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 == NULL || name == NULL || fops == NULL)
  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, const char *format, ...)
 111 {
 112         struct proc_dir_entry *entry;
 113         char *dest;
 114         va_list ap;
 115
 116         if (parent == NULL || format == NULL)
 117                 return NULL;
 118
 119         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
 120         if (dest == NULL)
 121                 return NULL;
 122
 123         va_start(ap, format);
 124         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
 125         va_end(ap);
 126
 127         entry = proc_symlink(name, parent, dest);
 128         if (entry == NULL)
 129                 CERROR("LprocFS: Could not create symbolic link from "
 130                        "%s to %s\n", name, dest);
 131
 132         OBD_FREE(dest, MAX_STRING_SIZE + 1);
 133         return entry;
 134 }
 135 EXPORT_SYMBOL(lprocfs_add_symlink);
 136
 137 static const struct file_operations lprocfs_generic_fops = { };
 138
 139 int ldebugfs_add_vars(struct dentry *parent, struct lprocfs_vars *list,
 140                       void *data)
 141 {
 142         if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
 143                 return -EINVAL;
 144
 145         while (list->name) {
 146                 struct dentry *entry;
 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                 entry = debugfs_create_file(list->name, mode, parent,
 158                                             list->data ? : data,
 159                                             list->fops ? : &lprocfs_generic_fops);
 160                 if (IS_ERR_OR_NULL(entry))
 161                         return entry ? PTR_ERR(entry) : -ENOMEM;
 162                 list++;
 163         }
 164         return 0;
 165 }
 166 EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
 167
 168 /**
 169  * Add /proc entries.
 170  *
 171  * \param root [in]  The parent proc entry on which new entry will be added.
 172  * \param list [in]  Array of proc entries to be added.
 173  * \param data [in]  The argument to be passed when entries read/write routines
 174  *                   are called through /proc file.
 175  *
 176  * \retval 0   on success
 177  *         < 0 on error
 178  */
 179 int
 180 lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
 181                  void *data)
 182 {
 183         if (root == NULL || list == NULL)
 184                 return -EINVAL;
 185
 186         while (list->name != NULL) {
 187                 struct proc_dir_entry *proc;
 188                 mode_t mode = 0;
 189
 190                 if (list->proc_mode != 0000) {
 191                         mode = list->proc_mode;
 192                 } else if (list->fops) {
 193                         if (list->fops->read)
 194                                 mode = 0444;
 195                         if (list->fops->write)
 196                                 mode |= 0200;
 197                 }
 198                 proc = proc_create_data(list->name, mode, root,
 199                                         list->fops ?: &lprocfs_generic_fops,
 200                                         list->data ?: data);
 201                 if (proc == NULL)
 202                         return -ENOMEM;
 203                 list++;
 204         }
 205         return 0;
 206 }
 207 EXPORT_SYMBOL(lprocfs_add_vars);
 208
 209 void ldebugfs_remove(struct dentry **entryp)
 210 {
 211         debugfs_remove_recursive(*entryp);
 212         *entryp = NULL;
 213 }
 214 EXPORT_SYMBOL_GPL(ldebugfs_remove);
 215
 216 #ifndef HAVE_REMOVE_PROC_SUBTREE
 217 /* for b=10866, global variable */
 218 DECLARE_RWSEM(_lprocfs_lock);
 219 EXPORT_SYMBOL(_lprocfs_lock);
 220
 221 static void lprocfs_remove_nolock(struct proc_dir_entry **proot)
 222 {
 223         struct proc_dir_entry *root = *proot;
 224         struct proc_dir_entry *temp = root;
 225         struct proc_dir_entry *rm_entry;
 226         struct proc_dir_entry *parent;
 227
 228         *proot = NULL;
 229         if (root == NULL || IS_ERR(root))
 230                 return;
 231
 232         parent = root->parent;
 233         LASSERT(parent != NULL);
 234
 235         while (1) {
 236                 while (temp->subdir != NULL)
 237                         temp = temp->subdir;
 238
 239                 rm_entry = temp;
 240                 temp = temp->parent;
 241
 242                 /* Memory corruption once caused this to fail, and
 243                    without this LASSERT we would loop here forever. */
 244                 LASSERTF(strlen(rm_entry->name) == rm_entry->namelen,
 245                          "0x%p  %s/%s len %d\n", rm_entry, temp->name,
 246                          rm_entry->name, (int)strlen(rm_entry->name));
 247
 248                 remove_proc_entry(rm_entry->name, temp);
 249                 if (temp == parent)
 250                         break;
 251         }
 252 }
 253
 254 int remove_proc_subtree(const char *name, struct proc_dir_entry *parent)
 255 {
 256         struct proc_dir_entry    *t = NULL;
 257         struct proc_dir_entry   **p;
 258         int                       len, busy = 0;
 259
 260         LASSERT(parent != NULL);
 261         len = strlen(name);
 262
 263         down_write(&_lprocfs_lock);
 264         /* lookup target name */
 265         for (p = &parent->subdir; *p; p = &(*p)->next) {
 266                 if ((*p)->namelen != len)
 267                         continue;
 268                 if (memcmp(name, (*p)->name, len))
 269                         continue;
 270                 t = *p;
 271                 break;
 272         }
 273
 274         if (t) {
 275                 /* verify it's empty: do not count "num_refs" */
 276                 for (p = &t->subdir; *p; p = &(*p)->next) {
 277                         if ((*p)->namelen != strlen("num_refs")) {
 278                                 busy = 1;
 279                                 break;
 280                         }
 281                         if (memcmp("num_refs", (*p)->name,
 282                                    strlen("num_refs"))) {
 283                                 busy = 1;
 284                                 break;
 285                         }
 286                 }
 287         }
 288
 289         if (busy == 0)
 290                 lprocfs_remove_nolock(&t);
 291
 292         up_write(&_lprocfs_lock);
 293         return 0;
 294 }
 295 #endif /* !HAVE_REMOVE_PROC_SUBTREE */
 296
 297 #ifndef HAVE_PROC_REMOVE
 298 void proc_remove(struct proc_dir_entry *de)
 299 {
 300 #ifndef HAVE_REMOVE_PROC_SUBTREE
 301         down_write(&_lprocfs_lock); /* search vs remove race */
 302         lprocfs_remove_nolock(&de);
 303         up_write(&_lprocfs_lock);
 304 #else
 305         if (de)
 306                 remove_proc_subtree(de->name, de->parent);
 307 #endif
 308 }
 309 #endif
 310
 311 void lprocfs_remove(struct proc_dir_entry **rooth)
 312 {
 313         proc_remove(*rooth);
 314         *rooth = NULL;
 315 }
 316 EXPORT_SYMBOL(lprocfs_remove);
 317
 318 void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
 319 {
 320         LASSERT(parent != NULL);
 321         remove_proc_entry(name, parent);
 322 }
 323 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
 324
 325 struct dentry *ldebugfs_register(const char *name, struct dentry *parent,
 326                                  struct lprocfs_vars *list, void *data)
 327 {
 328         struct dentry *entry;
 329
 330         entry = debugfs_create_dir(name, parent);
 331         if (IS_ERR_OR_NULL(entry)) {
 332                 entry = entry ?: ERR_PTR(-ENOMEM);
 333                 goto out;
 334         }
 335
 336         if (!IS_ERR_OR_NULL(list)) {
 337                 int rc;
 338
 339                 rc = ldebugfs_add_vars(entry, list, data);
 340                 if (rc) {
 341                         debugfs_remove(entry);
 342                         entry = ERR_PTR(rc);
 343                 }
 344         }
 345 out:
 346         return entry;
 347 }
 348 EXPORT_SYMBOL_GPL(ldebugfs_register);
 349
 350 struct proc_dir_entry *
 351 lprocfs_register(const char *name, struct proc_dir_entry *parent,
 352                  struct lprocfs_vars *list, void *data)
 353 {
 354         struct proc_dir_entry *newchild;
 355
 356         newchild = proc_mkdir(name, parent);
 357         if (newchild == NULL)
 358                 return ERR_PTR(-ENOMEM);
 359
 360         if (list != NULL) {
 361                 int rc = lprocfs_add_vars(newchild, list, data);
 362                 if (rc) {
 363                         lprocfs_remove(&newchild);
 364                         return ERR_PTR(rc);
 365                 }
 366         }
 367         return newchild;
 368 }
 369 EXPORT_SYMBOL(lprocfs_register);
 370
 371 /* Generic callbacks */
 372 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
 373 {
 374         struct obd_device *obd = data;
 375
 376         LASSERT(obd != NULL);
 377         seq_printf(m, "%s\n", obd->obd_uuid.uuid);
 378         return 0;
 379 }
 380 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
 381
 382 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 383                          char *buf)
 384 {
 385         struct obd_device *obd = container_of(kobj, struct obd_device,
 386                                               obd_kset.kobj);
 387
 388         return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
 389 }
 390 LUSTRE_RO_ATTR(uuid);
 391
 392 int lprocfs_name_seq_show(struct seq_file *m, void *data)
 393 {
 394         struct obd_device *dev = data;
 395
 396         LASSERT(dev != NULL);
 397         seq_printf(m, "%s\n", dev->obd_name);
 398         return 0;
 399 }
 400 EXPORT_SYMBOL(lprocfs_name_seq_show);
 401
