lustre/obdclass/lprocfs_status.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) 2011, 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  * lustre/obdclass/lprocfs_status.c
  33  *
  34  * Author: Hariharan Thantry <thantry@users.sourceforge.net>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_CLASS
  38
  39 #include <obd_class.h>
  40 #include <lprocfs_status.h>
  41
  42 #ifdef CONFIG_PROC_FS
  43
  44 static int lprocfs_no_percpu_stats = 0;
  45 module_param(lprocfs_no_percpu_stats, int, 0644);
  46 MODULE_PARM_DESC(lprocfs_no_percpu_stats, "Do not alloc percpu data for lprocfs stats");
  47
  48 #define MAX_STRING_SIZE 128
  49
  50 int lprocfs_single_release(struct inode *inode, struct file *file)
  51 {
  52         return single_release(inode, file);
  53 }
  54 EXPORT_SYMBOL(lprocfs_single_release);
  55
  56 int lprocfs_seq_release(struct inode *inode, struct file *file)
  57 {
  58         return seq_release(inode, file);
  59 }
  60 EXPORT_SYMBOL(lprocfs_seq_release);
  61
  62 struct dentry *ldebugfs_add_simple(struct dentry *root,
  63                                    char *name, void *data,
  64                                    const struct file_operations *fops)
  65 {
  66         struct dentry *entry;
  67         umode_t mode = 0;
  68
  69         if (!root || !name || !fops)
  70                 return ERR_PTR(-EINVAL);
  71
  72         if (fops->read)
  73                 mode = 0444;
  74         if (fops->write)
  75                 mode |= 0200;
  76         entry = debugfs_create_file(name, mode, root, data, fops);
  77         if (IS_ERR_OR_NULL(entry)) {
  78                 CERROR("LprocFS: No memory to create <debugfs> entry %s", name);
  79                 return entry ?: ERR_PTR(-ENOMEM);
  80         }
  81         return entry;
  82 }
  83 EXPORT_SYMBOL(ldebugfs_add_simple);
  84
  85 struct proc_dir_entry *
  86 lprocfs_add_simple(struct proc_dir_entry *root, char *name,
  87                    void *data, const struct file_operations *fops)
  88 {
  89         struct proc_dir_entry *proc;
  90         mode_t mode = 0;
  91
  92         if (!root || !name || !fops)
  93                 return ERR_PTR(-EINVAL);
  94
  95         if (fops->read)
  96                 mode = 0444;
  97         if (fops->write)
  98                 mode |= 0200;
  99         proc = proc_create_data(name, mode, root, fops, data);
 100         if (!proc) {
 101                 CERROR("LprocFS: No memory to create /proc entry %s\n",
 102                        name);
 103                 return ERR_PTR(-ENOMEM);
 104         }
 105         return proc;
 106 }
 107 EXPORT_SYMBOL(lprocfs_add_simple);
 108
 109 struct proc_dir_entry *lprocfs_add_symlink(const char *name,
 110                                            struct proc_dir_entry *parent,
 111                                            const char *format, ...)
 112 {
 113         struct proc_dir_entry *entry;
 114         char *dest;
 115         va_list ap;
 116
 117         if (!parent || !format)
 118                 return NULL;
 119
 120         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
 121         if (!dest)
 122                 return NULL;
 123
 124         va_start(ap, format);
 125         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
 126         va_end(ap);
 127
 128         entry = proc_symlink(name, parent, dest);
 129         if (!entry)
 130                 CERROR("LprocFS: Could not create symbolic link from "
 131                        "%s to %s\n", name, dest);
 132
 133         OBD_FREE(dest, MAX_STRING_SIZE + 1);
 134         return entry;
 135 }
 136 EXPORT_SYMBOL(lprocfs_add_symlink);
 137
 138 static const struct file_operations lprocfs_generic_fops = { };
 139
 140 void ldebugfs_add_vars(struct dentry *parent, struct lprocfs_vars *list,
 141                        void *data)
 142 {
 143         if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
 144                 return;
 145
 146         while (list->name) {
 147                 umode_t mode = 0;
 148
 149                 if (list->proc_mode != 0000) {
 150                         mode = list->proc_mode;
 151                 } else if (list->fops) {
 152                         if (list->fops->read)
 153                                 mode = 0444;
 154                         if (list->fops->write)
 155                                 mode |= 0200;
 156                 }
 157                 debugfs_create_file(list->name, mode, parent,
 158                                     list->data ? : data,
 159                                     list->fops ? : &lprocfs_generic_fops);
 160                 list++;
 161         }
 162 }
 163 EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
 164
 165 /**
 166  * Add /proc entries.
 167  *
 168  * \param root [in]  The parent proc entry on which new entry will be added.
 169  * \param list [in]  Array of proc entries to be added.
 170  * \param data [in]  The argument to be passed when entries read/write routines
 171  *                   are called through /proc file.
 172  *
 173  * \retval 0   on success
 174  *         < 0 on error
 175  */
 176 int
 177 lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
 178                  void *data)
 179 {
 180         if (!root || !list)
 181                 return -EINVAL;
 182
 183         while (list->name) {
 184                 struct proc_dir_entry *proc;
 185                 mode_t mode = 0;
 186
 187                 if (list->proc_mode != 0000) {
 188                         mode = list->proc_mode;
 189                 } else if (list->fops) {
 190                         if (list->fops->read)
 191                                 mode = 0444;
 192                         if (list->fops->write)
 193                                 mode |= 0200;
 194                 }
 195                 proc = proc_create_data(list->name, mode, root,
 196                                         list->fops ?: &lprocfs_generic_fops,
 197                                         list->data ?: data);
 198                 if (!proc)
 199                         return -ENOMEM;
 200                 list++;
 201         }
 202         return 0;
 203 }
 204 EXPORT_SYMBOL(lprocfs_add_vars);
 205
 206 void lprocfs_remove(struct proc_dir_entry **rooth)
 207 {
 208         proc_remove(*rooth);
 209         *rooth = NULL;
 210 }
 211 EXPORT_SYMBOL(lprocfs_remove);
 212
 213 void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
 214 {
 215         LASSERT(parent != NULL);
 216         remove_proc_entry(name, parent);
 217 }
 218 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
 219
 220 struct proc_dir_entry *
 221 lprocfs_register(const char *name, struct proc_dir_entry *parent,
 222                  struct lprocfs_vars *list, void *data)
 223 {
 224         struct proc_dir_entry *newchild;
 225
 226         newchild = proc_mkdir(name, parent);
 227         if (!newchild)
 228                 return ERR_PTR(-ENOMEM);
 229
 230         if (list) {
 231                 int rc = lprocfs_add_vars(newchild, list, data);
 232                 if (rc) {
 233                         lprocfs_remove(&newchild);
 234                         return ERR_PTR(rc);
 235                 }
 236         }
 237         return newchild;
 238 }
 239 EXPORT_SYMBOL(lprocfs_register);
 240
 241 /* Generic callbacks */
 242 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
 243 {
 244         struct obd_device *obd = data;
 245
 246         LASSERT(obd != NULL);
 247         seq_printf(m, "%s\n", obd->obd_uuid.uuid);
 248         return 0;
 249 }
 250 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
 251
 252 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 253                          char *buf)
 254 {
 255         struct obd_device *obd = container_of(kobj, struct obd_device,
 256                                               obd_kset.kobj);
 257
 258         return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
 259 }
 260 LUSTRE_RO_ATTR(uuid);
 261
 262 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 263                               char *buf)
 264 {
 265         struct obd_device *obd = container_of(kobj, struct obd_device,
 266                                               obd_kset.kobj);
 267         struct obd_statfs osfs;
 268         int rc;
 269
 270         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 271                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 272                         OBD_STATFS_NODELAY);
 273         if (!rc)
 274                 return sprintf(buf, "%u\n", osfs.os_bsize);
 275
 276         return rc;
 277 }
 278 LUSTRE_RO_ATTR(blocksize);
 279
 280 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 281                                 char *buf)
 282 {
 283         struct obd_device *obd = container_of(kobj, struct obd_device,
 284                                               obd_kset.kobj);
 285         struct obd_statfs osfs;
 286         int rc;
 287
 288         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 289                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 290                         OBD_STATFS_NODELAY);
 291         if (!rc) {
 292                 u32 blk_size = osfs.os_bsize >> 10;
 293                 u64 result = osfs.os_blocks;
 294
 295                 result *= rounddown_pow_of_two(blk_size ?: 1);
 296                 return sprintf(buf, "%llu\n", result);
 297         }
 298
 299         return rc;
 300 }
 301 LUSTRE_RO_ATTR(kbytestotal);
 302
 303 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 304                                char *buf)
 305 {
 306         struct obd_device *obd = container_of(kobj, struct obd_device,
 307                                               obd_kset.kobj);
 308         struct obd_statfs osfs;
 309         int rc;
 310
 311         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 312                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 313                         OBD_STATFS_NODELAY);
 314         if (!rc) {
 315                 u32 blk_size = osfs.os_bsize >> 10;
 316                 u64 result = osfs.os_bfree;
 317
 318                 while (blk_size >>= 1)
 319                         result <<= 1;
 320
 321                 return sprintf(buf, "%llu\n", result);
 322         }
 323
 324         return rc;
 325 }
 326 LUSTRE_RO_ATTR(kbytesfree);
 327
 328 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 329                                 char *buf)
 330 {
 331         struct obd_device *obd = container_of(kobj, struct obd_device,
 332                                               obd_kset.kobj);
 333         struct obd_statfs osfs;
 334         int rc;
 335
