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",  /* 0x01 */
 804         "lockaheadv2",  /* 0x02 */
 805         "dir_migrate",  /* 0x04 */
 806         "unknown",      /* 0x08 */
 807         "unknown",      /* 0x10 */
 808         "flr",          /* 0x20 */
 809         "wbc",          /* 0x40 */
 810         "lock_convert",  /* 0x80 */
 811         NULL
 812 };
 813
 814 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
 815                                const char *sep)
 816 {
 817         bool first = true;
 818         __u64 mask;
 819         int i;
 820
 821         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 822                 if (flags & mask) {
 823                         seq_printf(m, "%s%s",
 824                                    first ? "" : sep, obd_connect_names[i]);
 825                         first = false;
 826                 }
 827         }
 828
 829         if (flags & ~(mask - 1)) {
 830                 seq_printf(m, "%sunknown_%#llx",
 831                            first ? "" : sep, flags & ~(mask - 1));
 832                 first = false;
 833         }
 834
 835         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 836                 return;
 837
 838         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 839                 if (flags2 & mask) {
 840                         seq_printf(m, "%s%s",
 841                                    first ? "" : sep, obd_connect_names[i]);
 842                         first = false;
 843                 }
 844         }
 845
 846         if (flags2 & ~(mask - 1)) {
 847                 seq_printf(m, "%sunknown2_%#llx",
 848                            first ? "" : sep, flags2 & ~(mask - 1));
 849                 first = false;
 850         }
 851 }
 852 EXPORT_SYMBOL(obd_connect_seq_flags2str);
 853
 854 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 855                           const char *sep)
 856 {
 857         __u64 mask;
 858         int i, ret = 0;
 859
 860         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 861                 if (flags & mask)
 862                         ret += snprintf(page + ret, count - ret, "%s%s",
 863                                         ret ? sep : "", obd_connect_names[i]);
 864         }
 865
 866         if (flags & ~(mask - 1))
 867                 ret += snprintf(page + ret, count - ret,
 868                                 "%sunknown_%#llx",
 869                                 ret ? sep : "", flags & ~(mask - 1));
 870
 871         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 872                 return ret;
 873
 874         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 875                 if (flags2 & mask)
 876                         ret += snprintf(page + ret, count - ret, "%s%s",
 877                                         ret ? sep : "", obd_connect_names[i]);
 878         }
 879
 880         if (flags2 & ~(mask - 1))
 881                 ret += snprintf(page + ret, count - ret,
 882                                 "%sunknown2_%#llx",
 883                                 ret ? sep : "", flags2 & ~(mask - 1));
 884
 885         return ret;
 886 }
 887 EXPORT_SYMBOL(obd_connect_flags2str);
 888
 889 void
 890 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 891 {
 892         __u64 flags;
 893
 894         LASSERT(ocd != NULL);
 895         flags = ocd->ocd_connect_flags;
 896
 897         seq_printf(m, "    connect_data:\n"
 898                    "       flags: %#llx\n"
 899                    "       instance: %u\n",
 900                    ocd->ocd_connect_flags,
 901                    ocd->ocd_instance);
 902         if (flags & OBD_CONNECT_VERSION)
 903                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
 904                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
 905                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
 906                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
 907                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
 908         if (flags & OBD_CONNECT_MDS)
 909                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
 910         if (flags & OBD_CONNECT_GRANT)
 911                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
 912         if (flags & OBD_CONNECT_INDEX)
 913                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
 914         if (flags & OBD_CONNECT_BRW_SIZE)
 915                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
 916         if (flags & OBD_CONNECT_IBITS)
 917                 seq_printf(m, "       ibits_known: %#llx\n",
 918                            ocd->ocd_ibits_known);
 919         if (flags & OBD_CONNECT_GRANT_PARAM)
 920                 seq_printf(m, "       grant_block_size: %d\n"
 921                            "       grant_inode_size: %d\n"
 922                            "       grant_max_extent_size: %d\n"