 402 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 403                               char *buf)
 404 {
 405         struct obd_device *obd = container_of(kobj, struct obd_device,
 406                                               obd_kset.kobj);
 407         struct obd_statfs osfs;
 408         int rc;
 409
 410         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 411                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 412                         OBD_STATFS_NODELAY);
 413         if (!rc)
 414                 return sprintf(buf, "%u\n", osfs.os_bsize);
 415
 416         return rc;
 417 }
 418 LUSTRE_RO_ATTR(blocksize);
 419
 420 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 421                                 char *buf)
 422 {
 423         struct obd_device *obd = container_of(kobj, struct obd_device,
 424                                               obd_kset.kobj);
 425         struct obd_statfs osfs;
 426         int rc;
 427
 428         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 429                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 430                         OBD_STATFS_NODELAY);
 431         if (!rc) {
 432                 u32 blk_size = osfs.os_bsize >> 10;
 433                 u64 result = osfs.os_blocks;
 434
 435                 while (blk_size >>= 1)
 436                         result <<= 1;
 437
 438                 return sprintf(buf, "%llu\n", result);
 439         }
 440
 441         return rc;
 442 }
 443 LUSTRE_RO_ATTR(kbytestotal);
 444
 445 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 446                                char *buf)
 447 {
 448         struct obd_device *obd = container_of(kobj, struct obd_device,
 449                                               obd_kset.kobj);
 450         struct obd_statfs osfs;
 451         int rc;
 452
 453         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 454                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 455                         OBD_STATFS_NODELAY);
 456         if (!rc) {
 457                 u32 blk_size = osfs.os_bsize >> 10;
 458                 u64 result = osfs.os_bfree;
 459
 460                 while (blk_size >>= 1)
 461                         result <<= 1;
 462
 463                 return sprintf(buf, "%llu\n", result);
 464         }
 465
 466         return rc;
 467 }
 468 LUSTRE_RO_ATTR(kbytesfree);
 469
 470 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 471                                 char *buf)
 472 {
 473         struct obd_device *obd = container_of(kobj, struct obd_device,
 474                                               obd_kset.kobj);
 475         struct obd_statfs osfs;
 476         int rc;
 477
 478         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 479                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 480                         OBD_STATFS_NODELAY);
 481         if (!rc) {
 482                 u32 blk_size = osfs.os_bsize >> 10;
 483                 u64 result = osfs.os_bavail;
 484
 485                 while (blk_size >>= 1)
 486                         result <<= 1;
 487
 488                 return sprintf(buf, "%llu\n", result);
 489         }
 490
 491         return rc;
 492 }
 493 LUSTRE_RO_ATTR(kbytesavail);
 494
 495 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 496                                char *buf)
 497 {
 498         struct obd_device *obd = container_of(kobj, struct obd_device,
 499                                               obd_kset.kobj);
 500         struct obd_statfs osfs;
 501         int rc;
 502
 503         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 504                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 505                         OBD_STATFS_NODELAY);
 506         if (!rc)
 507                 return sprintf(buf, "%llu\n", osfs.os_files);
 508
 509         return rc;
 510 }
 511 LUSTRE_RO_ATTR(filestotal);
 512
 513 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 514                               char *buf)
 515 {
 516         struct obd_device *obd = container_of(kobj, struct obd_device,
 517                                               obd_kset.kobj);
 518         struct obd_statfs osfs;
 519         int rc;
 520
 521         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 522                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 523                         OBD_STATFS_NODELAY);
 524         if (!rc)
 525                 return sprintf(buf, "%llu\n", osfs.os_ffree);
 526
 527         return rc;
 528 }
 529 LUSTRE_RO_ATTR(filesfree);
 530
 531 ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
 532 {
 533         struct obd_device *obd = container_of(kobj, struct obd_device,
 534                                               obd_kset.kobj);
 535         struct ptlrpc_connection *conn;
 536         ssize_t count;
 537
 538         LPROCFS_CLIMP_CHECK(obd);
 539         conn = obd->u.cli.cl_import->imp_connection;
 540         if (conn && obd->u.cli.cl_import)
 541                 count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
 542         else
 543                 count = sprintf(buf, "%s\n", "<none>");
 544
 545         LPROCFS_CLIMP_EXIT(obd);
 546         return count;
 547 }
 548 EXPORT_SYMBOL(conn_uuid_show);
 549
 550 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
 551 {
 552         struct obd_device *obd = data;
 553         struct obd_import *imp;
 554         char *imp_state_name = NULL;
 555         int rc = 0;
 556
 557         LASSERT(obd != NULL);
 558         LPROCFS_CLIMP_CHECK(obd);
 559         imp = obd->u.cli.cl_import;
 560         imp_state_name = ptlrpc_import_state_name(imp->imp_state);
 561         seq_printf(m, "%s\t%s%s\n", obd2cli_tgt(obd), imp_state_name,
 562                    imp->imp_deactive ? "\tDEACTIVATED" : "");
 563
 564         LPROCFS_CLIMP_EXIT(obd);
 565         return rc;
 566 }
 567 EXPORT_SYMBOL(lprocfs_server_uuid_seq_show);
 568
 569 int lprocfs_conn_uuid_seq_show(struct seq_file *m, void *data)
 570 {
 571         struct obd_device *obd = data;
 572         struct ptlrpc_connection *conn;
 573         int rc = 0;
 574
 575         LASSERT(obd != NULL);
 576
 577         LPROCFS_CLIMP_CHECK(obd);
 578         conn = obd->u.cli.cl_import->imp_connection;
 579         if (conn && obd->u.cli.cl_import)
 580                 seq_printf(m, "%s\n", conn->c_remote_uuid.uuid);
 581         else
 582                 seq_printf(m, "%s\n", "<none>");
 583
 584         LPROCFS_CLIMP_EXIT(obd);
 585         return rc;
 586 }
 587 EXPORT_SYMBOL(lprocfs_conn_uuid_seq_show);
 588
 589 /** add up per-cpu counters */
 590
 591 /**
 592  * Lock statistics structure for access, possibly only on this CPU.
 593  *
 594  * The statistics struct may be allocated with per-CPU structures for
 595  * efficient concurrent update (usually only on server-wide stats), or
 596  * as a single global struct (e.g. for per-client or per-job statistics),
 597  * so the required locking depends on the type of structure allocated.
 598  *
 599  * For per-CPU statistics, pin the thread to the current cpuid so that
 600  * will only access the statistics for that CPU.  If the stats structure
 601  * for the current CPU has not been allocated (or previously freed),
 602  * allocate it now.  The per-CPU statistics do not need locking since
 603  * the thread is pinned to the CPU during update.
 604  *
 605  * For global statistics, lock the stats structure to prevent concurrent update.
 606  *
 607  * \param[in] stats     statistics structure to lock
 608  * \param[in] opc       type of operation:
 609  *                      LPROCFS_GET_SMP_ID: "lock" and return current CPU index
 610  *                              for incrementing statistics for that CPU
 611  *                      LPROCFS_GET_NUM_CPU: "lock" and return number of used
 612  *                              CPU indices to iterate over all indices
 613  * \param[out] flags    CPU interrupt saved state for IRQ-safe locking
 614  *
 615  * \retval cpuid of current thread or number of allocated structs
 616  * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
 617  */
 618 int lprocfs_stats_lock(struct lprocfs_stats *stats,
 619                        enum lprocfs_stats_lock_ops opc,
 620                        unsigned long *flags)
 621 {
 622         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 623                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 624                         spin_lock_irqsave(&stats->ls_lock, *flags);
 625                 else
 626                         spin_lock(&stats->ls_lock);
 627                 return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
 628         }
 629
 630         switch (opc) {
 631         case LPROCFS_GET_SMP_ID: {
 632                 unsigned int cpuid = get_cpu();
 633
 634                 if (unlikely(!stats->ls_percpu[cpuid])) {
 635                         int rc = lprocfs_stats_alloc_one(stats, cpuid);
 636
 637                         if (rc < 0) {
 638                                 put_cpu();
 639                                 return rc;
 640                         }
 641                 }
 642                 return cpuid;
 643         }
 644         case LPROCFS_GET_NUM_CPU:
 645                 return stats->ls_biggest_alloc_num;
 646         default:
 647                 LBUG();
 648         }
 649 }
 650
 651 /**
 652  * Unlock statistics structure after access.
 653  *
 654  * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
 655  * or unpin this thread from the current cpuid for per-CPU statistics.
 656  *
 657  * This function must be called using the same arguments as used when calling
 658  * lprocfs_stats_lock() so that the correct operation can be performed.