 336         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 337                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 338                         OBD_STATFS_NODELAY);
 339         if (!rc) {
 340                 u32 blk_size = osfs.os_bsize >> 10;
 341                 u64 result = osfs.os_bavail;
 342
 343                 while (blk_size >>= 1)
 344                         result <<= 1;
 345
 346                 return sprintf(buf, "%llu\n", result);
 347         }
 348
 349         return rc;
 350 }
 351 LUSTRE_RO_ATTR(kbytesavail);
 352
 353 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 354                                char *buf)
 355 {
 356         struct obd_device *obd = container_of(kobj, struct obd_device,
 357                                               obd_kset.kobj);
 358         struct obd_statfs osfs;
 359         int rc;
 360
 361         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 362                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 363                         OBD_STATFS_NODELAY);
 364         if (!rc)
 365                 return sprintf(buf, "%llu\n", osfs.os_files);
 366
 367         return rc;
 368 }
 369 LUSTRE_RO_ATTR(filestotal);
 370
 371 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 372                               char *buf)
 373 {
 374         struct obd_device *obd = container_of(kobj, struct obd_device,
 375                                               obd_kset.kobj);
 376         struct obd_statfs osfs;
 377         int rc;
 378
 379         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 380                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 381                         OBD_STATFS_NODELAY);
 382         if (!rc)
 383                 return sprintf(buf, "%llu\n", osfs.os_ffree);
 384
 385         return rc;
 386 }
 387 LUSTRE_RO_ATTR(filesfree);
 388
 389 ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
 390 {
 391         struct obd_device *obd = container_of(kobj, struct obd_device,
 392                                               obd_kset.kobj);
 393         struct obd_import *imp;
 394         struct ptlrpc_connection *conn;
 395         ssize_t count;
 396
 397         with_imp_locked(obd, imp, count) {
 398                 conn = imp->imp_connection;
 399                 if (conn)
 400                         count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
 401                 else
 402                         count = sprintf(buf, "%s\n", "<none>");
 403         }
 404
 405         return count;
 406 }
 407 EXPORT_SYMBOL(conn_uuid_show);
 408
 409 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
 410 {
 411         struct obd_device *obd = data;
 412         struct obd_import *imp;
 413         char *imp_state_name = NULL;
 414         int rc = 0;
 415
 416         LASSERT(obd != NULL);
 417         with_imp_locked(obd, imp, rc) {
 418                 imp_state_name = ptlrpc_import_state_name(imp->imp_state);
 419                 seq_printf(m, "%s\t%s%s\n", obd2cli_tgt(obd), imp_state_name,
 420                            imp->imp_deactive ? "\tDEACTIVATED" : "");
 421         }
 422
 423         return rc;
 424 }
 425 EXPORT_SYMBOL(lprocfs_server_uuid_seq_show);
 426
 427 /** add up per-cpu counters */
 428
 429 /**
 430  * Lock statistics structure for access, possibly only on this CPU.
 431  *
 432  * The statistics struct may be allocated with per-CPU structures for
 433  * efficient concurrent update (usually only on server-wide stats), or
 434  * as a single global struct (e.g. for per-client or per-job statistics),
 435  * so the required locking depends on the type of structure allocated.
 436  *
 437  * For per-CPU statistics, pin the thread to the current cpuid so that
 438  * will only access the statistics for that CPU.  If the stats structure
 439  * for the current CPU has not been allocated (or previously freed),
 440  * allocate it now.  The per-CPU statistics do not need locking since
 441  * the thread is pinned to the CPU during update.
 442  *
 443  * For global statistics, lock the stats structure to prevent concurrent update.
 444  *
 445  * \param[in] stats     statistics structure to lock
 446  * \param[in] opc       type of operation:
 447  *                      LPROCFS_GET_SMP_ID: "lock" and return current CPU index
 448  *                              for incrementing statistics for that CPU
 449  *                      LPROCFS_GET_NUM_CPU: "lock" and return number of used
 450  *                              CPU indices to iterate over all indices
 451  * \param[out] flags    CPU interrupt saved state for IRQ-safe locking
 452  *
 453  * \retval cpuid of current thread or number of allocated structs
 454  * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
 455  */
 456 int lprocfs_stats_lock(struct lprocfs_stats *stats,
 457                        enum lprocfs_stats_lock_ops opc,
 458                        unsigned long *flags)
 459 {
 460         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 461                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 462                         spin_lock_irqsave(&stats->ls_lock, *flags);
 463                 else
 464                         spin_lock(&stats->ls_lock);
 465                 return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
 466         }
 467
 468         switch (opc) {
 469         case LPROCFS_GET_SMP_ID: {
 470                 unsigned int cpuid = get_cpu();
 471
 472                 if (unlikely(!stats->ls_percpu[cpuid])) {
 473                         int rc = lprocfs_stats_alloc_one(stats, cpuid);
 474
 475                         if (rc < 0) {
 476                                 put_cpu();
 477                                 return rc;
 478                         }
 479                 }
 480                 return cpuid;
 481         }
 482         case LPROCFS_GET_NUM_CPU:
 483                 return stats->ls_biggest_alloc_num;
 484         default:
 485                 LBUG();
 486         }
 487 }
 488
 489 /**
 490  * Unlock statistics structure after access.
 491  *
 492  * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
 493  * or unpin this thread from the current cpuid for per-CPU statistics.
 494  *
 495  * This function must be called using the same arguments as used when calling
 496  * lprocfs_stats_lock() so that the correct operation can be performed.
 497  *
 498  * \param[in] stats     statistics structure to unlock
 499  * \param[in] opc       type of operation (current cpuid or number of structs)
 500  * \param[in] flags     CPU interrupt saved state for IRQ-safe locking
 501  */
 502 void lprocfs_stats_unlock(struct lprocfs_stats *stats,
 503                           enum lprocfs_stats_lock_ops opc,
 504                           unsigned long *flags)
 505 {
 506         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 507                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 508                         spin_unlock_irqrestore(&stats->ls_lock, *flags);
 509                 else
 510                         spin_unlock(&stats->ls_lock);
 511         } else if (opc == LPROCFS_GET_SMP_ID) {
 512                 put_cpu();
 513         }
 514 }
 515
 516 /** add up per-cpu counters */
 517 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
 518                            struct lprocfs_counter *cnt)
 519 {
 520         unsigned int num_entry;
 521         struct lprocfs_counter *percpu_cntr;
 522         int i;
 523         unsigned long flags = 0;
 524
 525         memset(cnt, 0, sizeof(*cnt));
 526
 527         if (!stats) {
 528                 /* set count to 1 to avoid divide-by-zero errs in callers */
 529                 cnt->lc_count = 1;
 530                 return;
 531         }
 532
 533         cnt->lc_min = LC_MIN_INIT;
 534
 535         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
 536
 537         for (i = 0; i < num_entry; i++) {
 538                 if (!stats->ls_percpu[i])
 539                         continue;
 540                 percpu_cntr = lprocfs_stats_counter_get(stats, i, idx);
 541
 542                 cnt->lc_count += percpu_cntr->lc_count;
 543                 cnt->lc_sum += percpu_cntr->lc_sum;
 544                 if (percpu_cntr->lc_min < cnt->lc_min)
 545                         cnt->lc_min = percpu_cntr->lc_min;
 546                 if (percpu_cntr->lc_max > cnt->lc_max)
 547                         cnt->lc_max = percpu_cntr->lc_max;
 548                 cnt->lc_sumsquare += percpu_cntr->lc_sumsquare;
 549         }
 550
 551         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
 552 }
 553
 554 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
 555 {
 556         bool first = true;
 557
 558         if (imp->imp_obd->obd_no_recov) {
 559                 seq_printf(m, "no_recov");
 560                 first = false;
 561         }
 562
 563         flag2str(imp, invalid);
 564         flag2str(imp, deactive);
 565         flag2str(imp, replayable);
 566         flag2str(imp, delayed_recovery);
 567         flag2str(imp, vbr_failed);
 568         flag2str(imp, pingable);
 569         flag2str(imp, resend_replay);
 570         flag2str(imp, no_pinger_recover);
 571         flag2str(imp, connect_tried);
 572 }
 573
 574 static const char *obd_connect_names[] = {
 575         /* flags names  */
 576         "read_only",
 577         "lov_index",
 578         "connect_from_mds",
 579         "write_grant",
 580         "server_lock",
 581         "version",
 582         "request_portal",
 583         "acl",
 584         "xattr",
 585         "create_on_write",
 586         "truncate_lock",
 587         "initial_transno",
 588         "inode_bit_locks",
 589         "barrier",
 590         "getattr_by_fid",
 591         "no_oh_for_devices",
 592         "remote_client",
 593         "remote_client_by_force",
 594         "max_byte_per_rpc",
 595         "64bit_qdata",
 596         "mds_capability",
 597         "oss_capability",
 598         "early_lock_cancel",
 599         "som",
 600         "adaptive_timeouts",
 601         "lru_resize",
 602         "mds_mds_connection",
 603         "real_conn",
 604         "change_qunit_size",
 605         "alt_checksum_algorithm",
 606         "fid_is_enabled",
 607         "version_recovery",
 608         "pools",
 609         "grant_shrink",
 610         "skip_orphan",
 611         "large_ea",
 612         "full20",
 613         "layout_lock",
 614         "64bithash",
 615         "object_max_bytes",
 616         "imp_recov",
 617         "jobstats",
 618         "umask",
 619         "einprogress",
 620         "grant_param",
 621         "flock_owner",
 622         "lvb_type",
 623         "nanoseconds_times",
 624         "lightweight_conn",