 923                            "       grant_extent_tax: %d\n",
 924                            1 << ocd->ocd_grant_blkbits,
 925                            1 << ocd->ocd_grant_inobits,
 926                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
 927                            ocd->ocd_grant_tax_kb << 10);
 928         if (flags & OBD_CONNECT_TRANSNO)
 929                 seq_printf(m, "       first_transno: %#llx\n",
 930                            ocd->ocd_transno);
 931         if (flags & OBD_CONNECT_CKSUM)
 932                 seq_printf(m, "       cksum_types: %#x\n",
 933                            ocd->ocd_cksum_types);
 934         if (flags & OBD_CONNECT_MAX_EASIZE)
 935                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
 936         if (flags & OBD_CONNECT_MAXBYTES)
 937                 seq_printf(m, "       max_object_bytes: %llu\n",
 938                            ocd->ocd_maxbytes);
 939         if (flags & OBD_CONNECT_MULTIMODRPCS)
 940                 seq_printf(m, "       max_mod_rpcs: %hu\n",
 941                            ocd->ocd_maxmodrpcs);
 942 }
 943
 944 int lprocfs_import_seq_show(struct seq_file *m, void *data)
 945 {
 946         char                            nidstr[LNET_NIDSTR_SIZE];
 947         struct lprocfs_counter          ret;
 948         struct lprocfs_counter_header   *header;
 949         struct obd_device               *obd    = (struct obd_device *)data;
 950         struct obd_import               *imp;
 951         struct obd_import_conn          *conn;
 952         struct obd_connect_data         *ocd;
 953         int                             j;
 954         int                             k;
 955         int                             rw      = 0;
 956
 957         LASSERT(obd != NULL);
 958         LPROCFS_CLIMP_CHECK(obd);
 959         imp = obd->u.cli.cl_import;
 960         ocd = &imp->imp_connect_data;
 961
 962         seq_printf(m, "import:\n"
 963                    "    name: %s\n"
 964                    "    target: %s\n"
 965                    "    state: %s\n"
 966                    "    connect_flags: [ ",
 967                    obd->obd_name,
 968                    obd2cli_tgt(obd),
 969                    ptlrpc_import_state_name(imp->imp_state));
 970         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
 971                                   imp->imp_connect_data.ocd_connect_flags2,
 972                                   ", ");
 973         seq_printf(m, " ]\n");
 974         obd_connect_data_seqprint(m, ocd);
 975         seq_printf(m, "    import_flags: [ ");
 976         obd_import_flags2str(imp, m);
 977
 978         seq_printf(m, " ]\n"
 979                    "    connection:\n"
 980                    "       failover_nids: [ ");
 981         spin_lock(&imp->imp_lock);
 982         j = 0;
 983         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 984                 libcfs_nid2str_r(conn->oic_conn->c_peer.nid,
 985                                  nidstr, sizeof(nidstr));
 986                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
 987                 j++;
 988         }
 989         if (imp->imp_connection != NULL)
 990                 libcfs_nid2str_r(imp->imp_connection->c_peer.nid,
 991                                  nidstr, sizeof(nidstr));
 992         else
 993                 strncpy(nidstr, "<none>", sizeof(nidstr));
 994         seq_printf(m, " ]\n"
 995                    "       current_connection: %s\n"
 996                    "       connection_attempts: %u\n"
 997                    "       generation: %u\n"
 998                    "       in-progress_invalidations: %u\n",
 999                    nidstr,
1000                    imp->imp_conn_cnt,
1001                    imp->imp_generation,
1002                    atomic_read(&imp->imp_inval_count));
1003         spin_unlock(&imp->imp_lock);
1004
1005         if (obd->obd_svc_stats == NULL)
1006                 goto out_climp;
1007
1008         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
1009         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
1010         if (ret.lc_count != 0) {
1011                 /* first argument to do_div MUST be __u64 */
1012                 __u64 sum = ret.lc_sum;
1013                 do_div(sum, ret.lc_count);
1014                 ret.lc_sum = sum;
1015         } else
1016                 ret.lc_sum = 0;
1017         seq_printf(m, "    rpcs:\n"
1018                    "       inflight: %u\n"
1019                    "       unregistering: %u\n"
1020                    "       timeouts: %u\n"
1021                    "       avg_waittime: %llu %s\n",
1022                    atomic_read(&imp->imp_inflight),
1023                    atomic_read(&imp->imp_unregistering),
1024                    atomic_read(&imp->imp_timeouts),
1025                    ret.lc_sum, header->lc_units);