 659  *
 660  * \param[in] stats     statistics structure to unlock
 661  * \param[in] opc       type of operation (current cpuid or number of structs)
 662  * \param[in] flags     CPU interrupt saved state for IRQ-safe locking
 663  */
 664 void lprocfs_stats_unlock(struct lprocfs_stats *stats,
 665                           enum lprocfs_stats_lock_ops opc,
 666                           unsigned long *flags)
 667 {
 668         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 669                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 670                         spin_unlock_irqrestore(&stats->ls_lock, *flags);
 671                 else
 672                         spin_unlock(&stats->ls_lock);
 673         } else if (opc == LPROCFS_GET_SMP_ID) {
 674                 put_cpu();
 675         }
 676 }
 677
 678 /** add up per-cpu counters */
 679 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
 680                            struct lprocfs_counter *cnt)
 681 {
 682         unsigned int                    num_entry;
 683         struct lprocfs_counter          *percpu_cntr;
 684         int                             i;
 685         unsigned long                   flags = 0;
 686
 687         memset(cnt, 0, sizeof(*cnt));
 688
 689         if (stats == NULL) {
 690                 /* set count to 1 to avoid divide-by-zero errs in callers */
 691                 cnt->lc_count = 1;
 692                 return;
 693         }
 694
 695         cnt->lc_min = LC_MIN_INIT;
 696
 697         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
 698
 699         for (i = 0; i < num_entry; i++) {
 700                 if (stats->ls_percpu[i] == NULL)
 701                         continue;
 702                 percpu_cntr = lprocfs_stats_counter_get(stats, i, idx);
 703
 704                 cnt->lc_count += percpu_cntr->lc_count;
 705                 cnt->lc_sum += percpu_cntr->lc_sum;
 706                 if (percpu_cntr->lc_min < cnt->lc_min)
 707                         cnt->lc_min = percpu_cntr->lc_min;
 708                 if (percpu_cntr->lc_max > cnt->lc_max)
 709                         cnt->lc_max = percpu_cntr->lc_max;
 710                 cnt->lc_sumsquare += percpu_cntr->lc_sumsquare;
 711         }
 712
 713         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
 714 }
 715
 716 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
 717 {
 718         bool first = true;
 719
 720         if (imp->imp_obd->obd_no_recov) {
 721                 seq_printf(m, "no_recov");
 722                 first = false;
 723         }
 724
 725         flag2str(imp, invalid);
 726         flag2str(imp, deactive);
 727         flag2str(imp, replayable);
 728         flag2str(imp, delayed_recovery);
 729         flag2str(imp, vbr_failed);
 730         flag2str(imp, pingable);
 731         flag2str(imp, resend_replay);
 732         flag2str(imp, no_pinger_recover);
 733         flag2str(imp, connect_tried);
 734 }
 735
 736 static const char *obd_connect_names[] = {
 737         /* flags names  */
 738         "read_only",
 739         "lov_index",
 740         "connect_from_mds",
 741         "write_grant",
 742         "server_lock",
 743         "version",
 744         "request_portal",
 745         "acl",
 746         "xattr",
 747         "create_on_write",
 748         "truncate_lock",
 749         "initial_transno",
 750         "inode_bit_locks",
 751         "barrier",
 752         "getattr_by_fid",
 753         "no_oh_for_devices",
 754         "remote_client",
 755         "remote_client_by_force",
 756         "max_byte_per_rpc",
 757         "64bit_qdata",
 758         "mds_capability",
 759         "oss_capability",
 760         "early_lock_cancel",
 761         "som",
 762         "adaptive_timeouts",
 763         "lru_resize",
 764         "mds_mds_connection",
 765         "real_conn",
 766         "change_qunit_size",
 767         "alt_checksum_algorithm",
 768         "fid_is_enabled",
 769         "version_recovery",
 770         "pools",
 771         "grant_shrink",
 772         "skip_orphan",
 773         "large_ea",
 774         "full20",
 775         "layout_lock",
 776         "64bithash",
 777         "object_max_bytes",
 778         "imp_recov",
 779         "jobstats",
 780         "umask",
 781         "einprogress",
 782         "grant_param",
 783         "flock_owner",
 784         "lvb_type",
 785         "nanoseconds_times",
 786         "lightweight_conn",
 787         "short_io",
 788         "pingless",
 789         "flock_deadlock",
 790         "disp_stripe",
 791         "open_by_fid",
 792         "lfsck",
 793         "unknown",
 794         "unlink_close",
 795         "multi_mod_rpcs",
 796         "dir_stripe",
 797         "subtree",
 798         "lockahead",
 799         "bulk_mbits",
 800         "compact_obdo",
 801         "second_flags",
 802         /* flags2 names */
 803         "file_secctx",
 804         "lockaheadv2",
 805         NULL
 806 };
 807
 808 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
 809                                const char *sep)
 810 {
 811         bool first = true;
 812         __u64 mask;
 813         int i;
 814
 815         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 816                 if (flags & mask) {
 817                         seq_printf(m, "%s%s",
 818                                    first ? "" : sep, obd_connect_names[i]);
 819                         first = false;
 820                 }
 821         }
 822
 823         if (flags & ~(mask - 1)) {
 824                 seq_printf(m, "%sunknown_%#llx",
 825                            first ? "" : sep, flags & ~(mask - 1));
 826                 first = false;
 827         }
 828
 829         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 830                 return;
 831
 832         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 833                 if (flags2 & mask) {
 834                         seq_printf(m, "%s%s",
 835                                    first ? "" : sep, obd_connect_names[i]);
 836                         first = false;
 837                 }
 838         }
 839
 840         if (flags2 & ~(mask - 1)) {
 841                 seq_printf(m, "%sunknown2_%#llx",
 842                            first ? "" : sep, flags2 & ~(mask - 1));
 843                 first = false;
 844         }
 845 }
 846 EXPORT_SYMBOL(obd_connect_seq_flags2str);
 847
 848 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 849                           const char *sep)
 850 {
 851         __u64 mask;
 852         int i, ret = 0;
 853
 854         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 855                 if (flags & mask)
 856                         ret += snprintf(page + ret, count - ret, "%s%s",
 857                                         ret ? sep : "", obd_connect_names[i]);
 858         }
 859
 860         if (flags & ~(mask - 1))
 861                 ret += snprintf(page + ret, count - ret,
 862                                 "%sunknown_%#llx",
 863                                 ret ? sep : "", flags & ~(mask - 1));
 864
 865         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 866                 return ret;
 867
 868         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 869                 if (flags2 & mask)
 870                         ret += snprintf(page + ret, count - ret, "%s%s",
 871                                         ret ? sep : "", obd_connect_names[i]);
 872         }
 873
 874         if (flags2 & ~(mask - 1))
 875                 ret += snprintf(page + ret, count - ret,
 876                                 "%sunknown2_%#llx",
 877                                 ret ? sep : "", flags2 & ~(mask - 1));
 878
 879         return ret;
 880 }
 881 EXPORT_SYMBOL(obd_connect_flags2str);
 882
 883 void
 884 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 885 {
 886         __u64 flags;
 887
 888         LASSERT(ocd != NULL);
 889         flags = ocd->ocd_connect_flags;
 890
 891         seq_printf(m, "    connect_data:\n"
 892                    "       flags: %#llx\n"
 893                    "       instance: %u\n",
 894                    ocd->ocd_connect_flags,
 895                    ocd->ocd_instance);
 896         if (flags & OBD_CONNECT_VERSION)
 897                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
 898                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
 899                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
 900                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
 901                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
 902         if (flags & OBD_CONNECT_MDS)
 903                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
 904         if (flags & OBD_CONNECT_GRANT)
 905                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
 906         if (flags & OBD_CONNECT_INDEX)
 907                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
 908         if (flags & OBD_CONNECT_BRW_SIZE)
 909                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
 910         if (flags & OBD_CONNECT_IBITS)
 911                 seq_printf(m, "       ibits_known: %#llx\n",
 912                            ocd->ocd_ibits_known);
 913         if (flags & OBD_CONNECT_GRANT_PARAM)
 914                 seq_printf(m, "       grant_block_size: %d\n"
 915                            "       grant_inode_size: %d\n"
 916                            "       grant_max_extent_size: %d\n"
 917                            "       grant_extent_tax: %d\n",
 918                            1 << ocd->ocd_grant_blkbits,
 919                            1 << ocd->ocd_grant_inobits,
 920                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
 921                            ocd->ocd_grant_tax_kb << 10);
 922         if (flags & OBD_CONNECT_TRANSNO)
 923                 seq_printf(m, "       first_transno: %#llx\n",
 924                            ocd->ocd_transno);
 925         if (flags & OBD_CONNECT_CKSUM)
 926                 seq_printf(m, "       cksum_types: %#x\n",
 927                            ocd->ocd_cksum_types);
 928         if (flags & OBD_CONNECT_MAX_EASIZE)
 929                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
 930         if (flags & OBD_CONNECT_MAXBYTES)
 931                 seq_printf(m, "       max_object_bytes: %llu\n",
 932                            ocd->ocd_maxbytes);
 933         if (flags & OBD_CONNECT_MULTIMODRPCS)
 934                 seq_printf(m, "       max_mod_rpcs: %hu\n",
 935                            ocd->ocd_maxmodrpcs);
 936 }
 937
 938 int lprocfs_import_seq_show(struct seq_file *m, void *data)
 939 {
 940         char                            nidstr[LNET_NIDSTR_SIZE];
 941         struct lprocfs_counter          ret;
 942         struct lprocfs_counter_header   *header;
 943         struct obd_device               *obd    = (struct obd_device *)data;
 944         struct obd_import               *imp;
 945         struct obd_import_conn          *conn;
 946         struct obd_connect_data         *ocd;
 947         int                             j;
 948         int                             k;
 949         int                             rw      = 0;
 950
 951         LASSERT(obd != NULL);
 952         LPROCFS_CLIMP_CHECK(obd);
 953         imp = obd->u.cli.cl_import;
 954         ocd = &imp->imp_connect_data;
 955
 956         seq_printf(m, "import:\n"
 957                    "    name: %s\n"
 958                    "    target: %s\n"
 959                    "    state: %s\n"
 960                    "    connect_flags: [ ",
 961                    obd->obd_name,
 962                    obd2cli_tgt(obd),
 963                    ptlrpc_import_state_name(imp->imp_state));
 964         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
 965                                   imp->imp_connect_data.ocd_connect_flags2,
 966                                   ", ");
 967         seq_printf(m, " ]\n");
 968         obd_connect_data_seqprint(m, ocd);
 969         seq_printf(m, "    import_flags: [ ");
 970         obd_import_flags2str(imp, m);
 971
 972         seq_printf(m, " ]\n"
 973                    "    connection:\n"
 974                    "       failover_nids: [ ");
 975         spin_lock(&imp->imp_lock);
 976         j = 0;
 977         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 978                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
 979                                  nidstr, sizeof(nidstr));
 980                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
 981                 j++;
 982         }
 983         if (imp->imp_connection != NULL)
 984                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
 985                                  nidstr, sizeof(nidstr));
 986         else
 987                 strncpy(nidstr, "<none>", sizeof(nidstr));