 625         "short_io",
 626         "pingless",
 627         "flock_deadlock",
 628         "disp_stripe",
 629         "open_by_fid",
 630         "lfsck",
 631         "unknown",
 632         "unlink_close",
 633         "multi_mod_rpcs",
 634         "dir_stripe",
 635         "subtree",
 636         "lockahead",
 637         "bulk_mbits",
 638         "compact_obdo",
 639         "second_flags",
 640         /* flags2 names */
 641         "file_secctx",  /* 0x01 */
 642         "lockaheadv2",  /* 0x02 */
 643         "dir_migrate",  /* 0x04 */
 644         "sum_statfs",   /* 0x08 */
 645         "overstriping", /* 0x10 */
 646         "flr",          /* 0x20 */
 647         "wbc",          /* 0x40 */
 648         "lock_convert",  /* 0x80 */
 649         "archive_id_array",     /* 0x100 */
 650         "increasing_xid",       /* 0x200 */
 651         "selinux_policy",       /* 0x400 */
 652         "lsom",                 /* 0x800 */
 653         "pcc",                  /* 0x1000 */
 654         "crush",                /* 0x2000 */
 655         "async_discard",        /* 0x4000 */
 656         "client_encryption",    /* 0x8000 */
 657         NULL
 658 };
 659
 660 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
 661                                const char *sep)
 662 {
 663         bool first = true;
 664         __u64 mask;
 665         int i;
 666
 667         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 668                 if (flags & mask) {
 669                         seq_printf(m, "%s%s",
 670                                    first ? "" : sep, obd_connect_names[i]);
 671                         first = false;
 672                 }
 673         }
 674
 675         if (flags & ~(mask - 1)) {
 676                 seq_printf(m, "%sunknown_%#llx",
 677                            first ? "" : sep, flags & ~(mask - 1));
 678                 first = false;
 679         }
 680
 681         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 682                 return;
 683
 684         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 685                 if (flags2 & mask) {
 686                         seq_printf(m, "%s%s",
 687                                    first ? "" : sep, obd_connect_names[i]);
 688                         first = false;
 689                 }
 690         }
 691
 692         if (flags2 & ~(mask - 1)) {
 693                 seq_printf(m, "%sunknown2_%#llx",
 694                            first ? "" : sep, flags2 & ~(mask - 1));
 695                 first = false;
 696         }
 697 }
 698 EXPORT_SYMBOL(obd_connect_seq_flags2str);
 699
 700 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 701                           const char *sep)
 702 {
 703         __u64 mask;
 704         int i, ret = 0;
 705
 706         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 707                 if (flags & mask)
 708                         ret += snprintf(page + ret, count - ret, "%s%s",
 709                                         ret ? sep : "", obd_connect_names[i]);
 710         }
 711
 712         if (flags & ~(mask - 1))
 713                 ret += snprintf(page + ret, count - ret,
 714                                 "%sunknown_%#llx",
 715                                 ret ? sep : "", flags & ~(mask - 1));
 716
 717         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 718                 return ret;
 719
 720         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 721                 if (flags2 & mask)
 722                         ret += snprintf(page + ret, count - ret, "%s%s",
 723                                         ret ? sep : "", obd_connect_names[i]);
 724         }
 725
 726         if (flags2 & ~(mask - 1))
 727                 ret += snprintf(page + ret, count - ret,
 728                                 "%sunknown2_%#llx",
 729                                 ret ? sep : "", flags2 & ~(mask - 1));
 730
 731         return ret;
 732 }
 733 EXPORT_SYMBOL(obd_connect_flags2str);
 734
 735 void
 736 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 737 {
 738         __u64 flags;
 739
 740         LASSERT(ocd != NULL);
 741         flags = ocd->ocd_connect_flags;
 742
 743         seq_printf(m, "    connect_data:\n"
 744                    "       flags: %#llx\n"
 745                    "       instance: %u\n",
 746                    ocd->ocd_connect_flags,
 747                    ocd->ocd_instance);
 748         if (flags & OBD_CONNECT_VERSION)
 749                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
 750                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
 751                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
 752                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
 753                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
 754         if (flags & OBD_CONNECT_MDS)
 755                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
 756         if (flags & OBD_CONNECT_GRANT)
 757                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
 758         if (flags & OBD_CONNECT_INDEX)
 759                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
 760         if (flags & OBD_CONNECT_BRW_SIZE)
 761                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
 762         if (flags & OBD_CONNECT_IBITS)
 763                 seq_printf(m, "       ibits_known: %#llx\n",
 764                            ocd->ocd_ibits_known);
 765         if (flags & OBD_CONNECT_GRANT_PARAM)
 766                 seq_printf(m, "       grant_block_size: %d\n"
 767                            "       grant_inode_size: %d\n"
 768                            "       grant_max_extent_size: %d\n"
 769                            "       grant_extent_tax: %d\n",
 770                            1 << ocd->ocd_grant_blkbits,
 771                            1 << ocd->ocd_grant_inobits,
 772                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
 773                            ocd->ocd_grant_tax_kb << 10);
 774         if (flags & OBD_CONNECT_TRANSNO)
 775                 seq_printf(m, "       first_transno: %#llx\n",
 776                            ocd->ocd_transno);
 777         if (flags & OBD_CONNECT_CKSUM)
 778                 seq_printf(m, "       cksum_types: %#x\n",
 779                            ocd->ocd_cksum_types);
 780         if (flags & OBD_CONNECT_MAX_EASIZE)
 781                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
 782         if (flags & OBD_CONNECT_MAXBYTES)
 783                 seq_printf(m, "       max_object_bytes: %llu\n",
 784                            ocd->ocd_maxbytes);
 785         if (flags & OBD_CONNECT_MULTIMODRPCS)
 786                 seq_printf(m, "       max_mod_rpcs: %hu\n",
 787                            ocd->ocd_maxmodrpcs);
 788 }
 789
 790 static void lprocfs_import_seq_show_locked(struct seq_file *m,
 791                                            struct obd_device *obd,
 792                                            struct obd_import *imp)
 793 {
 794         char nidstr[LNET_NIDSTR_SIZE];
 795         struct lprocfs_counter ret;
 796         struct lprocfs_counter_header *header;
 797         struct obd_import_conn *conn;
 798         struct obd_connect_data *ocd;
 799         int j;
 800         int k;
 801         int rw = 0;
 802
 803         ocd = &imp->imp_connect_data;
 804
 805         seq_printf(m, "import:\n"
 806                    "    name: %s\n"
 807                    "    target: %s\n"
 808                    "    state: %s\n"
 809                    "    connect_flags: [ ",
 810                    obd->obd_name,
 811                    obd2cli_tgt(obd),
 812                    ptlrpc_import_state_name(imp->imp_state));
 813         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
 814                                   imp->imp_connect_data.ocd_connect_flags2,
 815                                   ", ");
 816         seq_printf(m, " ]\n");
 817         obd_connect_data_seqprint(m, ocd);
 818         seq_printf(m, "    import_flags: [ ");
 819         obd_import_flags2str(imp, m);
 820
 821         seq_printf(m, " ]\n"
 822                    "    connection:\n"
 823                    "       failover_nids: [ ");
 824         spin_lock(&imp->imp_lock);
 825         j = 0;
 826         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 827                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
 828                                  nidstr, sizeof(nidstr));
 829                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
 830                 j++;
 831         }
 832         if (imp->imp_connection)
 833                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
 834                                  nidstr, sizeof(nidstr));
 835         else
 836                 strncpy(nidstr, "<none>", sizeof(nidstr));
 837         seq_printf(m, " ]\n"
 838                    "       current_connection: %s\n"
 839                    "       connection_attempts: %u\n"
 840                    "       generation: %u\n"
 841                    "       in-progress_invalidations: %u\n"
 842                    "       idle: %lld sec\n",
 843                    nidstr,
 844                    imp->imp_conn_cnt,
 845                    imp->imp_generation,
 846                    atomic_read(&imp->imp_inval_count),
 847                    ktime_get_real_seconds() - imp->imp_last_reply_time);
 848         spin_unlock(&imp->imp_lock);
 849
 850         if (!obd->obd_svc_stats)
 851                 return;
 852
 853         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
 854         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
 855         if (ret.lc_count != 0)
 856                 ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 857         else
 858                 ret.lc_sum = 0;
 859         seq_printf(m, "    rpcs:\n"
 860                    "       inflight: %u\n"
 861                    "       unregistering: %u\n"
 862                    "       timeouts: %u\n"
 863                    "       avg_waittime: %llu %s\n",
 864                    atomic_read(&imp->imp_inflight),
 865                    atomic_read(&imp->imp_unregistering),
 866                    atomic_read(&imp->imp_timeouts),
 867                    ret.lc_sum, header->lc_units);
 868
 869         k = 0;
 870         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
 871                 if (imp->imp_at.iat_portal[j] == 0)
 872                         break;
 873                 k = max_t(unsigned int, k,
 874                           at_get(&imp->imp_at.iat_service_estimate[j]));
 875         }
 876         seq_printf(m, "    service_estimates:\n"