1026
1027         k = 0;
1028         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
1029                 if (imp->imp_at.iat_portal[j] == 0)
1030                         break;
1031                 k = max_t(unsigned int, k,
1032                           at_get(&imp->imp_at.iat_service_estimate[j]));
1033         }
1034         seq_printf(m, "    service_estimates:\n"
1035                    "       services: %u sec\n"
1036                    "       network: %u sec\n",
1037                    k,
1038                    at_get(&imp->imp_at.iat_net_latency));
1039
1040         seq_printf(m, "    transactions:\n"
1041                    "       last_replay: %llu\n"
1042                    "       peer_committed: %llu\n"
1043                    "       last_checked: %llu\n",
1044                    imp->imp_last_replay_transno,
1045                    imp->imp_peer_committed_transno,
1046                    imp->imp_last_transno_checked);
1047
1048         /* avg data rates */
1049         for (rw = 0; rw <= 1; rw++) {
1050                 lprocfs_stats_collect(obd->obd_svc_stats,
1051                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
1052                                       &ret);
1053                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
1054                         /* first argument to do_div MUST be __u64 */
1055                         __u64 sum = ret.lc_sum;
1056                         do_div(sum, ret.lc_count);
1057                         ret.lc_sum = sum;
1058                         seq_printf(m, "    %s_data_averages:\n"
1059                                    "       bytes_per_rpc: %llu\n",
1060                                    rw ? "write" : "read",
1061                                    ret.lc_sum);
1062                 }
1063                 k = (int)ret.lc_sum;
1064                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
1065                 header = &obd->obd_svc_stats->ls_cnt_header[j];
1066                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
1067                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
1068                         /* first argument to do_div MUST be __u64 */
1069                         __u64 sum = ret.lc_sum;
1070                         do_div(sum, ret.lc_count);
1071                         ret.lc_sum = sum;
1072                         seq_printf(m, "       %s_per_rpc: %llu\n",
1073                                    header->lc_units, ret.lc_sum);
1074                         j = (int)ret.lc_sum;
1075                         if (j > 0)
1076                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
1077                                            k / j, (100 * k / j) % 100);
1078                 }
1079         }
1080
1081 out_climp:
1082         LPROCFS_CLIMP_EXIT(obd);
1083         return 0;
1084 }
1085 EXPORT_SYMBOL(lprocfs_import_seq_show);
1086
1087 int lprocfs_state_seq_show(struct seq_file *m, void *data)
1088 {
1089         struct obd_device *obd = (struct obd_device *)data;
1090         struct obd_import *imp;
1091         int j, k;
1092
1093         LASSERT(obd != NULL);
1094         LPROCFS_CLIMP_CHECK(obd);
1095         imp = obd->u.cli.cl_import;
1096
1097         seq_printf(m, "current_state: %s\n",
1098                    ptlrpc_import_state_name(imp->imp_state));
1099         seq_printf(m, "state_history:\n");
1100         k = imp->imp_state_hist_idx;
1101         for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
1102                 struct import_state_hist *ish =
1103                         &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
1104                 if (ish->ish_state == 0)
1105                         continue;
1106                 seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
1107                            ptlrpc_import_state_name(ish->ish_state));
1108         }
1109
1110         LPROCFS_CLIMP_EXIT(obd);
1111         return 0;
1112 }
1113 EXPORT_SYMBOL(lprocfs_state_seq_show);
1114
1115 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
1116 {
1117         int i;
1118         for (i = 0; i < AT_BINS; i++)
1119                 seq_printf(m, "%3u ", at->at_hist[i]);
1120         seq_printf(m, "\n");
1121         return 0;
1122 }
1123 EXPORT_SYMBOL(lprocfs_at_hist_helper);
1124
1125 /* See also ptlrpc_lprocfs_timeouts_show_seq */
1126 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1127 {
1128         struct obd_device *obd = (struct obd_device *)data;
1129         struct obd_import *imp;
1130         unsigned int cur, worst;
1131         time64_t now, worstt;
1132         int i;
1133
1134         LASSERT(obd != NULL);
1135         LPROCFS_CLIMP_CHECK(obd);
1136         imp = obd->u.cli.cl_import;
1137
1138         now = ktime_get_real_seconds();
1139
1140         /* Some network health info for kicks */
1141         seq_printf(m, "%-10s : %lld, %llds ago\n",