 988         seq_printf(m, " ]\n"
 989                    "       current_connection: %s\n"
 990                    "       connection_attempts: %u\n"
 991                    "       generation: %u\n"
 992                    "       in-progress_invalidations: %u\n",
 993                    nidstr,
 994                    imp->imp_conn_cnt,
 995                    imp->imp_generation,
 996                    atomic_read(&imp->imp_inval_count));
 997         spin_unlock(&imp->imp_lock);
 998
 999         if (obd->obd_svc_stats == NULL)
1000                 goto out_climp;
1001
1002         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
1003         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
1004         if (ret.lc_count != 0) {
1005                 /* first argument to do_div MUST be __u64 */
1006                 __u64 sum = ret.lc_sum;
1007                 do_div(sum, ret.lc_count);
1008                 ret.lc_sum = sum;
1009         } else
1010                 ret.lc_sum = 0;
1011         seq_printf(m, "    rpcs:\n"
1012                    "       inflight: %u\n"
1013                    "       unregistering: %u\n"
1014                    "       timeouts: %u\n"
1015                    "       avg_waittime: %llu %s\n",
1016                    atomic_read(&imp->imp_inflight),
1017                    atomic_read(&imp->imp_unregistering),
1018                    atomic_read(&imp->imp_timeouts),
1019                    ret.lc_sum, header->lc_units);
1020
1021         k = 0;
1022         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1023                 if (imp->imp_at.iat_portal[j] == 0)
1024                         break;
1025                 k = max_t(unsigned int, k,
1026                           at_get(&imp->imp_at.iat_service_estimate[j]));
1027         }
1028         seq_printf(m, "    service_estimates:\n"
1029                    "       services: %u sec\n"
1030                    "       network: %u sec\n",
1031                    k,
1032                    at_get(&imp->imp_at.iat_net_latency));
1033
1034         seq_printf(m, "    transactions:\n"
1035                    "       last_replay: %llu\n"
1036                    "       peer_committed: %llu\n"
1037                    "       last_checked: %llu\n",
1038                    imp->imp_last_replay_transno,
1039                    imp->imp_peer_committed_transno,
1040                    imp->imp_last_transno_checked);
1041
1042         /* avg data rates */
1043         for (rw = 0; rw <= 1; rw++) {
1044                 lprocfs_stats_collect(obd->obd_svc_stats,
1045                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1046                                       &ret);
1047                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1048                         /* first argument to do_div MUST be __u64 */
1049                         __u64 sum = ret.lc_sum;
1050                         do_div(sum, ret.lc_count);
1051                         ret.lc_sum = sum;
1052                         seq_printf(m, "    %s_data_averages:\n"
1053                                    "       bytes_per_rpc: %llu\n",
1054                                    rw ? "write" : "read",
1055                                    ret.lc_sum);
1056                 }
1057                 k = (int)ret.lc_sum;
1058                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1059                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1060                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1061                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1062                         /* first argument to do_div MUST be __u64 */
1063                         __u64 sum = ret.lc_sum;
1064                         do_div(sum, ret.lc_count);
1065                         ret.lc_sum = sum;
1066                         seq_printf(m, "       %s_per_rpc: %llu\n",
1067                                    header->lc_units, ret.lc_sum);
1068                         j = (int)ret.lc_sum;
1069                         if (j > 0)
1070                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1071                                            k / j, (100 * k / j) % 100);
1072                 }
1073         }
1074
1075 out_climp:
1076         LPROCFS_CLIMP_EXIT(obd);
1077         return 0;
1078 }
1079 EXPORT_SYMBOL(lprocfs_import_seq_show);
1080
1081 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1082 {
1083         struct obd_device *obd = (struct obd_device *)data;
1084         struct obd_import *imp;
1085         int j, k;
1086
1087         LASSERT(obd != NULL);
1088         LPROCFS_CLIMP_CHECK(obd);
1089         imp = obd->u.cli.cl_import;
1090
1091         seq_printf(m, "current_state: %s\n",
1092                    ptlrpc_import_state_name(imp->imp_state));
1093         seq_printf(m, "state_history:\n");
1094         k = imp->imp_state_hist_idx;
1095         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1096                 struct import_state_hist *ish =
1097                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1098                 if (ish->ish_state == 0)
1099                         continue;
1100                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1101                            ptlrpc_import_state_name(ish->ish_state));
1102         }
1103
1104         LPROCFS_CLIMP_EXIT(obd);
1105         return 0;
1106 }
1107 EXPORT_SYMBOL(lprocfs_state_seq_show);
1108
1109 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1110 {
1111         int i;
1112         for (i = 0; i < AT_BINS; i++)
1113                 seq_printf(m, "%3u ", at->at_hist[i]);
1114         seq_printf(m, "\n");
1115         return 0;
1116 }
1117 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1118
1119 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1120 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1121 {
1122         struct obd_device *obd = (struct obd_device *)data;
1123         struct obd_import *imp;
1124         unsigned int cur, worst;
1125         time64_t now, worstt;
1126         int i;
1127
1128         LASSERT(obd != NULL);
1129         LPROCFS_CLIMP_CHECK(obd);
1130         imp = obd->u.cli.cl_import;
1131
1132         now = ktime_get_real_seconds();
1133
1134         /* Some network health info for kicks */
1135         seq_printf(m, "%-10s : %lld, %llds ago\n",
1136                    "last reply", (s64)imp->imp_last_reply_time,
1137                    (s64)(now - imp->imp_last_reply_time));
1138
1139         cur = at_get(&imp->imp_at.iat_net_latency);
1140         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1141         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1142         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1143                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1144         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1145
1146         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1147                 if (imp->imp_at.iat_portal[i] == 0)
1148                         break;
1149                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1150                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1151                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1152                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1153                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1154                            (s64)(now - worstt));
1155                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1156         }
1157
1158         LPROCFS_CLIMP_EXIT(obd);
1159         return 0;
1160 }
1161 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1162
1163 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1164 {
1165         struct obd_device *obd = data;
1166         __u64 flags;
1167         __u64 flags2;
1168
1169         LPROCFS_CLIMP_CHECK(obd);
1170         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1171         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1172         seq_printf(m, "flags=%#llx\n", flags);
1173         seq_printf(m, "flags2=%#llx\n", flags2);
1174         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1175         seq_printf(m, "\n");
1176         LPROCFS_CLIMP_EXIT(obd);
1177         return 0;
1178 }
1179 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1180
1181 static struct attribute *obd_def_uuid_attrs[] = {
1182         &lustre_attr_uuid.attr,
1183         NULL,
1184 };
1185
1186 static struct attribute *obd_def_attrs[] = {
1187         &lustre_attr_blocksize.attr,
1188         &lustre_attr_kbytestotal.attr,
1189         &lustre_attr_kbytesfree.attr,
1190         &lustre_attr_kbytesavail.attr,
1191         &lustre_attr_filestotal.attr,
1192         &lustre_attr_filesfree.attr,
1193         &lustre_attr_uuid.attr,
1194         NULL,
1195 };
1196
1197 static void obd_sysfs_release(struct kobject *kobj)
1198 {
1199         struct obd_device *obd = container_of(kobj, struct obd_device,
1200                                               obd_kset.kobj);
1201
1202         complete(&obd->obd_kobj_unregister);
1203 }
1204
1205 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1206 {
1207         struct lprocfs_vars *debugfs_vars = NULL;
1208         int rc;
1209
1210         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1211                 return -ENODEV;
1212
1213         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1214         if (rc)
1215                 return rc;
1216
1217         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1218         obd->obd_ktype.release = obd_sysfs_release;
1219         if (obd->obd_attrs)
1220                 obd->obd_ktype.default_attrs = obd->obd_attrs;
1221
1222         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1223         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1224         init_completion(&obd->obd_kobj_unregister);
1225         rc = kset_register(&obd->obd_kset);
1226         if (rc)
1227                 return rc;
1228
1229         if (uuid_only)
1230                 obd->obd_attrs_group.attrs = obd_def_uuid_attrs;
1231         else
1232                 obd->obd_attrs_group.attrs = obd_def_attrs;
1233
1234         rc = sysfs_create_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
1235         if (rc) {
1236                 kset_unregister(&obd->obd_kset);
1237                 return rc;
1238         }
1239
1240         if (!obd->obd_type->typ_procroot)
1241                 debugfs_vars = obd->obd_vars;
1242         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1243                                                    obd->obd_type->typ_debugfs_entry,
1244                                                    debugfs_vars, obd);
1245         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1246                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1247                                             : -ENOMEM;
1248                 CERROR("error %d setting up debugfs for %s\n",
1249                        rc, obd->obd_name);
1250                 obd->obd_debugfs_entry = NULL;
1251                 lprocfs_obd_cleanup(obd);
1252                 return rc;
1253         }
1254
1255         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1256                 GOTO(already_registered, rc);
1257
1258         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1259                                                obd->obd_type->typ_procroot,
1260                                                obd->obd_vars, obd);
1261         if (IS_ERR(obd->obd_proc_entry)) {
1262                 rc = PTR_ERR(obd->obd_proc_entry);
1263                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1264                 obd->obd_proc_entry = NULL;
1265                 lprocfs_obd_cleanup(obd);
1266         }
1267 already_registered:
1268         return rc;
1269 }
1270 EXPORT_SYMBOL(lprocfs_obd_setup);
1271
1272 int lprocfs_obd_cleanup(struct obd_device *obd)
1273 {
1274         if (!obd)
1275                 return -EINVAL;
1276
1277         if (obd->obd_proc_exports_entry) {
1278                 /* Should be no exports left */
1279                 lprocfs_remove(&obd->obd_proc_exports_entry);
1280                 obd->obd_proc_exports_entry = NULL;
1281         }
1282
1283         if (obd->obd_proc_entry) {
1284                 lprocfs_remove(&obd->obd_proc_entry);
1285                 obd->obd_proc_entry = NULL;
1286         }
1287
1288         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1289                 ldebugfs_remove(&obd->obd_debugfs_entry);
1290
1291         sysfs_remove_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
1292         kset_unregister(&obd->obd_kset);
1293         wait_for_completion(&obd->obd_kobj_unregister);
1294
1295         return 0;
1296 }
1297 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1298