 877                    "       services: %u sec\n"
 878                    "       network: %u sec\n",
 879                    k,
 880                    at_get(&imp->imp_at.iat_net_latency));
 881
 882         seq_printf(m, "    transactions:\n"
 883                    "       last_replay: %llu\n"
 884                    "       peer_committed: %llu\n"
 885                    "       last_checked: %llu\n",
 886                    imp->imp_last_replay_transno,
 887                    imp->imp_peer_committed_transno,
 888                    imp->imp_last_transno_checked);
 889
 890         /* avg data rates */
 891         for (rw = 0; rw <= 1; rw++) {
 892                 lprocfs_stats_collect(obd->obd_svc_stats,
 893                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
 894                                       &ret);
 895                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
 896                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 897                         seq_printf(m, "    %s_data_averages:\n"
 898                                    "       bytes_per_rpc: %llu\n",
 899                                    rw ? "write" : "read",
 900                                    ret.lc_sum);
 901                 }
 902                 k = (int)ret.lc_sum;
 903                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
 904                 header = &obd->obd_svc_stats->ls_cnt_header[j];
 905                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
 906                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
 907                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 908                         seq_printf(m, "       %s_per_rpc: %llu\n",
 909                                    header->lc_units, ret.lc_sum);
 910                         j = (int)ret.lc_sum;
 911                         if (j > 0)
 912                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
 913                                            k / j, (100 * k / j) % 100);
 914                 }
 915         }
 916 }
 917
 918 int lprocfs_import_seq_show(struct seq_file *m, void *data)
 919 {
 920         struct obd_device *obd = (struct obd_device *)data;
 921         struct obd_import *imp;
 922         int rv;
 923
 924         LASSERT(obd != NULL);
 925         with_imp_locked(obd, imp, rv)
 926                 lprocfs_import_seq_show_locked(m, obd, imp);
 927         return rv;
 928 }
 929 EXPORT_SYMBOL(lprocfs_import_seq_show);
 930
 931 int lprocfs_state_seq_show(struct seq_file *m, void *data)
 932 {
 933         struct obd_device *obd = (struct obd_device *)data;
 934         struct obd_import *imp;
 935         int j, k;
 936         int rc;
 937
 938         LASSERT(obd != NULL);
 939         with_imp_locked(obd, imp, rc) {
 940                 seq_printf(m, "current_state: %s\n",
 941                            ptlrpc_import_state_name(imp->imp_state));
 942                 seq_printf(m, "state_history:\n");
 943                 k = imp->imp_state_hist_idx;
 944                 for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
 945                         struct import_state_hist *ish =
 946                                 &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
 947                         if (ish->ish_state == 0)
 948                                 continue;
 949                         seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
 950                                    ptlrpc_import_state_name(ish->ish_state));
 951                 }
 952         }
 953
 954         return rc;
 955 }
 956 EXPORT_SYMBOL(lprocfs_state_seq_show);
 957
 958 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
 959 {
 960         int i;
 961         for (i = 0; i < AT_BINS; i++)
 962                 seq_printf(m, "%3u ", at->at_hist[i]);
 963         seq_printf(m, "\n");
 964         return 0;
 965 }
 966 EXPORT_SYMBOL(lprocfs_at_hist_helper);
 967
 968 /* See also ptlrpc_lprocfs_timeouts_show_seq */
 969 static void lprocfs_timeouts_seq_show_locked(struct seq_file *m,
 970                                              struct obd_device *obd,
 971                                              struct obd_import *imp)
 972 {
 973         unsigned int cur, worst;
 974         time64_t now, worstt;
 975         int i;
 976
 977         LASSERT(obd != NULL);
 978
 979         now = ktime_get_real_seconds();
 980
 981         /* Some network health info for kicks */
 982         seq_printf(m, "%-10s : %lld, %llds ago\n",
 983                    "last reply", (s64)imp->imp_last_reply_time,
 984                    (s64)(now - imp->imp_last_reply_time));
 985
 986         cur = at_get(&imp->imp_at.iat_net_latency);
 987         worst = imp->imp_at.iat_net_latency.at_worst_ever;
 988         worstt = imp->imp_at.iat_net_latency.at_worst_time;
 989         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
 990                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
 991         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
 992
 993         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
 994                 if (imp->imp_at.iat_portal[i] == 0)
 995                         break;
 996                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
 997                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
 998                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
 999                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1000                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1001                            (s64)(now - worstt));
1002                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1003         }
1004 }
1005
1006 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1007 {
1008         struct obd_device *obd = (struct obd_device *)data;
1009         struct obd_import *imp;
1010         int rc;
1011
1012         with_imp_locked(obd, imp, rc)
1013                 lprocfs_timeouts_seq_show_locked(m, obd, imp);
1014         return rc;
1015 }
1016 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1017
1018 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1019 {
1020         struct obd_device *obd = data;
1021         __u64 flags;
1022         __u64 flags2;
1023         struct obd_import *imp;
1024         int rc;
1025
1026         with_imp_locked(obd, imp, rc) {
1027                 flags = imp->imp_connect_data.ocd_connect_flags;
1028                 flags2 = imp->imp_connect_data.ocd_connect_flags2;
1029                 seq_printf(m, "flags=%#llx\n", flags);
1030                 seq_printf(m, "flags2=%#llx\n", flags2);
1031                 obd_connect_seq_flags2str(m, flags, flags2, "\n");
1032                 seq_printf(m, "\n");
1033         }
1034
1035         return rc;
1036 }
1037 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1038
1039 static const struct attribute *obd_def_uuid_attrs[] = {
1040         &lustre_attr_uuid.attr,
1041         NULL,
1042 };
1043
1044 static const struct attribute *obd_def_attrs[] = {
1045         &lustre_attr_blocksize.attr,
1046         &lustre_attr_kbytestotal.attr,
1047         &lustre_attr_kbytesfree.attr,
1048         &lustre_attr_kbytesavail.attr,
1049         &lustre_attr_filestotal.attr,
1050         &lustre_attr_filesfree.attr,
1051         &lustre_attr_uuid.attr,
1052         NULL,
1053 };
1054
1055 static void obd_sysfs_release(struct kobject *kobj)
1056 {
1057         struct obd_device *obd = container_of(kobj, struct obd_device,
1058                                               obd_kset.kobj);
1059
1060         complete(&obd->obd_kobj_unregister);
1061 }
1062
1063 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1064 {
1065         struct lprocfs_vars *debugfs_vars = NULL;
1066         int rc;
1067
1068         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1069                 return -ENODEV;
1070
1071         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1072         if (rc)
1073                 return rc;
1074
1075         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1076         obd->obd_ktype.release = obd_sysfs_release;
1077
1078         obd->obd_kset.kobj.parent = &obd->obd_type->typ_kobj;
1079         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1080         init_completion(&obd->obd_kobj_unregister);
1081         rc = kset_register(&obd->obd_kset);
1082         if (rc)
1083                 return rc;
1084
1085         if (uuid_only)
1086                 obd->obd_attrs = obd_def_uuid_attrs;
1087         else
1088                 obd->obd_attrs = obd_def_attrs;
1089
1090         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1091         if (rc) {
1092                 kset_unregister(&obd->obd_kset);
1093                 return rc;
1094         }
1095
1096         if (!obd->obd_type->typ_procroot)
1097                 debugfs_vars = obd->obd_vars;
1098         obd->obd_debugfs_entry = debugfs_create_dir(
1099                 obd->obd_name, obd->obd_type->typ_debugfs_entry);
1100         ldebugfs_add_vars(obd->obd_debugfs_entry, debugfs_vars, obd);
1101
1102         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1103                 GOTO(already_registered, rc);
1104
1105         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1106                                                obd->obd_type->typ_procroot,
1107                                                obd->obd_vars, obd);
1108         if (IS_ERR(obd->obd_proc_entry)) {
1109                 rc = PTR_ERR(obd->obd_proc_entry);
1110                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1111                 obd->obd_proc_entry = NULL;
1112
1113                 debugfs_remove_recursive(obd->obd_debugfs_entry);
1114                 obd->obd_debugfs_entry = NULL;
1115
1116                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1117                 obd->obd_attrs = NULL;
1118                 kset_unregister(&obd->obd_kset);
1119                 return rc;
1120         }
1121 already_registered:
1122         return rc;
1123 }
1124 EXPORT_SYMBOL(lprocfs_obd_setup);
1125
1126 int lprocfs_obd_cleanup(struct obd_device *obd)
1127 {
1128         if (!obd)
1129                 return -EINVAL;
1130
1131         if (obd->obd_proc_exports_entry) {
1132                 /* Should be no exports left */
1133                 lprocfs_remove(&obd->obd_proc_exports_entry);
1134                 obd->obd_proc_exports_entry = NULL;
1135         }
1136
1137         if (obd->obd_proc_entry) {
1138                 lprocfs_remove(&obd->obd_proc_entry);
1139                 obd->obd_proc_entry = NULL;
1140         }
1141
1142         debugfs_remove_recursive(obd->obd_debugfs_entry);