1142                    "last reply", (s64)imp->imp_last_reply_time,
1143                    (s64)(now - imp->imp_last_reply_time));
1144
1145         cur = at_get(&imp->imp_at.iat_net_latency);
1146         worst = imp->imp_at.iat_net_latency.at_worst_ever;
1147         worstt = imp->imp_at.iat_net_latency.at_worst_time;
1148         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
1149                    "network", cur, worst, (s64)worstt, (s64)(now - worstt));
1150         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
1151
1152         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
1153                 if (imp->imp_at.iat_portal[i] == 0)
1154                         break;
1155                 cur = at_get(&imp->imp_at.iat_service_estimate[i]);
1156                 worst = imp->imp_at.iat_service_estimate[i].at_worst_ever;
1157                 worstt = imp->imp_at.iat_service_estimate[i].at_worst_time;
1158                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1159                            imp->imp_at.iat_portal[i], cur, worst, (s64)worstt,
1160                            (s64)(now - worstt));
1161                 lprocfs_at_hist_helper(m, &imp->imp_at.iat_service_estimate[i]);
1162         }
1163
1164         LPROCFS_CLIMP_EXIT(obd);
1165         return 0;
1166 }
1167 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1168
1169 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1170 {
1171         struct obd_device *obd = data;
1172         __u64 flags;
1173         __u64 flags2;
1174
1175         LPROCFS_CLIMP_CHECK(obd);
1176         flags = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags;
1177         flags2 = obd->u.cli.cl_import->imp_connect_data.ocd_connect_flags2;
1178         seq_printf(m, "flags=%#llx\n", flags);
1179         seq_printf(m, "flags2=%#llx\n", flags2);
1180         obd_connect_seq_flags2str(m, flags, flags2, "\n");
1181         seq_printf(m, "\n");
1182         LPROCFS_CLIMP_EXIT(obd);
1183         return 0;
1184 }
1185 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1186
1187 static struct attribute *obd_def_uuid_attrs[] = {
1188         &lustre_attr_uuid.attr,
1189         NULL,
1190 };
1191
1192 static struct attribute *obd_def_attrs[] = {
1193         &lustre_attr_blocksize.attr,
1194         &lustre_attr_kbytestotal.attr,
1195         &lustre_attr_kbytesfree.attr,
1196         &lustre_attr_kbytesavail.attr,
1197         &lustre_attr_filestotal.attr,
1198         &lustre_attr_filesfree.attr,
1199         &lustre_attr_uuid.attr,
1200         NULL,
1201 };
1202
1203 static void obd_sysfs_release(struct kobject *kobj)
1204 {
1205         struct obd_device *obd = container_of(kobj, struct obd_device,
1206                                               obd_kset.kobj);
1207
1208         complete(&obd->obd_kobj_unregister);
1209 }
1210
1211 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1212 {
1213         struct lprocfs_vars *debugfs_vars = NULL;
1214         int rc;
1215
1216         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1217                 return -ENODEV;
1218
1219         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1220         if (rc)
1221                 return rc;
1222
1223         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1224         obd->obd_ktype.release = obd_sysfs_release;
1225         if (obd->obd_attrs)
1226                 obd->obd_ktype.default_attrs = obd->obd_attrs;
1227
1228         obd->obd_kset.kobj.parent = obd->obd_type->typ_kobj;
1229         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1230         init_completion(&obd->obd_kobj_unregister);
1231         rc = kset_register(&obd->obd_kset);
1232         if (rc)
1233                 return rc;
1234
1235         if (uuid_only)
1236                 obd->obd_attrs_group.attrs = obd_def_uuid_attrs;
1237         else
1238                 obd->obd_attrs_group.attrs = obd_def_attrs;
1239
1240         rc = sysfs_create_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
1241         if (rc) {
1242                 kset_unregister(&obd->obd_kset);
1243                 return rc;
1244         }
1245
1246         if (!obd->obd_type->typ_procroot)
1247                 debugfs_vars = obd->obd_vars;
1248         obd->obd_debugfs_entry = ldebugfs_register(obd->obd_name,
1249                                                    obd->obd_type->typ_debugfs_entry,
1250                                                    debugfs_vars, obd);
1251         if (IS_ERR_OR_NULL(obd->obd_debugfs_entry)) {
1252                 rc = obd->obd_debugfs_entry ? PTR_ERR(obd->obd_debugfs_entry)
1253                                             : -ENOMEM;
1254                 CERROR("error %d setting up debugfs for %s\n",
1255                        rc, obd->obd_name);
1256                 obd->obd_debugfs_entry = NULL;
1257                 lprocfs_obd_cleanup(obd);
1258                 return rc;