1299 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1300 {
1301         struct lprocfs_counter  *cntr;
1302         unsigned int            percpusize;
1303         int                     rc = -ENOMEM;
1304         unsigned long           flags = 0;
1305         int                     i;
1306
1307         LASSERT(stats->ls_percpu[cpuid] == NULL);
1308         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1309
1310         percpusize = lprocfs_stats_counter_size(stats);
1311         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1312         if (stats->ls_percpu[cpuid] != NULL) {
1313                 rc = 0;
1314                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1315                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1316                                 spin_lock_irqsave(&stats->ls_lock, flags);
1317                         else
1318                                 spin_lock(&stats->ls_lock);
1319                         if (stats->ls_biggest_alloc_num <= cpuid)
1320                                 stats->ls_biggest_alloc_num = cpuid + 1;
1321                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1322                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1323                         } else {
1324                                 spin_unlock(&stats->ls_lock);
1325                         }
1326                 }
1327                 /* initialize the ls_percpu[cpuid] non-zero counter */
1328                 for (i = 0; i < stats->ls_num; ++i) {
1329                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1330                         cntr->lc_min = LC_MIN_INIT;
1331                 }
1332         }
1333         return rc;
1334 }
1335
1336 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1337                                           enum lprocfs_stats_flags flags)
1338 {
1339         struct lprocfs_stats    *stats;
1340         unsigned int            num_entry;
1341         unsigned int            percpusize = 0;
1342         int                     i;
1343
1344         if (num == 0)
1345                 return NULL;
1346
1347         if (lprocfs_no_percpu_stats != 0)
1348                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1349
1350         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1351                 num_entry = 1;
1352         else
1353                 num_entry = num_possible_cpus();
1354
1355         /* alloc percpu pointers for all possible cpu slots */
1356         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1357         if (stats == NULL)
1358                 return NULL;
1359
1360         stats->ls_num = num;
1361         stats->ls_flags = flags;
1362         spin_lock_init(&stats->ls_lock);
1363
1364         /* alloc num of counter headers */
1365         LIBCFS_ALLOC(stats->ls_cnt_header,
1366                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1367         if (stats->ls_cnt_header == NULL)
1368                 goto fail;
1369
1370         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1371                 /* contains only one set counters */
1372                 percpusize = lprocfs_stats_counter_size(stats);
1373                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1374                 if (stats->ls_percpu[0] == NULL)
1375                         goto fail;
1376                 stats->ls_biggest_alloc_num = 1;
1377         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1378                 /* alloc all percpu data, currently only obd_memory use this */
1379                 for (i = 0; i < num_entry; ++i)
1380                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1381                                 goto fail;
1382         }
1383
1384         return stats;
1385
1386 fail:
1387         lprocfs_free_stats(&stats);
1388         return NULL;
1389 }
1390 EXPORT_SYMBOL(lprocfs_alloc_stats);
1391
1392 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1393 {
1394         struct lprocfs_stats *stats = *statsh;
1395         unsigned int num_entry;
1396         unsigned int percpusize;
1397         unsigned int i;
1398
1399         if (stats == NULL || stats->ls_num == 0)
1400                 return;
1401         *statsh = NULL;
1402
1403         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1404                 num_entry = 1;
1405         else
1406                 num_entry = num_possible_cpus();
1407
1408         percpusize = lprocfs_stats_counter_size(stats);
1409         for (i = 0; i < num_entry; i++)
1410                 if (stats->ls_percpu[i] != NULL)
1411                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1412         if (stats->ls_cnt_header != NULL)
1413                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1414                                         sizeof(struct lprocfs_counter_header));
1415         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1416 }
1417 EXPORT_SYMBOL(lprocfs_free_stats);
1418
1419 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1420                             enum lprocfs_fields_flags field)
1421 {
1422         unsigned long flags = 0;
1423         unsigned int num_cpu;
1424         unsigned int i;
1425         u64 ret = 0;
1426
1427         LASSERT(stats);
1428
1429         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1430         for (i = 0; i < num_cpu; i++) {
1431                 struct lprocfs_counter *cntr;
1432
1433                 if (!stats->ls_percpu[i])
1434                         continue;
1435
1436                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1437                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1438                                            stats->ls_flags, field);
1439         }
1440         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1441         return ret;
1442 }
1443 EXPORT_SYMBOL(lprocfs_stats_collector);
1444
1445 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1446 {
1447         struct lprocfs_counter          *percpu_cntr;
1448         int                             i;
1449         int                             j;
1450         unsigned int                    num_entry;
1451         unsigned long                   flags = 0;
1452
1453         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1454
1455         for (i = 0; i < num_entry; i++) {
1456                 if (stats->ls_percpu[i] == NULL)
1457                         continue;
1458                 for (j = 0; j < stats->ls_num; j++) {
1459                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1460                         percpu_cntr->lc_count           = 0;
1461                         percpu_cntr->lc_min             = LC_MIN_INIT;
1462                         percpu_cntr->lc_max             = 0;
1463                         percpu_cntr->lc_sumsquare       = 0;
1464                         percpu_cntr->lc_sum             = 0;
1465                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1466                                 percpu_cntr->lc_sum_irq = 0;
1467                 }
1468         }
1469
1470         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1471 }
1472 EXPORT_SYMBOL(lprocfs_clear_stats);
1473
1474 static ssize_t lprocfs_stats_seq_write(struct file *file,
1475                                        const char __user *buf,
1476                                        size_t len, loff_t *off)
1477 {
1478         struct seq_file *seq = file->private_data;
1479         struct lprocfs_stats *stats = seq->private;
1480
1481         lprocfs_clear_stats(stats);
1482
1483         return len;
1484 }
1485
1486 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1487 {
1488         struct lprocfs_stats *stats = p->private;
1489
1490         return (*pos < stats->ls_num) ? pos : NULL;
1491 }
1492
1493 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1494 {
1495 }
1496
1497 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1498 {
1499         (*pos)++;
1500
1501         return lprocfs_stats_seq_start(p, pos);
1502 }
1503
1504 /* seq file export of one lprocfs counter */
1505 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1506 {
1507         struct lprocfs_stats            *stats  = p->private;
1508         struct lprocfs_counter_header   *hdr;
1509         struct lprocfs_counter           ctr;
1510         int                              idx    = *(loff_t *)v;
1511
1512         if (idx == 0) {
1513                 struct timespec64 now;
1514
1515                 ktime_get_real_ts64(&now);
1516                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1517                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1518         }
1519
1520         hdr = &stats->ls_cnt_header[idx];
1521         lprocfs_stats_collect(stats, idx, &ctr);
1522
1523         if (ctr.lc_count == 0)
1524                 return 0;
1525
1526         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1527                    ctr.lc_count, hdr->lc_units);
1528
1529         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1530                 seq_printf(p, " %lld %lld %lld",
1531                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1532                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1533                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1534         }
1535         seq_putc(p, '\n');
1536         return 0;
1537 }
1538
1539 static const struct seq_operations lprocfs_stats_seq_sops = {
1540         .start  = lprocfs_stats_seq_start,
1541         .stop   = lprocfs_stats_seq_stop,
1542         .next   = lprocfs_stats_seq_next,
1543         .show   = lprocfs_stats_seq_show,
1544 };
1545
1546 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1547 {
1548         struct seq_file *seq;
1549         int rc;
1550
1551         rc = LPROCFS_ENTRY_CHECK(inode);
1552         if (rc < 0)
1553                 return rc;
1554
1555         rc = seq_open(file, &lprocfs_stats_seq_sops);
1556         if (rc)
1557                 return rc;
1558         seq = file->private_data;
1559         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1560         return 0;
1561 }
1562
1563 static const struct file_operations lprocfs_stats_seq_fops = {
1564         .owner   = THIS_MODULE,
1565         .open    = lprocfs_stats_seq_open,
1566         .read    = seq_read,
1567         .write   = lprocfs_stats_seq_write,
1568         .llseek  = seq_lseek,
1569         .release = lprocfs_seq_release,
1570 };
1571
1572 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1573                             struct lprocfs_stats *stats)
1574 {
1575         struct dentry *entry;
1576
1577         LASSERT(!IS_ERR_OR_NULL(parent));
1578
1579         entry = debugfs_create_file(name, 0644, parent, stats,
1580                                     &lprocfs_stats_seq_fops);
1581         if (IS_ERR_OR_NULL(entry))
1582                 return entry ? PTR_ERR(entry) : -ENOMEM;
1583
1584         return 0;
1585 }
1586 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1587
1588 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1589                            struct lprocfs_stats *stats)
1590 {
1591         struct proc_dir_entry *entry;
1592         LASSERT(root != NULL);
1593
1594         entry = proc_create_data(name, 0644, root,
1595                                  &lprocfs_stats_seq_fops, stats);
1596         if (entry == NULL)
1597                 return -ENOMEM;
1598         return 0;
1599 }
1600 EXPORT_SYMBOL(lprocfs_register_stats);
1601
1602 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1603                           unsigned conf, const char *name, const char *units)
1604 {
1605         struct lprocfs_counter_header   *header;
1606         struct lprocfs_counter          *percpu_cntr;
1607         unsigned long                   flags = 0;
1608         unsigned int                    i;
1609         unsigned int                    num_cpu;
1610
1611         LASSERT(stats != NULL);
1612
1613         header = &stats->ls_cnt_header[index];
1614         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1615                  index, name, units);
1616
1617         header->lc_config = conf;
1618         header->lc_name   = name;
1619         header->lc_units  = units;
1620
1621         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1622         for (i = 0; i < num_cpu; ++i) {
1623                 if (stats->ls_percpu[i] == NULL)
1624                         continue;
1625                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1626                 percpu_cntr->lc_count           = 0;
1627                 percpu_cntr->lc_min             = LC_MIN_INIT;
1628                 percpu_cntr->lc_max             = 0;
1629                 percpu_cntr->lc_sumsquare       = 0;
1630                 percpu_cntr->lc_sum             = 0;