1143         obd->obd_debugfs_entry = NULL;
1144
1145         /* obd device never allocated a kset */
1146         if (!obd->obd_kset.kobj.state_initialized)
1147                 return 0;
1148
1149         if (obd->obd_attrs) {
1150                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1151                 obd->obd_attrs = NULL;
1152         }
1153
1154         kset_unregister(&obd->obd_kset);
1155         wait_for_completion(&obd->obd_kobj_unregister);
1156         return 0;
1157 }
1158 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1159
1160 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1161 {
1162         struct lprocfs_counter *cntr;
1163         unsigned int percpusize;
1164         int rc = -ENOMEM;
1165         unsigned long flags = 0;
1166         int i;
1167
1168         LASSERT(stats->ls_percpu[cpuid] == NULL);
1169         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1170
1171         percpusize = lprocfs_stats_counter_size(stats);
1172         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1173         if (stats->ls_percpu[cpuid]) {
1174                 rc = 0;
1175                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1176                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1177                                 spin_lock_irqsave(&stats->ls_lock, flags);
1178                         else
1179                                 spin_lock(&stats->ls_lock);
1180                         if (stats->ls_biggest_alloc_num <= cpuid)
1181                                 stats->ls_biggest_alloc_num = cpuid + 1;
1182                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1183                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1184                         } else {
1185                                 spin_unlock(&stats->ls_lock);
1186                         }
1187                 }
1188                 /* initialize the ls_percpu[cpuid] non-zero counter */
1189                 for (i = 0; i < stats->ls_num; ++i) {
1190                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1191                         cntr->lc_min = LC_MIN_INIT;
1192                 }
1193         }
1194         return rc;
1195 }
1196
1197 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1198                                           enum lprocfs_stats_flags flags)
1199 {
1200         struct lprocfs_stats *stats;
1201         unsigned int num_entry;
1202         unsigned int percpusize = 0;
1203         int i;
1204
1205         if (num == 0)
1206                 return NULL;
1207
1208         if (lprocfs_no_percpu_stats != 0)
1209                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1210
1211         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1212                 num_entry = 1;
1213         else
1214                 num_entry = num_possible_cpus();
1215
1216         /* alloc percpu pointers for all possible cpu slots */
1217         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1218         if (!stats)
1219                 return NULL;
1220
1221         stats->ls_num = num;
1222         stats->ls_flags = flags;
1223         spin_lock_init(&stats->ls_lock);
1224
1225         /* alloc num of counter headers */
1226         CFS_ALLOC_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1227         if (!stats->ls_cnt_header)
1228                 goto fail;
1229
1230         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1231                 /* contains only one set counters */
1232                 percpusize = lprocfs_stats_counter_size(stats);
1233                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1234                 if (!stats->ls_percpu[0])
1235                         goto fail;
1236                 stats->ls_biggest_alloc_num = 1;
1237         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1238                 /* alloc all percpu data, currently only obd_memory use this */
1239                 for (i = 0; i < num_entry; ++i)
1240                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1241                                 goto fail;
1242         }
1243
1244         return stats;
1245
1246 fail:
1247         lprocfs_free_stats(&stats);
1248         return NULL;
1249 }
1250 EXPORT_SYMBOL(lprocfs_alloc_stats);
1251
1252 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1253 {
1254         struct lprocfs_stats *stats = *statsh;
1255         unsigned int num_entry;
1256         unsigned int percpusize;
1257         unsigned int i;
1258
1259         if (!stats || stats->ls_num == 0)
1260                 return;
1261         *statsh = NULL;
1262
1263         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1264                 num_entry = 1;
1265         else
1266                 num_entry = num_possible_cpus();
1267
1268         percpusize = lprocfs_stats_counter_size(stats);
1269         for (i = 0; i < num_entry; i++)
1270                 if (stats->ls_percpu[i])
1271                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1272         if (stats->ls_cnt_header)
1273                 CFS_FREE_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1274         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1275 }
1276 EXPORT_SYMBOL(lprocfs_free_stats);
1277
1278 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1279                             enum lprocfs_fields_flags field)
1280 {
1281         unsigned long flags = 0;
1282         unsigned int num_cpu;
1283         unsigned int i;
1284         u64 ret = 0;
1285
1286         LASSERT(stats);
1287
1288         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1289         for (i = 0; i < num_cpu; i++) {
1290                 struct lprocfs_counter *cntr;
1291
1292                 if (!stats->ls_percpu[i])
1293                         continue;
1294
1295                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1296                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1297                                            stats->ls_flags, field);
1298         }
1299         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1300         return ret;
1301 }
1302 EXPORT_SYMBOL(lprocfs_stats_collector);
1303
1304 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1305 {
1306         struct lprocfs_counter *percpu_cntr;
1307         int i;
1308         int j;
1309         unsigned int num_entry;
1310         unsigned long flags = 0;
1311
1312         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1313
1314         for (i = 0; i < num_entry; i++) {
1315                 if (!stats->ls_percpu[i])
1316                         continue;
1317                 for (j = 0; j < stats->ls_num; j++) {
1318                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1319                         percpu_cntr->lc_count           = 0;
1320                         percpu_cntr->lc_min             = LC_MIN_INIT;
1321                         percpu_cntr->lc_max             = 0;
1322                         percpu_cntr->lc_sumsquare       = 0;
1323                         percpu_cntr->lc_sum             = 0;
1324                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1325                                 percpu_cntr->lc_sum_irq = 0;
1326                 }
1327         }
1328
1329         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1330 }
1331 EXPORT_SYMBOL(lprocfs_clear_stats);
1332
1333 static ssize_t lprocfs_stats_seq_write(struct file *file,
1334                                        const char __user *buf,
1335                                        size_t len, loff_t *off)
1336 {
1337         struct seq_file *seq = file->private_data;
1338         struct lprocfs_stats *stats = seq->private;
1339
1340         lprocfs_clear_stats(stats);
1341
1342         return len;
1343 }
1344
1345 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1346 {
1347         struct lprocfs_stats *stats = p->private;
1348
1349         return (*pos < stats->ls_num) ? pos : NULL;
1350 }
1351
1352 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1353 {
1354 }
1355
1356 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1357 {
1358         (*pos)++;
1359
1360         return lprocfs_stats_seq_start(p, pos);
1361 }
1362
1363 /* seq file export of one lprocfs counter */
1364 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1365 {
1366         struct lprocfs_stats *stats = p->private;
1367         struct lprocfs_counter_header *hdr;
1368         struct lprocfs_counter ctr;
1369         int idx = *(loff_t *)v;
1370
1371         if (idx == 0) {
1372                 struct timespec64 now;
1373
1374                 ktime_get_real_ts64(&now);
1375                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1376                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1377         }
1378
1379         hdr = &stats->ls_cnt_header[idx];
1380         lprocfs_stats_collect(stats, idx, &ctr);
1381
1382         if (ctr.lc_count == 0)
1383                 return 0;
1384
1385         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1386                    ctr.lc_count, hdr->lc_units);
1387
1388         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1389                 seq_printf(p, " %lld %lld %lld",
1390                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1391                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1392                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1393         }
1394         seq_putc(p, '\n');
1395         return 0;
1396 }
1397
1398 static const struct seq_operations lprocfs_stats_seq_sops = {
1399         .start  = lprocfs_stats_seq_start,
1400         .stop   = lprocfs_stats_seq_stop,
1401         .next   = lprocfs_stats_seq_next,
1402         .show   = lprocfs_stats_seq_show,
1403 };
1404
1405 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1406 {
1407         struct seq_file *seq;
1408         int rc;
1409
1410         rc = seq_open(file, &lprocfs_stats_seq_sops);
1411         if (rc)
1412                 return rc;
1413         seq = file->private_data;
1414         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1415         return 0;
1416 }
1417
1418 static const struct file_operations lprocfs_stats_seq_fops = {
1419         .owner   = THIS_MODULE,
1420         .open    = lprocfs_stats_seq_open,
1421         .read    = seq_read,
1422         .write   = lprocfs_stats_seq_write,
1423         .llseek  = seq_lseek,
1424         .release = lprocfs_seq_release,
1425 };
1426
1427 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1428                             struct lprocfs_stats *stats)
1429 {
1430         struct dentry *entry;
1431
1432         LASSERT(!IS_ERR_OR_NULL(parent));
