcode,
1654                                      (type & LPROCFS_CNTR_AVGMINMAX),
1655                                      llite_opcode_table[id].opname, ptr);
1656         }
1657
1658         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "stats",
1659                                       sbi->ll_stats);
1660         if (err)
1661                 GOTO(out_stats, err);
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         err = ldebugfs_register_stats(sbi->ll_debugfs_entry, "read_ahead_stats",
1673                                       sbi->ll_ra_stats);
1674         if (err)
1675                 GOTO(out_ra_stats, err);
1676
1677 out_ll_kset:
1678         /* Yes we also register sysfs mount kset here as well */
1679         sbi->ll_kset.kobj.parent = llite_kobj;
1680         sbi->ll_kset.kobj.ktype = &sbi_ktype;
1681         init_completion(&sbi->ll_kobj_unregister);
1682         err = kobject_set_name(&sbi->ll_kset.kobj, "%s", name);
1683         if (err)
1684                 GOTO(out_ra_stats, err);
1685
1686         err = kset_register(&sbi->ll_kset);
1687         if (err)
1688                 GOTO(out_ra_stats, err);
1689
1690         lsi->lsi_kobj = kobject_get(&sbi->ll_kset.kobj);
1691
1692         RETURN(0);
1693 out_ra_stats:
1694         lprocfs_free_stats(&sbi->ll_ra_stats);
1695 out_stats:
1696         lprocfs_free_stats(&sbi->ll_stats);
1697 out_debugfs:
1698         ldebugfs_remove(&sbi->ll_debugfs_entry);
1699
1700         RETURN(err);
1701 }
1702
1703 void ll_debugfs_unregister_super(struct super_block *sb)
1704 {
1705         struct lustre_sb_info *lsi = s2lsi(sb);
1706         struct ll_sb_info *sbi = ll_s2sbi(sb);
1707
1708         if (!IS_ERR_OR_NULL(sbi->ll_debugfs_entry))
1709                 ldebugfs_remove(&sbi->ll_debugfs_entry);
1710
1711         if (sbi->ll_dt_obd)
1712                 sysfs_remove_link(&sbi->ll_kset.kobj,
1713                                   sbi->ll_dt_obd->obd_type->typ_name);
1714
1715         if (sbi->ll_md_obd)
1716                 sysfs_remove_link(&sbi->ll_kset.kobj,
1717                                   sbi->ll_md_obd->obd_type->typ_name);
1718
1719         kobject_put(lsi->lsi_kobj);
1720
1721         kset_unregister(&sbi->ll_kset);
1722         wait_for_completion(&sbi->ll_kobj_unregister);
1723
1724         lprocfs_free_stats(&sbi->ll_ra_stats);
1725         lprocfs_free_stats(&sbi->ll_stats);
1726 }
1727 #undef MAX_STRING_SIZE
1728
1729 static void ll_display_extents_info(struct ll_rw_extents_info *io_extents,
1730                                    struct seq_file *seq, int which)
1731 {
1732         unsigned long read_tot = 0, write_tot = 0, read_cum, write_cum;
1733         unsigned long start, end, r, w;
1734         char *unitp = "KMGTPEZY";
1735         int i, units = 10;
1736         struct per_process_info *pp_info = &io_extents->pp_extents[which];
1737
1738         read_cum = 0;
1739         write_cum = 0;
1740         start = 0;
1741
1742         for(i = 0; i < LL_HIST_MAX; i++) {
1743                 read_tot += pp_info->pp_r_hist.oh_buckets[i];
1744                 write_tot += pp_info->pp_w_hist.oh_buckets[i];
1745         }
1746
1747         for(i = 0; i < LL_HIST_MAX; i++) {
1748                 r = pp_info->pp_r_hist.oh_buckets[i];
1749                 w = pp_info->pp_w_hist.oh_buckets[i];
1750                 read_cum += r;
1751                 write_cum += w;
1752                 end = BIT(i + LL_HIST_START - units);
1753                 seq_printf(seq, "%4lu%c - %4lu%c%c: %14lu %4u %4u  | "
1754                            "%14lu %4u %4u\n", start, *unitp, end, *unitp,
1755                            (i == LL_HIST_MAX - 1) ? '+' : ' ',
1756                            r, pct(r, read_tot), pct(read_cum, read_tot),
1757                            w, pct(w, write_tot), pct(write_cum, write_tot));
1758                 start = end;
1759                 if (start == BIT(10)) {
1760                         start = 1;
1761                         units += 10;
1762                         unitp++;
1763                 }
1764                 if (read_cum == read_tot && write_cum == write_tot)
1765                         break;
1766         }
1767 }
1768
1769 static int ll_rw_extents_stats_pp_seq_show(struct seq_file *seq, void *v)
1770 {
1771         struct timespec64 now;
1772         struct ll_sb_info *sbi = seq->private;
1773         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1774         int k;
1775
1776         ktime_get_real_ts64(&now);
1777
1778         if (!sbi->ll_rw_stats_on) {
1779                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1780                 return 0;