1631                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1632                         percpu_cntr->lc_sum_irq = 0;
1633         }
1634         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1635 }
1636 EXPORT_SYMBOL(lprocfs_counter_init);
1637
1638 /* Note that we only init md counters for ops whose offset is less
1639  * than NUM_MD_STATS. This is explained in a comment in the definition
1640  * of struct md_ops. */
1641 #define LPROCFS_MD_OP_INIT(base, stats, op)                                    \
1642         do {                                                                   \
1643                 unsigned int _idx = base + MD_COUNTER_OFFSET(op);              \
1644                                                                                \
1645                 if (MD_COUNTER_OFFSET(op) < NUM_MD_STATS) {                    \
1646                         LASSERT(_idx < stats->ls_num);                         \
1647                         lprocfs_counter_init(stats, _idx, 0, #op, "reqs");     \
1648                 }                                                              \
1649         } while (0)
1650
1651 void lprocfs_init_mps_stats(int num_private_stats, struct lprocfs_stats *stats)
1652 {
1653         LPROCFS_MD_OP_INIT(num_private_stats, stats, get_root);
1654         LPROCFS_MD_OP_INIT(num_private_stats, stats, null_inode);
1655         LPROCFS_MD_OP_INIT(num_private_stats, stats, close);
1656         LPROCFS_MD_OP_INIT(num_private_stats, stats, create);
1657         LPROCFS_MD_OP_INIT(num_private_stats, stats, enqueue);
1658         LPROCFS_MD_OP_INIT(num_private_stats, stats, getattr);
1659         LPROCFS_MD_OP_INIT(num_private_stats, stats, getattr_name);
1660         LPROCFS_MD_OP_INIT(num_private_stats, stats, intent_lock);
1661         LPROCFS_MD_OP_INIT(num_private_stats, stats, link);
1662         LPROCFS_MD_OP_INIT(num_private_stats, stats, rename);
1663         LPROCFS_MD_OP_INIT(num_private_stats, stats, setattr);
1664         LPROCFS_MD_OP_INIT(num_private_stats, stats, fsync);
1665         LPROCFS_MD_OP_INIT(num_private_stats, stats, read_page);
1666         LPROCFS_MD_OP_INIT(num_private_stats, stats, unlink);
1667         LPROCFS_MD_OP_INIT(num_private_stats, stats, setxattr);
1668         LPROCFS_MD_OP_INIT(num_private_stats, stats, getxattr);
1669         LPROCFS_MD_OP_INIT(num_private_stats, stats, init_ea_size);
1670         LPROCFS_MD_OP_INIT(num_private_stats, stats, get_lustre_md);
1671         LPROCFS_MD_OP_INIT(num_private_stats, stats, free_lustre_md);
1672         LPROCFS_MD_OP_INIT(num_private_stats, stats, merge_attr);
1673         LPROCFS_MD_OP_INIT(num_private_stats, stats, set_open_replay_data);
1674         LPROCFS_MD_OP_INIT(num_private_stats, stats, clear_open_replay_data);
1675         LPROCFS_MD_OP_INIT(num_private_stats, stats, set_lock_data);
1676         LPROCFS_MD_OP_INIT(num_private_stats, stats, lock_match);
1677         LPROCFS_MD_OP_INIT(num_private_stats, stats, cancel_unused);
1678         LPROCFS_MD_OP_INIT(num_private_stats, stats, intent_getattr_async);
1679         LPROCFS_MD_OP_INIT(num_private_stats, stats, revalidate_lock);
1680 }
1681
1682 int lprocfs_alloc_md_stats(struct obd_device *obd,
1683                            unsigned int num_private_stats)
1684 {
1685         struct lprocfs_stats *stats;
1686         unsigned int num_stats;
1687         int rc, i;
1688
1689         CLASSERT(offsetof(struct md_ops, MD_STATS_FIRST_OP) == 0);
1690         CLASSERT(_MD_COUNTER_OFFSET(MD_STATS_FIRST_OP) == 0);
1691         CLASSERT(_MD_COUNTER_OFFSET(MD_STATS_LAST_OP) > 0);
1692
1693         /* TODO Ensure that this function is only used where
1694          * appropriate by adding an assertion to the effect that
1695          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1696          * because mdt_procfs_init() uses this function to allocate
1697          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1698          * mdt layer does not use the md_ops interface. This is
1699          * confusing and a waste of memory. See LU-2484.
1700          */
1701         LASSERT(obd->obd_proc_entry != NULL);
1702         LASSERT(obd->obd_md_stats == NULL);
1703         LASSERT(obd->obd_md_cntr_base == 0);
1704
1705         num_stats = NUM_MD_STATS + num_private_stats;
1706         stats = lprocfs_alloc_stats(num_stats, 0);
1707         if (stats == NULL)
1708                 return -ENOMEM;
1709
1710         lprocfs_init_mps_stats(num_private_stats, stats);
1711
1712         for (i = num_private_stats; i < num_stats; i++) {
1713                 if (stats->ls_cnt_header[i].lc_name == NULL) {
1714                         CERROR("Missing md_stat initializer md_op "
1715                                "operation at offset %d. Aborting.\n",
1716                                i - num_private_stats);
1717                         LBUG();
1718                 }
1719         }
1720
1721         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1722         if (rc < 0) {
1723                 lprocfs_free_stats(&stats);
1724         } else {
1725                 obd->obd_md_stats = stats;
1726                 obd->obd_md_cntr_base = num_private_stats;
1727         }
1728
1729         return rc;
1730 }
1731 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1732
1733 void lprocfs_free_md_stats(struct obd_device *obd)
1734 {
1735         struct lprocfs_stats *stats = obd->obd_md_stats;
1736
1737         if (stats != NULL) {
1738                 obd->obd_md_stats = NULL;
1739                 obd->obd_md_cntr_base = 0;
1740                 lprocfs_free_stats(&stats);
1741         }
1742 }
1743 EXPORT_SYMBOL(lprocfs_free_md_stats);
1744
1745 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1746 {
1747         lprocfs_counter_init(ldlm_stats,
1748                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1749                              0, "ldlm_enqueue", "reqs");
1750         lprocfs_counter_init(ldlm_stats,
1751                              LDLM_CONVERT - LDLM_FIRST_OPC,
1752                              0, "ldlm_convert", "reqs");
1753         lprocfs_counter_init(ldlm_stats,
1754                              LDLM_CANCEL - LDLM_FIRST_OPC,
1755                              0, "ldlm_cancel", "reqs");
1756         lprocfs_counter_init(ldlm_stats,
1757                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1758                              0, "ldlm_bl_callback", "reqs");
1759         lprocfs_counter_init(ldlm_stats,
1760                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1761                              0, "ldlm_cp_callback", "reqs");
1762         lprocfs_counter_init(ldlm_stats,
1763                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1764                              0, "ldlm_gl_callback", "reqs");
1765 }
1766 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1767
1768 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1769                           struct lprocfs_counter_header *header,
1770                           enum lprocfs_stats_flags flags,
1771                           enum lprocfs_fields_flags field)
1772 {
1773         __s64 ret = 0;
1774
1775         if (lc == NULL || header == NULL)
1776                 RETURN(0);
1777
1778         switch (field) {
1779                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1780                         ret = header->lc_config;
1781                         break;
1782                 case LPROCFS_FIELDS_FLAGS_SUM:
1783                         ret = lc->lc_sum;
1784                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1785                                 ret += lc->lc_sum_irq;
1786                         break;
1787                 case LPROCFS_FIELDS_FLAGS_MIN:
1788                         ret = lc->lc_min;
1789                         break;
1790                 case LPROCFS_FIELDS_FLAGS_MAX:
1791                         ret = lc->lc_max;
1792                         break;
1793                 case LPROCFS_FIELDS_FLAGS_AVG:
1794                         ret = (lc->lc_max - lc->lc_min) / 2;
1795                         break;
1796                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1797                         ret = lc->lc_sumsquare;
1798                         break;
1799                 case LPROCFS_FIELDS_FLAGS_COUNT:
1800                         ret = lc->lc_count;
1801                         break;
1802                 default:
1803                         break;
1804         };
1805         RETURN(ret);
1806 }
1807 EXPORT_SYMBOL(lprocfs_read_helper);
1808
1809 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1810                              int mult)
1811 {
1812         long decimal_val, frac_val;
1813         int prtn;
1814
1815         if (count < 10)
1816                 return -EINVAL;
1817
1818         decimal_val = val / mult;
1819         prtn = snprintf(buffer, count, "%ld", decimal_val);
1820         frac_val = val % mult;
1821
1822         if (prtn < (count - 4) && frac_val > 0) {
1823                 long temp_frac;
1824                 int i, temp_mult = 1, frac_bits = 0;
1825
1826                 temp_frac = frac_val * 10;
1827                 buffer[prtn++] = '.';
1828                 while (frac_bits < 2 && (temp_frac / mult) < 1 ) {
1829                         /* only reserved 2 bits fraction */
1830                         buffer[prtn++] ='0';
1831                         temp_frac *= 10;
1832                         frac_bits++;
1833                 }
1834                 /*
1835                  * Need to think these cases :
1836                  *      1. #echo x.00 > /proc/xxx       output result : x
1837                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1838                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1839                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1840                  *      Only reserved 2 bits fraction.
1841                  */
1842                 for (i = 0; i < (5 - prtn); i++)
1843                         temp_mult *= 10;
1844
1845                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1846                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1847                                  frac_val * temp_mult / mult);
1848
1849                 prtn--;
1850                 while(buffer[prtn] < '1' || buffer[prtn] > '9') {
1851                         prtn--;
1852                         if (buffer[prtn] == '.') {
1853                                 prtn--;
1854                                 break;
1855                         }
1856                 }
1857                 prtn++;
1858         }
1859         buffer[prtn++] ='\n';
1860         return prtn;
1861 }
1862 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1863
1864 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1865 {
1866         long decimal_val, frac_val;
1867
1868         decimal_val = val / mult;
1869         seq_printf(m, "%ld", decimal_val);
1870         frac_val = val % mult;
1871
1872         if (frac_val > 0) {
1873                 frac_val *= 100;
1874                 frac_val /= mult;
1875         }
1876         if (frac_val > 0) {
1877                 /* Three cases: x0, xx, 0x */
1878                 if ((frac_val % 10) != 0)
1879                         seq_printf(m, ".%ld", frac_val);
1880                 else
1881                         seq_printf(m, ".%ld", frac_val / 10);
1882         }
1883
1884         seq_printf(m, "\n");
1885         return 0;
1886 }
1887 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1888
1889 /* Obtains the conversion factor for the unit specified */
1890 static int get_mult(char unit, __u64 *mult)
1891 {
1892         __u64 units = 1;
1893
1894         switch (unit) {
1895         /* peta, tera, giga, mega, and kilo */
1896         case 'p':
1897         case 'P':
1898                 units <<= 10;
1899         case 't':
1900         case 'T':
1901                 units <<= 10;
1902         case 'g':
1903         case 'G':
1904                 units <<= 10;
1905         case 'm':
1906         case 'M':
1907                 units <<= 10;
1908         case 'k':
1909         case 'K':
1910                 units <<= 10;
1911                 break;
1912         /* some tests expect % to be accepted */
1913         case '%':
1914                 units = 1;
1915                 break;
1916         default:
1917                 return -EINVAL;
1918         }
1919
1920         *mult = units;
1921
1922         return 0;
1923 }
1924
1925 /*
1926  * Ensures the numeric string is valid. The function provides the final
1927  * multiplier in the case a unit exists at the end of the string. It also
1928  * locates the start of the whole and fractional parts (if any). This
1929  * function modifies the string so kstrtoull can be used to parse both
1930  * the whole and fraction portions. This function also figures out
1931  * the base of the number.