1433
1434         entry = debugfs_create_file(name, 0644, parent, stats,
1435                                     &lprocfs_stats_seq_fops);
1436         if (IS_ERR_OR_NULL(entry))
1437                 return entry ? PTR_ERR(entry) : -ENOMEM;
1438
1439         return 0;
1440 }
1441 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1442
1443 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1444                            struct lprocfs_stats *stats)
1445 {
1446         struct proc_dir_entry *entry;
1447         LASSERT(root != NULL);
1448
1449         entry = proc_create_data(name, 0644, root,
1450                                  &lprocfs_stats_seq_fops, stats);
1451         if (!entry)
1452                 return -ENOMEM;
1453         return 0;
1454 }
1455 EXPORT_SYMBOL(lprocfs_register_stats);
1456
1457 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1458                           unsigned conf, const char *name, const char *units)
1459 {
1460         struct lprocfs_counter_header *header;
1461         struct lprocfs_counter *percpu_cntr;
1462         unsigned long flags = 0;
1463         unsigned int i;
1464         unsigned int num_cpu;
1465
1466         LASSERT(stats != NULL);
1467
1468         header = &stats->ls_cnt_header[index];
1469         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1470                  index, name, units);
1471
1472         header->lc_config = conf;
1473         header->lc_name   = name;
1474         header->lc_units  = units;
1475
1476         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1477         for (i = 0; i < num_cpu; ++i) {
1478                 if (!stats->ls_percpu[i])
1479                         continue;
1480                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1481                 percpu_cntr->lc_count           = 0;
1482                 percpu_cntr->lc_min             = LC_MIN_INIT;
1483                 percpu_cntr->lc_max             = 0;
1484                 percpu_cntr->lc_sumsquare       = 0;
1485                 percpu_cntr->lc_sum             = 0;
1486                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1487                         percpu_cntr->lc_sum_irq = 0;
1488         }
1489         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1490 }
1491 EXPORT_SYMBOL(lprocfs_counter_init);
1492
1493 static const char * const mps_stats[] = {
1494         [LPROC_MD_CLOSE]                = "close",
1495         [LPROC_MD_CREATE]               = "create",
1496         [LPROC_MD_ENQUEUE]              = "enqueue",
1497         [LPROC_MD_GETATTR]              = "getattr",
1498         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1499         [LPROC_MD_LINK]                 = "link",
1500         [LPROC_MD_RENAME]               = "rename",
1501         [LPROC_MD_SETATTR]              = "setattr",
1502         [LPROC_MD_FSYNC]                = "fsync",
1503         [LPROC_MD_READ_PAGE]            = "read_page",
1504         [LPROC_MD_UNLINK]               = "unlink",
1505         [LPROC_MD_SETXATTR]             = "setxattr",
1506         [LPROC_MD_GETXATTR]             = "getxattr",
1507         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1508         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1509 };
1510
1511 int lprocfs_alloc_md_stats(struct obd_device *obd,
1512                            unsigned int num_private_stats)
1513 {
1514         struct lprocfs_stats *stats;
1515         unsigned int num_stats;
1516         int rc, i;
1517
1518         /*
1519          * TODO Ensure that this function is only used where
1520          * appropriate by adding an assertion to the effect that
1521          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1522          * because mdt_procfs_init() uses this function to allocate
1523          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1524          * mdt layer does not use the md_ops interface. This is
1525          * confusing and a waste of memory. See LU-2484.
1526          */
1527         LASSERT(obd->obd_proc_entry != NULL);
1528         LASSERT(obd->obd_md_stats == NULL);
1529
1530         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1531         stats = lprocfs_alloc_stats(num_stats, 0);
1532         if (!stats)
1533                 return -ENOMEM;
1534
1535         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1536                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1537                 if (!stats->ls_cnt_header[i].lc_name) {
1538                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1539                                i);
1540                         LBUG();
1541                 }
1542         }
1543
1544         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1545         if (rc < 0) {
1546                 lprocfs_free_stats(&stats);
1547         } else {
1548                 obd->obd_md_stats = stats;
1549         }
1550
1551         return rc;
1552 }
1553 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1554
1555 void lprocfs_free_md_stats(struct obd_device *obd)
1556 {
1557         struct lprocfs_stats *stats = obd->obd_md_stats;
1558
1559         if (stats) {
1560                 obd->obd_md_stats = NULL;
1561                 lprocfs_free_stats(&stats);
1562         }
1563 }
1564 EXPORT_SYMBOL(lprocfs_free_md_stats);
1565
1566 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1567 {
1568         lprocfs_counter_init(ldlm_stats,
1569                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1570                              0, "ldlm_enqueue", "reqs");
1571         lprocfs_counter_init(ldlm_stats,
1572                              LDLM_CONVERT - LDLM_FIRST_OPC,
1573                              0, "ldlm_convert", "reqs");
1574         lprocfs_counter_init(ldlm_stats,
1575                              LDLM_CANCEL - LDLM_FIRST_OPC,
1576                              0, "ldlm_cancel", "reqs");
1577         lprocfs_counter_init(ldlm_stats,
1578                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1579                              0, "ldlm_bl_callback", "reqs");
1580         lprocfs_counter_init(ldlm_stats,
1581                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1582                              0, "ldlm_cp_callback", "reqs");
1583         lprocfs_counter_init(ldlm_stats,
1584                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1585                              0, "ldlm_gl_callback", "reqs");
1586 }
1587 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1588
1589 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1590                           struct lprocfs_counter_header *header,
1591                           enum lprocfs_stats_flags flags,
1592                           enum lprocfs_fields_flags field)
1593 {
1594         __s64 ret = 0;
1595
1596         if (!lc || !header)
1597                 RETURN(0);
1598
1599         switch (field) {
1600                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1601                         ret = header->lc_config;
1602                         break;
1603                 case LPROCFS_FIELDS_FLAGS_SUM:
1604                         ret = lc->lc_sum;
1605                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1606                                 ret += lc->lc_sum_irq;
1607                         break;
1608                 case LPROCFS_FIELDS_FLAGS_MIN:
1609                         ret = lc->lc_min;
1610                         break;
1611                 case LPROCFS_FIELDS_FLAGS_MAX:
1612                         ret = lc->lc_max;
1613                         break;
1614                 case LPROCFS_FIELDS_FLAGS_AVG:
1615                         ret = (lc->lc_max - lc->lc_min) / 2;
1616                         break;
1617                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1618                         ret = lc->lc_sumsquare;
1619                         break;
1620                 case LPROCFS_FIELDS_FLAGS_COUNT:
1621                         ret = lc->lc_count;
1622                         break;
1623                 default:
1624                         break;
1625         };
1626         RETURN(ret);
1627 }
1628 EXPORT_SYMBOL(lprocfs_read_helper);
1629
1630 /**
1631  * string_to_size - convert ASCII string representing a numerical
1632  *                  value with optional units to 64-bit binary value
1633  *
1634  * @size:       The numerical value extract out of @buffer
1635  * @buffer:     passed in string to parse
1636  * @count:      length of the @buffer
1637  *
1638  * This function returns a 64-bit binary value if @buffer contains a valid
1639  * numerical string. The string is parsed to 3 significant figures after
1640  * the decimal point. Support the string containing an optional units at
1641  * the end which can be base 2 or base 10 in value. If no units are given
1642  * the string is assumed to just a numerical value.
1643  *
1644  * Returns:     @count if the string is successfully parsed,
1645  *              -errno on invalid input strings. Error values:
1646  *
1647  *  - ``-EINVAL``: @buffer is not a proper numerical string
1648  *  - ``-EOVERFLOW``: results does not fit into 64 bits.
1649  *  - ``-E2BIG ``: @buffer is too large (not a valid number)
1650  */
1651 int string_to_size(u64 *size, const char *buffer, size_t count)
1652 {
1653         /* For string_get_size() it can support values above exabytes,
1654          * (ZiB, YiB) due to breaking the return value into a size and
1655          * bulk size to avoid 64 bit overflow. We don't break the size
1656          * up into block size units so we don't support ZiB or YiB.
1657          */
1658         static const char *const units_10[] = {
1659                 "kB", "MB", "GB", "TB", "PB", "EB",
1660         };
1661         static const char *const units_2[] = {
1662                 "K",  "M",  "G",  "T",  "P",  "E",
1663         };
1664         static const char *const *const units_str[] = {
1665                 [STRING_UNITS_2] = units_2,
1666                 [STRING_UNITS_10] = units_10,
1667         };
1668         static const unsigned int coeff[] = {
1669                 [STRING_UNITS_10] = 1000,
1670                 [STRING_UNITS_2] = 1024,
1671         };
1672         enum string_size_units unit = STRING_UNITS_2;
1673         u64 whole, blk_size = 1;
1674         char kernbuf[22], *end;
1675         size_t len = count;
1676         int rc;
1677         int i;
1678
1679         if (count >= sizeof(kernbuf)) {
1680                 CERROR("count %zd > buffer %zd\n", count, sizeof(kernbuf));
1681                 return -E2BIG;
1682         }
1683
1684         *size = 0;
1685         /* The "iB" suffix is optionally allowed for indicating base-2 numbers.