1781         }
1782         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1783                    (s64)now.tv_sec, now.tv_nsec);
1784         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1785         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1786                    "extents", "calls", "%", "cum%",
1787                    "calls", "%", "cum%");
1788         spin_lock(&sbi->ll_pp_extent_lock);
1789         for (k = 0; k < LL_PROCESS_HIST_MAX; k++) {
1790                 if (io_extents->pp_extents[k].pid != 0) {
1791                         seq_printf(seq, "\nPID: %d\n",
1792                                    io_extents->pp_extents[k].pid);
1793                         ll_display_extents_info(io_extents, seq, k);
1794                 }
1795         }
1796         spin_unlock(&sbi->ll_pp_extent_lock);
1797         return 0;
1798 }
1799
1800 static ssize_t ll_rw_extents_stats_pp_seq_write(struct file *file,
1801                                                 const char __user *buf,
1802                                                 size_t len,
1803                                                 loff_t *off)
1804 {
1805         struct seq_file *seq = file->private_data;
1806         struct ll_sb_info *sbi = seq->private;
1807         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1808         int i;
1809         __s64 value;
1810
1811         if (len == 0)
1812                 return -EINVAL;
1813
1814         value = ll_stats_pid_write(buf, len);
1815
1816         if (value == 0)
1817                 sbi->ll_rw_stats_on = 0;
1818         else
1819                 sbi->ll_rw_stats_on = 1;
1820
1821         spin_lock(&sbi->ll_pp_extent_lock);
1822         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1823                 io_extents->pp_extents[i].pid = 0;
1824                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1825                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1826         }
1827         spin_unlock(&sbi->ll_pp_extent_lock);
1828         return len;
1829 }
1830
1831 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats_pp);
1832
1833 static int ll_rw_extents_stats_seq_show(struct seq_file *seq, void *v)
1834 {
1835         struct timespec64 now;
1836         struct ll_sb_info *sbi = seq->private;
1837         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1838
1839         ktime_get_real_ts64(&now);
1840
1841         if (!sbi->ll_rw_stats_on) {
1842                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
1843                 return 0;
1844         }
1845         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
1846                    (s64)now.tv_sec, now.tv_nsec);
1847
1848         seq_printf(seq, "%15s %19s       | %20s\n", " ", "read", "write");
1849         seq_printf(seq, "%13s   %14s %4s %4s  | %14s %4s %4s\n",
1850                    "extents", "calls", "%", "cum%",
1851                    "calls", "%", "cum%");
1852         spin_lock(&sbi->ll_lock);
1853         ll_display_extents_info(io_extents, seq, LL_PROCESS_HIST_MAX);
1854         spin_unlock(&sbi->ll_lock);
1855
1856         return 0;
1857 }
1858
1859 static ssize_t ll_rw_extents_stats_seq_write(struct file *file,
1860                                              const char __user *buf,
1861                                              size_t len, loff_t *off)
1862 {
1863         struct seq_file *seq = file->private_data;
1864         struct ll_sb_info *sbi = seq->private;
1865         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1866         int i;
1867         __s64 value;
1868
1869         if (len == 0)
1870                 return -EINVAL;
1871
1872         value = ll_stats_pid_write(buf, len);
1873
1874         if (value == 0)
1875                 sbi->ll_rw_stats_on = 0;
1876         else
1877                 sbi->ll_rw_stats_on = 1;
1878
1879         spin_lock(&sbi->ll_pp_extent_lock);
1880         for (i = 0; i <= LL_PROCESS_HIST_MAX; i++) {
1881                 io_extents->pp_extents[i].pid = 0;
1882                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_r_hist);
1883                 lprocfs_oh_clear(&io_extents->pp_extents[i].pp_w_hist);
1884         }
1885         spin_unlock(&sbi->ll_pp_extent_lock);
1886
1887         return len;
1888 }
1889
1890 LDEBUGFS_SEQ_FOPS(ll_rw_extents_stats);
1891
1892 void ll_rw_stats_tally(struct ll_sb_info *sbi, pid_t pid,
1893                        struct ll_file_data *file, loff_t pos,
1894                        size_t count, int rw)
1895 {
1896         int i, cur = -1;
1897         struct ll_rw_process_info *process;
1898         struct ll_rw_process_info *offset;