1932  */
1933 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1934                                   bool allow_units, char **whole, char **frac,
1935                                   unsigned int *base)
1936 {
1937         bool hit_decimal = false;
1938         bool hit_unit = false;
1939         int rc = 0;
1940         char *start;
1941         *mult = def_mult;
1942         *whole = NULL;
1943         *frac = NULL;
1944         *base = 10;
1945
1946         /* a hex string if it starts with "0x" */
1947         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1948                 *base = 16;
1949                 buffer += 2;
1950         }
1951
1952         start = buffer;
1953
1954         while (*buffer) {
1955                 /* allow for a single new line before the null terminator */
1956                 if (*buffer == '\n') {
1957                         *buffer = '\0';
1958                         buffer++;
1959
1960                         if (*buffer)
1961                                 return -EINVAL;
1962
1963                         break;
1964                 }
1965
1966                 /* any chars after our unit indicates a malformed string */
1967                 if (hit_unit)
1968                         return -EINVAL;
1969
1970                 /* ensure we only hit one decimal */
1971                 if (*buffer == '.') {
1972                         if (hit_decimal)
1973                                 return -EINVAL;
1974
1975                         /* if past start, there's a whole part */
1976                         if (start != buffer)
1977                                 *whole = start;
1978
1979                         *buffer = '\0';
1980                         start = buffer + 1;
1981                         hit_decimal = true;
1982                 } else if (!isdigit(*buffer) &&
1983                            !(*base == 16 && isxdigit(*buffer))) {
1984                         if (allow_units) {
1985                                 /* if we allow units, attempt to get mult */
1986                                 hit_unit = true;
1987                                 rc = get_mult(*buffer, mult);
1988                                 if (rc)
1989                                         return rc;
1990
1991                                 /* string stops here, but keep processing */
1992                                 *buffer = '\0';
1993                         } else {
1994                                 /* bad string */
1995                                 return -EINVAL;
1996                         }
1997                 }
1998
1999                 buffer++;
2000         }
2001
2002         if (hit_decimal) {
2003                 /* hit a decimal, make sure there's a fractional part */
2004                 if (!*start)
2005                         return -EINVAL;
2006
2007                 *frac = start;
2008         } else {
2009                 /* didn't hit a decimal, but may have a whole part */
2010                 if (start != buffer && *start)
2011                         *whole = start;
2012         }
2013
2014         /* malformed string if we didn't get anything */
2015         if (!*frac && !*whole)
2016                 return -EINVAL;
2017
2018         return 0;
2019 }
2020
2021 /*
2022  * Parses a numeric string which can contain a whole and fraction portion
2023  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
2024  * allows the string to have a unit at the end. The function handles
2025  * wrapping of the final unsigned value.
2026  */
2027 static int str_to_u64_parse(char *buffer, unsigned long count,
2028                             __u64 *val, __u64 def_mult, bool allow_units)
2029 {
2030         __u64 whole = 0;
2031         __u64 frac = 0;
2032         unsigned int frac_d = 1;
2033         __u64 wrap_indicator = ULLONG_MAX;
2034         int rc = 0;
2035         __u64 mult;
2036         char *strwhole;
2037         char *strfrac;
2038         unsigned int base = 10;
2039
2040         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2041                                     &strwhole, &strfrac, &base);
2042
2043         if (rc)
2044                 return rc;
2045
2046         if (mult == 0) {
2047                 *val = 0;
2048                 return 0;
2049         }
2050
2051         /* the multiplier limits how large the value can be */
2052         wrap_indicator /=  mult;
2053
2054         if (strwhole) {
2055                 rc = kstrtoull(strwhole, base, &whole);
2056                 if (rc)
2057                         return rc;
2058
2059                 if (whole > wrap_indicator)
2060                         return -ERANGE;
2061
2062                 whole *= mult;
2063         }
2064
2065         if (strfrac) {
2066                 if (strlen(strfrac) > 10)
2067                         strfrac[10] = '\0';
2068
2069                 rc = kstrtoull(strfrac, base, &frac);
2070                 if (rc)
2071                         return rc;
2072
2073                 /* determine power of fractional portion */
2074                 while (*strfrac) {
2075                         frac_d *= base;
2076                         strfrac++;
2077                 }
2078
2079                 /* fractional portion is too large to perform calculation */
2080                 if (frac > wrap_indicator)
2081                         return -ERANGE;
2082
2083                 frac *= mult;
2084                 do_div(frac, frac_d);
2085         }
2086
2087         /* check that the sum of whole and fraction fits in u64 */
2088         if (whole > (ULLONG_MAX - frac))
2089                 return -ERANGE;
2090
2091         *val = whole + frac;
2092
2093         return 0;
2094 }
2095
2096 /*
2097  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2098  * which will apply to the value parsed. It also can allow the string to
2099  * have a unit as the last character. The function handles overflow/underflow
2100  * of the signed integer.
2101  */
2102 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2103                                __s64 *val, __u64 def_mult, bool allow_units)
2104 {
2105         char kernbuf[22];
2106         __u64 tmp;
2107         unsigned int offset = 0;
2108         int signed sign = 1;
2109         __u64 max = LLONG_MAX;
2110         int rc = 0;
2111
2112         if (count > (sizeof(kernbuf) - 1))
2113                 return -EINVAL;
2114
2115         if (copy_from_user(kernbuf, buffer, count))
2116                 return -EFAULT;
2117
2118         kernbuf[count] = '\0';
2119
2120         /* keep track of our sign */
2121         if (*kernbuf == '-') {
2122                 sign = -1;
2123                 offset++;
2124                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2125                 max++;
2126         }
2127
2128         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2129                               &tmp, def_mult, allow_units);
2130         if (rc)
2131                 return rc;
2132
2133         /* check for overflow/underflow */
2134         if (max < tmp)
2135                 return -ERANGE;
2136
2137         *val = (__s64)tmp * sign;
2138
2139         return 0;
2140 }
2141
2142 /**
2143  * Convert a user string into a signed 64 bit number. This function produces
2144  * an error when the value parsed from the string times multiplier underflows or
2145  * overflows. This function only accepts strings that contains digits, an
2146  * optional decimal, and a char representing a unit at the end. If a unit is
2147  * specified in the string, the multiplier provided by the caller is ignored.
2148  * This function can also accept hexadecimal strings which are prefixed with
2149  * "0x".
2150  *
2151  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2152  * \param[in] count     buffer length
2153  * \param[in] val       if successful, the value represented by the string
2154  * \param[in] defunit   default unit if string doesn't contain one
2155  *
2156  * \retval              0 on success
2157  * \retval              negative number on error
2158  */
2159 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2160                                   unsigned long count, __s64 *val, char defunit)
2161 {
2162         __u64 mult = 1;
2163         int rc;
2164
2165         if (defunit != '1') {
2166                 rc = get_mult(defunit, &mult);
2167                 if (rc)
2168                         return rc;
2169         }
2170
2171         return str_to_s64_internal(buffer, count, val, mult, true);
2172 }
2173 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2174
2175 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2176 {
2177         size_t l2;
2178
2179         l2 = strlen(s2);
2180         if (!l2)
2181                 return (char *)s1;
2182         while (len >= l2) {
2183                 len--;
2184                 if (!memcmp(s1, s2, l2))
2185                         return (char *)s1;
2186                 s1++;
2187         }
2188         return NULL;
2189 }
2190 EXPORT_SYMBOL(lprocfs_strnstr);
2191
2192 /**
2193  * Find the string \a name in the input \a buffer, and return a pointer to the
2194  * value immediately following \a name, reducing \a count appropriately.
2195  * If \a name is not found the original \a buffer is returned.