1686          * If suffix is only "B" and not "iB" then we treat it as base-10.
1687          */
1688         end = strstr(buffer, "B");
1689         if (end && *(end - 1) != 'i')
1690                 unit = STRING_UNITS_10;
1691
1692         i = unit == STRING_UNITS_2 ? ARRAY_SIZE(units_2) - 1 :
1693                                      ARRAY_SIZE(units_10) - 1;
1694         do {
1695                 end = strnstr(buffer, units_str[unit][i], count);
1696                 if (end) {
1697                         for (; i >= 0; i--)
1698                                 blk_size *= coeff[unit];
1699                         len = end - buffer;
1700                         break;
1701                 }
1702         } while (i--);
1703
1704         /* as 'B' is a substring of all units, we need to handle it
1705          * separately.
1706          */
1707         if (!end) {
1708                 /* 'B' is only acceptable letter at this point */
1709                 end = strnchr(buffer, count, 'B');
1710                 if (end) {
1711                         len = end - buffer;
1712
1713                         if (count - len > 2 ||
1714                             (count - len == 2 && strcmp(end, "B\n") != 0)) {
1715                                 CDEBUG(D_INFO, "unknown suffix '%s'\n", buffer);
1716                                 return -EINVAL;
1717                         }
1718                 }
1719                 /* kstrtoull will error out if it has non digits */
1720                 goto numbers_only;
1721         }
1722
1723         end = strnchr(buffer, count, '.');
1724         if (end) {
1725                 /* need to limit 3 decimal places */
1726                 char rem[4] = "000";
1727                 u64 frac = 0;
1728                 size_t off;
1729
1730                 len = end - buffer;
1731                 end++;
1732
1733                 /* limit to 3 decimal points */
1734                 off = min_t(size_t, 3, strspn(end, "0123456789"));
1735                 /* need to limit frac_d to a u32 */
1736                 memcpy(rem, end, off);
1737                 rc = kstrtoull(rem, 10, &frac);
1738                 if (rc)
1739                         return rc;
1740
1741                 if (fls64(frac) + fls64(blk_size) - 1 > 64)
1742                         return -EOVERFLOW;
1743
1744                 frac *= blk_size;
1745                 do_div(frac, 1000);
1746                 *size += frac;
1747         }
1748 numbers_only:
1749         snprintf(kernbuf, sizeof(kernbuf), "%.*s", (int)len, buffer);
1750         rc = kstrtoull(kernbuf, 10, &whole);
1751         if (rc)
1752                 return rc;
1753
1754         if (whole != 0 && fls64(whole) + fls64(blk_size) - 1 > 64)
1755                 return -EOVERFLOW;
1756
1757         *size += whole * blk_size;
1758
1759         return count;
1760 }
1761 EXPORT_SYMBOL(string_to_size);
1762
1763 /**
1764  * sysfs_memparse - parse a ASCII string to 64-bit binary value,
1765  *                  with optional units
1766  *
1767  * @buffer:     kernel pointer to input string
1768  * @count:      number of bytes in the input @buffer
1769  * @val:        (output) binary value returned to caller
1770  * @defunit:    default unit suffix to use if none is provided
1771  *
1772  * Parses a string into a number. The number stored at @buffer is
1773  * potentially suffixed with K, M, G, T, P, E. Besides these other
1774  * valid suffix units are shown in the string_to_size() function.
1775  * If the string lacks a suffix then the defunit is used. The defunit
1776  * should be given as a binary unit (e.g. MiB) as that is the standard
1777  * for tunables in Lustre. If no unit suffix is given (e.g. 'G'), then
1778  * it is assumed to be in binary units.
1779  *
1780  * Returns:     0 on success or -errno on failure.
1781  */
1782 int sysfs_memparse(const char *buffer, size_t count, u64 *val,
1783                    const char *defunit)
1784 {
1785         const char *param = buffer;
1786         char tmp_buf[23];
1787         int rc;
1788
1789         count = strlen(buffer);
1790         while (count > 0 && isspace(buffer[count - 1]))
1791                 count--;
1792
1793         if (!count)
1794                 RETURN(-EINVAL);
1795
1796         /* If there isn't already a unit on this value, append @defunit.
1797          * Units of 'B' don't affect the value, so don't bother adding.
1798          */
1799         if (!isalpha(buffer[count - 1]) && defunit[0] != 'B') {
1800                 if (count + 3 >= sizeof(tmp_buf)) {
1801                         CERROR("count %zd > size %zd\n", count, sizeof(param));
1802                         RETURN(-E2BIG);
1803                 }
1804
1805                 scnprintf(tmp_buf, sizeof(tmp_buf), "%.*s%s", (int)count,
1806                           buffer, defunit);
1807                 param = tmp_buf;
1808                 count = strlen(param);
1809         }
1810
1811         rc = string_to_size(val, param, count);
1812
1813         return rc < 0 ? rc : 0;
1814 }
1815 EXPORT_SYMBOL(sysfs_memparse);
1816
1817 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
1818 {
1819         size_t l2;
1820
1821         l2 = strlen(s2);
1822         if (!l2)
1823                 return (char *)s1;
1824         while (len >= l2) {
1825                 len--;
1826                 if (!memcmp(s1, s2, l2))
1827                         return (char *)s1;
1828                 s1++;
1829         }
1830         return NULL;
1831 }
1832 EXPORT_SYMBOL(lprocfs_strnstr);
1833
1834 /**
1835  * Find the string \a name in the input \a buffer, and return a pointer to the
1836  * value immediately following \a name, reducing \a count appropriately.
1837  * If \a name is not found the original \a buffer is returned.
1838  */
1839 char *lprocfs_find_named_value(const char *buffer, const char *name,
1840                                 size_t *count)
1841 {
1842         char *val;
1843         size_t buflen = *count;
1844
1845         /* there is no strnstr() in rhel5 and ubuntu kernels */
1846         val = lprocfs_strnstr(buffer, name, buflen);
1847         if (!val)
1848                 return (char *)buffer;
1849
1850         val += strlen(name);                             /* skip prefix */
1851         while (val < buffer + buflen && isspace(*val)) /* skip separator */
1852                 val++;
1853
1854         *count = 0;
1855         while (val < buffer + buflen && isalnum(*val)) {
1856                 ++*count;
1857                 ++val;
1858         }
1859
1860         return val - *count;
1861 }
1862 EXPORT_SYMBOL(lprocfs_find_named_value);
1863
1864 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
1865                         const struct file_operations *seq_fops, void *data)
1866 {
1867         struct dentry *entry;
1868
1869         /* Disallow secretly (un)writable entries. */
1870         LASSERT((!seq_fops->write) == (!(mode & 0222)));
1871
1872         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
1873         if (IS_ERR_OR_NULL(entry))
1874                 return entry ? PTR_ERR(entry) : -ENOMEM;
1875
1876         return 0;
1877 }
1878 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
1879
1880 int lprocfs_seq_create(struct proc_dir_entry *parent,
1881                        const char *name,
1882                        mode_t mode,
1883                        const struct file_operations *seq_fops,
1884                        void *data)
1885 {
1886         struct proc_dir_entry *entry;
1887         ENTRY;
1888
1889         /* Disallow secretly (un)writable entries. */
1890         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
1891
1892         entry = proc_create_data(name, mode, parent, seq_fops, data);
1893
1894         if (!entry)
1895                 RETURN(-ENOMEM);
1896
1897         RETURN(0);
1898 }
1899 EXPORT_SYMBOL(lprocfs_seq_create);
1900
1901 int lprocfs_obd_seq_create(struct obd_device *obd,
1902                            const char *name,
1903                            mode_t mode,
1904                            const struct file_operations *seq_fops,
1905                            void *data)
1906 {
1907         return lprocfs_seq_create(obd->obd_proc_entry, name,
1908                                   mode, seq_fops, data);
1909 }
1910 EXPORT_SYMBOL(lprocfs_obd_seq_create);
1911
1912 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
1913 {
1914         if (value >= OBD_HIST_MAX)
1915                 value = OBD_HIST_MAX - 1;
1916
1917         spin_lock(&oh->oh_lock);
1918         oh->oh_buckets[value]++;
1919         spin_unlock(&oh->oh_lock);
1920 }
1921 EXPORT_SYMBOL(lprocfs_oh_tally);
1922
1923 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
1924 {
1925         unsigned int val = 0;
1926
1927         if (likely(value != 0))
1928                 val = min(fls(value - 1), OBD_HIST_MAX);
1929
1930         lprocfs_oh_tally(oh, val);
1931 }
1932 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
1933
1934 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
1935 {
1936         unsigned long ret = 0;
1937         int i;
1938
1939         for (i = 0; i < OBD_HIST_MAX; i++)
1940                 ret +=  oh->oh_buckets[i];
1941         return ret;
1942 }
1943 EXPORT_SYMBOL(lprocfs_oh_sum);
1944
1945 void lprocfs_oh_clear(struct obd_histogram *oh)
1946 {
1947         spin_lock(&oh->oh_lock);
1948         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
1949         spin_unlock(&oh->oh_lock);
1950 }
1951 EXPORT_SYMBOL(lprocfs_oh_clear);
1952
1953 ssize_t lustre_attr_show(struct kobject *kobj,
1954                          struct attribute *attr, char *buf)