1899         int *off_count = &sbi->ll_rw_offset_entry_count;
1900         int *process_count = &sbi->ll_offset_process_count;
1901         struct ll_rw_extents_info *io_extents = &sbi->ll_rw_extents_info;
1902
1903         if(!sbi->ll_rw_stats_on)
1904                 return;
1905         process = sbi->ll_rw_process_info;
1906         offset = sbi->ll_rw_offset_info;
1907
1908         spin_lock(&sbi->ll_pp_extent_lock);
1909         /* Extent statistics */
1910         for(i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1911                 if(io_extents->pp_extents[i].pid == pid) {
1912                         cur = i;
1913                         break;
1914                 }
1915         }
1916
1917         if (cur == -1) {
1918                 /* new process */
1919                 sbi->ll_extent_process_count =
1920                         (sbi->ll_extent_process_count + 1) % LL_PROCESS_HIST_MAX;
1921                 cur = sbi->ll_extent_process_count;
1922                 io_extents->pp_extents[cur].pid = pid;
1923                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_r_hist);
1924                 lprocfs_oh_clear(&io_extents->pp_extents[cur].pp_w_hist);
1925         }
1926
1927         for(i = 0; (count >= BIT(LL_HIST_START << i)) &&
1928              (i < (LL_HIST_MAX - 1)); i++);
1929         if (rw == 0) {
1930                 io_extents->pp_extents[cur].pp_r_hist.oh_buckets[i]++;
1931                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_r_hist.oh_buckets[i]++;
1932         } else {
1933                 io_extents->pp_extents[cur].pp_w_hist.oh_buckets[i]++;
1934                 io_extents->pp_extents[LL_PROCESS_HIST_MAX].pp_w_hist.oh_buckets[i]++;
1935         }
1936         spin_unlock(&sbi->ll_pp_extent_lock);
1937
1938         spin_lock(&sbi->ll_process_lock);
1939         /* Offset statistics */
1940         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
1941                 if (process[i].rw_pid == pid) {
1942                         if (process[i].rw_last_file != file) {
1943                                 process[i].rw_range_start = pos;
1944                                 process[i].rw_last_file_pos = pos + count;
1945                                 process[i].rw_smallest_extent = count;
1946                                 process[i].rw_largest_extent = count;
1947                                 process[i].rw_offset = 0;
1948                                 process[i].rw_last_file = file;
1949                                 spin_unlock(&sbi->ll_process_lock);
1950                                 return;
1951                         }
1952                         if (process[i].rw_last_file_pos != pos) {
1953                                 *off_count =
1954                                     (*off_count + 1) % LL_OFFSET_HIST_MAX;
1955                                 offset[*off_count].rw_op = process[i].rw_op;
1956                                 offset[*off_count].rw_pid = pid;
1957                                 offset[*off_count].rw_range_start =
1958                                         process[i].rw_range_start;
1959                                 offset[*off_count].rw_range_end =
1960                                         process[i].rw_last_file_pos;
1961                                 offset[*off_count].rw_smallest_extent =
1962                                         process[i].rw_smallest_extent;
1963                                 offset[*off_count].rw_largest_extent =
1964                                         process[i].rw_largest_extent;
1965                                 offset[*off_count].rw_offset =
1966                                         process[i].rw_offset;
1967                                 process[i].rw_op = rw;
1968                                 process[i].rw_range_start = pos;
1969                                 process[i].rw_smallest_extent = count;
1970                                 process[i].rw_largest_extent = count;
1971                                 process[i].rw_offset = pos -
1972                                         process[i].rw_last_file_pos;
1973                         }
1974                         if(process[i].rw_smallest_extent > count)
1975                                 process[i].rw_smallest_extent = count;
1976                         if(process[i].rw_largest_extent < count)
1977                                 process[i].rw_largest_extent = count;
1978                         process[i].rw_last_file_pos = pos + count;
1979                         spin_unlock(&sbi->ll_process_lock);
1980                         return;
1981                 }