2196  */
2197 char *lprocfs_find_named_value(const char *buffer, const char *name,
2198                                 size_t *count)
2199 {
2200         char *val;
2201         size_t buflen = *count;
2202
2203         /* there is no strnstr() in rhel5 and ubuntu kernels */
2204         val = lprocfs_strnstr(buffer, name, buflen);
2205         if (val == NULL)
2206                 return (char *)buffer;
2207
2208         val += strlen(name);                             /* skip prefix */
2209         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2210                 val++;
2211
2212         *count = 0;
2213         while (val < buffer + buflen && isalnum(*val)) {
2214                 ++*count;
2215                 ++val;
2216         }
2217
2218         return val - *count;
2219 }
2220 EXPORT_SYMBOL(lprocfs_find_named_value);
2221
2222 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2223                         const struct file_operations *seq_fops, void *data)
2224 {
2225         struct dentry *entry;
2226
2227         /* Disallow secretly (un)writable entries. */
2228         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2229
2230         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2231         if (IS_ERR_OR_NULL(entry))
2232                 return entry ? PTR_ERR(entry) : -ENOMEM;
2233
2234         return 0;
2235 }
2236 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2237
2238 int lprocfs_seq_create(struct proc_dir_entry *parent,
2239                        const char *name,
2240                        mode_t mode,
2241                        const struct file_operations *seq_fops,
2242                        void *data)
2243 {
2244         struct proc_dir_entry *entry;
2245         ENTRY;
2246
2247         /* Disallow secretly (un)writable entries. */
2248         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2249
2250         entry = proc_create_data(name, mode, parent, seq_fops, data);
2251
2252         if (entry == NULL)
2253                 RETURN(-ENOMEM);
2254
2255         RETURN(0);
2256 }
2257 EXPORT_SYMBOL(lprocfs_seq_create);
2258
2259 int lprocfs_obd_seq_create(struct obd_device *dev,
2260                            const char *name,
2261                            mode_t mode,
2262                            const struct file_operations *seq_fops,
2263                            void *data)
2264 {
2265         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2266                                    mode, seq_fops, data));
2267 }
2268 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2269
2270 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2271 {
2272         if (value >= OBD_HIST_MAX)
2273                 value = OBD_HIST_MAX - 1;
2274
2275         spin_lock(&oh->oh_lock);
2276         oh->oh_buckets[value]++;
2277         spin_unlock(&oh->oh_lock);
2278 }
2279 EXPORT_SYMBOL(lprocfs_oh_tally);
2280
2281 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2282 {
2283         unsigned int val = 0;
2284
2285         if (likely(value != 0))
2286                 val = min(fls(value - 1), OBD_HIST_MAX);
2287
2288         lprocfs_oh_tally(oh, val);
2289 }
2290 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2291
2292 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2293 {
2294         unsigned long ret = 0;
2295         int i;
2296
2297         for (i = 0; i < OBD_HIST_MAX; i++)
2298                 ret +=  oh->oh_buckets[i];
2299         return ret;
2300 }
2301 EXPORT_SYMBOL(lprocfs_oh_sum);
2302
2303 void lprocfs_oh_clear(struct obd_histogram *oh)
2304 {
2305         spin_lock(&oh->oh_lock);
2306         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2307         spin_unlock(&oh->oh_lock);
2308 }
2309 EXPORT_SYMBOL(lprocfs_oh_clear);
2310
2311 ssize_t lustre_attr_show(struct kobject *kobj,
2312                          struct attribute *attr, char *buf)
2313 {
2314         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2315
2316         return a->show ? a->show(kobj, attr, buf) : 0;
2317 }
2318 EXPORT_SYMBOL_GPL(lustre_attr_show);
2319
2320 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2321                           const char *buf, size_t len)
2322 {
2323         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2324
2325         return a->store ? a->store(kobj, attr, buf, len) : len;
2326 }
2327 EXPORT_SYMBOL_GPL(lustre_attr_store);
2328
2329 const struct sysfs_ops lustre_sysfs_ops = {
2330         .show  = lustre_attr_show,
2331         .store = lustre_attr_store,
2332 };
2333 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2334
2335 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2336 {
2337         struct obd_device *dev = data;
2338         struct client_obd *cli = &dev->u.cli;
2339
2340         spin_lock(&cli->cl_loi_list_lock);
2341         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2342         spin_unlock(&cli->cl_loi_list_lock);
2343         return 0;
2344 }
2345 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2346
2347 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2348                                                 const char __user *buffer,
2349                                                 size_t count, loff_t *off)
2350 {
2351         struct obd_device *dev =
2352                 ((struct seq_file *)file->private_data)->private;
2353         struct client_obd *cli = &dev->u.cli;
2354         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2355         int chunk_mask, rc;
2356         s64 val;
2357
2358         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2359         if (rc)
2360                 return rc;
2361         if (val < 0)
2362                 return -ERANGE;
2363
2364         /* if the max_pages is specified in bytes, convert to pages */
2365         if (val >= ONE_MB_BRW_SIZE)
2366                 val >>= PAGE_SHIFT;
2367
2368         LPROCFS_CLIMP_CHECK(dev);
2369
2370         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2371         /* max_pages_per_rpc must be chunk aligned */
2372         val = (val + ~chunk_mask) & chunk_mask;
2373         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2374                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2375                 LPROCFS_CLIMP_EXIT(dev);
2376                 return -ERANGE;
2377         }
2378         spin_lock(&cli->cl_loi_list_lock);
2379         cli->cl_max_pages_per_rpc = val;
2380         client_adjust_max_dirty(cli);
2381         spin_unlock(&cli->cl_loi_list_lock);
2382
2383         LPROCFS_CLIMP_EXIT(dev);
2384         return count;
2385 }
2386 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2387
2388 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2389                             char *buf)
2390 {
2391         struct obd_device *dev = container_of(kobj, struct obd_device,
2392                                               obd_kset.kobj);
2393         struct client_obd *cli = &dev->u.cli;
2394         int rc;
2395
2396         spin_lock(&cli->cl_loi_list_lock);
2397         rc = sprintf(buf, "%d\n", cli->cl_short_io_bytes);
2398         spin_unlock(&cli->cl_loi_list_lock);
2399         return rc;
2400 }
2401 EXPORT_SYMBOL(short_io_bytes_show);
2402
2403 /* Used to catch people who think they're specifying pages. */
2404 #define MIN_SHORT_IO_BYTES 64
2405
2406 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2407                              const char *buffer, size_t count)
2408 {
2409         struct obd_device *dev = container_of(kobj, struct obd_device,
2410                                               obd_kset.kobj);
2411         struct client_obd *cli = &dev->u.cli;
2412         u32 val;
2413         int rc;
2414
2415         LPROCFS_CLIMP_CHECK(dev);
2416
2417         rc = kstrtouint(buffer, 0, &val);
2418         if (rc)
2419                 GOTO(out, rc);
2420
2421         if (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES)
2422                 GOTO(out, rc = -ERANGE);
2423
2424         rc = count;
2425
2426         spin_lock(&cli->cl_loi_list_lock);
2427         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2428                 rc = -ERANGE;
2429         else
2430                 cli->cl_short_io_bytes = val;
2431         spin_unlock(&cli->cl_loi_list_lock);
2432
2433 out:
2434         LPROCFS_CLIMP_EXIT(dev);
2435         return rc;
2436 }
2437 EXPORT_SYMBOL(short_io_bytes_store);
2438
2439 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2440                            struct root_squash_info *squash, char *name)
2441 {
2442         int rc;
2443         char kernbuf[64], *tmp, *errmsg;
2444         unsigned long uid, gid;
2445         ENTRY;
2446
2447         if (count >= sizeof(kernbuf)) {
2448                 errmsg = "string too long";
2449                 GOTO(failed_noprint, rc = -EINVAL);
2450         }
2451         if (copy_from_user(kernbuf, buffer, count)) {
2452                 errmsg = "bad address";
2453                 GOTO(failed_noprint, rc = -EFAULT);
2454         }
2455         kernbuf[count] = '\0';
2456
2457         /* look for uid gid separator */
2458         tmp = strchr(kernbuf, ':');
2459         if (tmp == NULL) {
2460                 errmsg = "needs uid:gid format";
2461                 GOTO(failed, rc = -EINVAL);
2462         }
2463         *tmp = '\0';
2464         tmp++;
2465
2466         /* parse uid */
2467         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2468                 errmsg = "bad uid";
2469                 GOTO(failed, rc = -EINVAL);
2470         }
2471
2472         /* parse gid */
2473         if (kstrtoul(tmp, 0, &gid) != 0) {
2474                 errmsg = "bad gid";
2475                 GOTO(failed, rc = -EINVAL);
2476         }
2477
2478         squash->rsi_uid = uid;
2479         squash->rsi_gid = gid;
2480
2481         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2482                       name, squash->rsi_uid, squash->rsi_gid);
2483         RETURN(count);
2484
2485 failed:
2486         if (tmp != NULL) {
2487                 tmp--;
2488                 *tmp = ':';
2489         }
2490         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2491               name, kernbuf, errmsg, rc);
2492         RETURN(rc);
2493 failed_noprint:
2494         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2495               name, errmsg, rc);
2496         RETURN(rc);
2497 }
2498 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2499
2500
2501 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2502                              struct root_squash_info *squash, char *name)
2503 {
2504         int rc;
2505         char *kernbuf = NULL;
2506         char *errmsg;
2507         struct list_head tmp;
2508         int len = count;
2509         ENTRY;
2510
2511         if (count > 4096) {
2512                 errmsg = "string too long";
2513                 GOTO(failed, rc = -EINVAL);
2514         }
2515
2516         OBD_ALLOC(kernbuf, count + 1);
2517         if (kernbuf == NULL) {
2518                 errmsg = "no memory";
2519                 GOTO(failed, rc = -ENOMEM);
2520         }
2521         if (copy_from_user(kernbuf, buffer, count)) {
2522                 errmsg = "bad address";
2523                 GOTO(failed, rc = -EFAULT);
2524         }
2525         kernbuf[count] = '\0';
2526
2527         if (count > 0 && kernbuf[count - 1] == '\n')
2528                 len = count - 1;
2529
2530         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2531             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2532                 /* empty string is special case */
2533                 down_write(&squash->rsi_sem);
2534                 if (!list_empty(&squash->rsi_nosquash_nids))
2535                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2536                 up_write(&squash->rsi_sem);
2537                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2538                 OBD_FREE(kernbuf, count + 1);
2539                 RETURN(count);
2540         }
2541
2542         INIT_LIST_HEAD(&tmp);
2543         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2544                 errmsg = "can't parse";
2545                 GOTO(failed, rc = -EINVAL);
2546         }
2547         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2548                       name, kernbuf);
2549         OBD_FREE(kernbuf, count + 1);
2550         kernbuf = NULL;
2551
2552         down_write(&squash->rsi_sem);
2553         if (!list_empty(&squash->rsi_nosquash_nids))
2554                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2555         list_splice(&tmp, &squash->rsi_nosquash_nids);
2556         up_write(&squash->rsi_sem);
2557
2558         RETURN(count);
2559
2560 failed:
2561         if (kernbuf) {
2562                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2563                       name, kernbuf, errmsg, rc);
2564                 OBD_FREE(kernbuf, count + 1);
2565         } else {
2566                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2567                       name, errmsg, rc);
2568         }
2569         RETURN(rc);
2570 }
2571 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2572
2573 #endif /* CONFIG_PROC_FS*/