1955 {
1956         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
1957
1958         return a->show ? a->show(kobj, attr, buf) : 0;
1959 }
1960 EXPORT_SYMBOL_GPL(lustre_attr_show);
1961
1962 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
1963                           const char *buf, size_t len)
1964 {
1965         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
1966
1967         return a->store ? a->store(kobj, attr, buf, len) : len;
1968 }
1969 EXPORT_SYMBOL_GPL(lustre_attr_store);
1970
1971 const struct sysfs_ops lustre_sysfs_ops = {
1972         .show  = lustre_attr_show,
1973         .store = lustre_attr_store,
1974 };
1975 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
1976
1977 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
1978 {
1979         struct obd_device *obd = data;
1980         struct client_obd *cli = &obd->u.cli;
1981
1982         spin_lock(&cli->cl_loi_list_lock);
1983         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
1984         spin_unlock(&cli->cl_loi_list_lock);
1985         return 0;
1986 }
1987 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
1988
1989 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
1990                                                 const char __user *buffer,
1991                                                 size_t count, loff_t *off)
1992 {
1993         struct seq_file *m = file->private_data;
1994         struct obd_device *obd = m->private;
1995         struct client_obd *cli = &obd->u.cli;
1996         struct obd_import *imp;
1997         struct obd_connect_data *ocd;
1998         int chunk_mask, rc;
1999         char kernbuf[22];
2000         u64 val;
2001
2002         if (count > sizeof(kernbuf) - 1)
2003                 return -EINVAL;
2004
2005         if (copy_from_user(kernbuf, buffer, count))
2006                 return -EFAULT;
2007
2008         kernbuf[count] = '\0';
2009
2010         rc = sysfs_memparse(kernbuf, count, &val, "B");
2011         if (rc)
2012                 return rc;
2013
2014         /* if the max_pages is specified in bytes, convert to pages */
2015         if (val >= ONE_MB_BRW_SIZE)
2016                 val >>= PAGE_SHIFT;
2017
2018         with_imp_locked(obd, imp, rc) {
2019                 ocd = &imp->imp_connect_data;
2020                 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2021                 /* max_pages_per_rpc must be chunk aligned */
2022                 val = (val + ~chunk_mask) & chunk_mask;
2023                 if (val == 0 || (ocd->ocd_brw_size != 0 &&
2024                                  val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2025                         rc = -ERANGE;
2026                 } else {
2027                         spin_lock(&cli->cl_loi_list_lock);
2028                         cli->cl_max_pages_per_rpc = val;
2029                         client_adjust_max_dirty(cli);
2030                         spin_unlock(&cli->cl_loi_list_lock);
2031                 }
2032         }
2033
2034         return rc ?: count;
2035 }
2036 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2037
2038 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2039                             char *buf)
2040 {
2041         struct obd_device *obd = container_of(kobj, struct obd_device,
2042                                               obd_kset.kobj);
2043         struct client_obd *cli = &obd->u.cli;
2044         int rc;
2045
2046         spin_lock(&cli->cl_loi_list_lock);
2047         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2048         spin_unlock(&cli->cl_loi_list_lock);
2049         return rc;
2050 }
2051 EXPORT_SYMBOL(short_io_bytes_show);
2052
2053 /* Used to catch people who think they're specifying pages. */
2054 #define MIN_SHORT_IO_BYTES 64U
2055
2056 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2057                              const char *buffer, size_t count)
2058 {
2059         struct obd_device *obd = container_of(kobj, struct obd_device,
2060                                               obd_kset.kobj);
2061         struct client_obd *cli = &obd->u.cli;
2062         u64 val;
2063         int rc;
2064
2065         if (strcmp(buffer, "-1") == 0) {
2066                 val = OBD_DEF_SHORT_IO_BYTES;
2067         } else {
2068                 rc = sysfs_memparse(buffer, count, &val, "B");
2069                 if (rc)
2070                         GOTO(out, rc);
2071         }
2072
2073         if (val && (val < MIN_SHORT_IO_BYTES || val > LNET_MTU))
2074                 GOTO(out, rc = -ERANGE);
2075
2076         rc = count;
2077
2078         spin_lock(&cli->cl_loi_list_lock);
2079         cli->cl_max_short_io_bytes = min_t(u64, val, OST_MAX_SHORT_IO_BYTES);
2080         spin_unlock(&cli->cl_loi_list_lock);
2081
2082 out:
2083         return rc;
2084 }
2085 EXPORT_SYMBOL(short_io_bytes_store);
2086
2087 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2088                            struct root_squash_info *squash, char *name)
2089 {
2090         int rc;
2091         char kernbuf[64], *tmp, *errmsg;
2092         unsigned long uid, gid;
2093         ENTRY;
2094
2095         if (count >= sizeof(kernbuf)) {
2096                 errmsg = "string too long";
2097                 GOTO(failed_noprint, rc = -EINVAL);
2098         }
2099         if (copy_from_user(kernbuf, buffer, count)) {
2100                 errmsg = "bad address";
2101                 GOTO(failed_noprint, rc = -EFAULT);
2102         }
2103         kernbuf[count] = '\0';
2104
2105         /* look for uid gid separator */
2106         tmp = strchr(kernbuf, ':');
2107         if (!tmp) {
2108                 errmsg = "needs uid:gid format";
2109                 GOTO(failed, rc = -EINVAL);
2110         }
2111         *tmp = '\0';
2112         tmp++;
2113
2114         /* parse uid */
2115         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2116                 errmsg = "bad uid";
2117                 GOTO(failed, rc = -EINVAL);
2118         }
2119
2120         /* parse gid */
2121         if (kstrtoul(tmp, 0, &gid) != 0) {
2122                 errmsg = "bad gid";
2123                 GOTO(failed, rc = -EINVAL);
2124         }
2125
2126         squash->rsi_uid = uid;
2127         squash->rsi_gid = gid;
2128
2129         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2130                       name, squash->rsi_uid, squash->rsi_gid);
2131         RETURN(count);
2132
2133 failed:
2134         if (tmp) {
2135                 tmp--;
2136                 *tmp = ':';
2137         }
2138         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2139               name, kernbuf, errmsg, rc);
2140         RETURN(rc);
2141 failed_noprint:
2142         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2143               name, errmsg, rc);
2144         RETURN(rc);
2145 }
2146 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2147
2148
2149 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2150                              struct root_squash_info *squash, char *name)
2151 {
2152         int rc;
2153         char *kernbuf = NULL;
2154         char *errmsg;
2155         LIST_HEAD(tmp);
2156         int len = count;
2157         ENTRY;
2158
2159         if (count > 4096) {
2160                 errmsg = "string too long";
2161                 GOTO(failed, rc = -EINVAL);
2162         }
2163
2164         OBD_ALLOC(kernbuf, count + 1);
2165         if (!kernbuf) {
2166                 errmsg = "no memory";
2167                 GOTO(failed, rc = -ENOMEM);
2168         }
2169         if (copy_from_user(kernbuf, buffer, count)) {
2170                 errmsg = "bad address";
2171                 GOTO(failed, rc = -EFAULT);
2172         }
2173         kernbuf[count] = '\0';
2174
2175         if (count > 0 && kernbuf[count - 1] == '\n')
2176                 len = count - 1;
2177
2178         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2179             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2180                 /* empty string is special case */
2181                 spin_lock(&squash->rsi_lock);
2182                 if (!list_empty(&squash->rsi_nosquash_nids))
2183                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2184                 spin_unlock(&squash->rsi_lock);
2185                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2186                 OBD_FREE(kernbuf, count + 1);
2187                 RETURN(count);
2188         }
2189
2190         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2191                 errmsg = "can't parse";
2192                 GOTO(failed, rc = -EINVAL);
2193         }
2194         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2195                       name, kernbuf);
2196         OBD_FREE(kernbuf, count + 1);
2197         kernbuf = NULL;
2198
2199         spin_lock(&squash->rsi_lock);
2200         if (!list_empty(&squash->rsi_nosquash_nids))
2201                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2202         list_splice(&tmp, &squash->rsi_nosquash_nids);
2203         spin_unlock(&squash->rsi_lock);
2204
2205         RETURN(count);
2206
2207 failed:
2208         if (kernbuf) {
2209                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2210                       name, kernbuf, errmsg, rc);
2211                 OBD_FREE(kernbuf, count + 1);
2212         } else {
2213                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2214                       name, errmsg, rc);
2215         }
2216         RETURN(rc);
2217 }
2218 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2219
2220 #endif /* CONFIG_PROC_FS*/