1982         }
1983         *process_count = (*process_count + 1) % LL_PROCESS_HIST_MAX;
1984         process[*process_count].rw_pid = pid;
1985         process[*process_count].rw_op = rw;
1986         process[*process_count].rw_range_start = pos;
1987         process[*process_count].rw_last_file_pos = pos + count;
1988         process[*process_count].rw_smallest_extent = count;
1989         process[*process_count].rw_largest_extent = count;
1990         process[*process_count].rw_offset = 0;
1991         process[*process_count].rw_last_file = file;
1992         spin_unlock(&sbi->ll_process_lock);
1993 }
1994
1995 static int ll_rw_offset_stats_seq_show(struct seq_file *seq, void *v)
1996 {
1997         struct timespec64 now;
1998         struct ll_sb_info *sbi = seq->private;
1999         struct ll_rw_process_info *offset = sbi->ll_rw_offset_info;
2000         struct ll_rw_process_info *process = sbi->ll_rw_process_info;
2001         int i;
2002
2003         ktime_get_real_ts64(&now);
2004
2005         if (!sbi->ll_rw_stats_on) {
2006                 seq_puts(seq, "disabled\n write anything to this file to activate, then '0' or 'disable' to deactivate\n");
2007                 return 0;
2008         }
2009         spin_lock(&sbi->ll_process_lock);
2010
2011         seq_printf(seq, "snapshot_time:         %llu.%09lu (secs.nsecs)\n",
2012                    (s64)now.tv_sec, now.tv_nsec);
2013         seq_printf(seq, "%3s %10s %14s %14s %17s %17s %14s\n",
2014                    "R/W", "PID", "RANGE START", "RANGE END",
2015                    "SMALLEST EXTENT", "LARGEST EXTENT", "OFFSET");
2016
2017         /* We stored the discontiguous offsets here; print them first */
2018         for (i = 0; i < LL_OFFSET_HIST_MAX; i++) {
2019                 if (offset[i].rw_pid != 0)
2020                         seq_printf(seq,
2021                                   "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
2022                                    offset[i].rw_op == READ ? 'R' : 'W',
2023                                    offset[i].rw_pid,
2024                                    offset[i].rw_range_start,
2025                                    offset[i].rw_range_end,
2026                                    (unsigned long)offset[i].rw_smallest_extent,
2027                                    (unsigned long)offset[i].rw_largest_extent,
2028                                    offset[i].rw_offset);
2029         }
2030
2031         /* Then print the current offsets for each process */
2032         for (i = 0; i < LL_PROCESS_HIST_MAX; i++) {
2033                 if (process[i].rw_pid != 0)
2034                         seq_printf(seq,
2035                                   "%3c %10d %14llu %14llu %17lu %17lu %14llu\n",
2036                                    process[i].rw_op == READ ? 'R' : 'W',
2037                                    process[i].rw_pid,
2038                                    process[i].rw_range_start,
2039                                    process[i].rw_last_file_pos,
2040                                    (unsigned long)process[i].rw_smallest_extent,
2041                                    (unsigned long)process[i].rw_largest_extent,
2042                                    process[i].rw_offset);
2043         }
2044         spin_unlock(&sbi->ll_process_lock);
2045
2046         return 0;
2047 }
2048
2049 static ssize_t ll_rw_offset_stats_seq_write(struct file *file,
2050                                             const char __user *buf,
2051                                             size_t len, loff_t *off)
2052 {
2053         struct seq_file *seq = file->private_data;
2054         struct ll_sb_info *sbi = seq->private;
2055         struct ll_rw_process_info *process_info = sbi->ll_rw_process_info;
2056         struct ll_rw_process_info *offset_info = sbi->ll_rw_offset_info;
2057         __s64 value;
2058
2059         if (len == 0)
2060                 return -EINVAL;
2061
2062         value = ll_stats_pid_write(buf, len);
2063
2064         if (value == 0)
2065                 sbi->ll_rw_stats_on = 0;
2066         else
2067                 sbi->ll_rw_stats_on = 1;
2068
2069         spin_lock(&sbi->ll_process_lock);
2070         sbi->ll_offset_process_count = 0;
2071         sbi->ll_rw_offset_entry_count = 0;
2072         memset(process_info, 0, sizeof(struct ll_rw_process_info) *
2073                LL_PROCESS_HIST_MAX);
2074         memset(offset_info, 0, sizeof(struct ll_rw_process_info) *
2075                LL_OFFSET_HIST_MAX);
2076         spin_unlock(&sbi->ll_process_lock);
2077
2078         return len;
2079 }
2080
2081 LDEBUGFS_SEQ_FOPS(ll_rw_offset_stats);