1259         }
1260
1261         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1262                 GOTO(already_registered, rc);
1263
1264         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1265                                                obd->obd_type->typ_procroot,
1266                                                obd->obd_vars, obd);
1267         if (IS_ERR(obd->obd_proc_entry)) {
1268                 rc = PTR_ERR(obd->obd_proc_entry);
1269                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1270                 obd->obd_proc_entry = NULL;
1271                 lprocfs_obd_cleanup(obd);
1272         }
1273 already_registered:
1274         return rc;
1275 }
1276 EXPORT_SYMBOL(lprocfs_obd_setup);
1277
1278 int lprocfs_obd_cleanup(struct obd_device *obd)
1279 {
1280         if (!obd)
1281                 return -EINVAL;
1282
1283         if (obd->obd_proc_exports_entry) {
1284                 /* Should be no exports left */
1285                 lprocfs_remove(&obd->obd_proc_exports_entry);
1286                 obd->obd_proc_exports_entry = NULL;
1287         }
1288
1289         if (obd->obd_proc_entry) {
1290                 lprocfs_remove(&obd->obd_proc_entry);
1291                 obd->obd_proc_entry = NULL;
1292         }
1293
1294         if (!IS_ERR_OR_NULL(obd->obd_debugfs_entry))
1295                 ldebugfs_remove(&obd->obd_debugfs_entry);
1296
1297         sysfs_remove_group(&obd->obd_kset.kobj, &obd->obd_attrs_group);
1298         kset_unregister(&obd->obd_kset);
1299         wait_for_completion(&obd->obd_kobj_unregister);
1300
1301         return 0;
1302 }
1303 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1304
1305 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1306 {
1307         struct lprocfs_counter  *cntr;
1308         unsigned int            percpusize;
1309         int                     rc = -ENOMEM;
1310         unsigned long           flags = 0;
1311         int                     i;
1312
1313         LASSERT(stats->ls_percpu[cpuid] == NULL);
1314         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1315
1316         percpusize = lprocfs_stats_counter_size(stats);
1317         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1318         if (stats->ls_percpu[cpuid] != NULL) {
1319                 rc = 0;
1320                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1321                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1322                                 spin_lock_irqsave(&stats->ls_lock, flags);
1323                         else
1324                                 spin_lock(&stats->ls_lock);
1325                         if (stats->ls_biggest_alloc_num <= cpuid)
1326                                 stats->ls_biggest_alloc_num = cpuid + 1;
1327                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1328                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1329                         } else {
1330                                 spin_unlock(&stats->ls_lock);
1331                         }
1332                 }
1333                 /* initialize the ls_percpu[cpuid] non-zero counter */
1334                 for (i = 0; i < stats->ls_num; ++i) {
1335                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1336                         cntr->lc_min = LC_MIN_INIT;
1337                 }
1338         }
1339         return rc;
1340 }
1341
1342 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1343                                           enum lprocfs_stats_flags flags)
1344 {
1345         struct lprocfs_stats    *stats;
1346         unsigned int            num_entry;
1347         unsigned int            percpusize = 0;
1348         int                     i;
1349
1350         if (num == 0)
1351                 return NULL;
1352
1353         if (lprocfs_no_percpu_stats != 0)
1354                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1355
1356         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1357                 num_entry = 1;
1358         else
1359                 num_entry = num_possible_cpus();
1360
1361         /* alloc percpu pointers for all possible cpu slots */
1362         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1363         if (stats == NULL)
1364                 return NULL;
1365
1366         stats->ls_num = num;
1367         stats->ls_flags = flags;
1368         spin_lock_init(&stats->ls_lock);
1369
1370         /* alloc num of counter headers */
1371         LIBCFS_ALLOC(stats->ls_cnt_header,
1372                      stats->ls_num * sizeof(struct lprocfs_counter_header));
1373         if (stats->ls_cnt_header == NULL)
1374                 goto fail;
1375
1376         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1377                 /* contains only one set counters */
1378                 percpusize = lprocfs_stats_counter_size(stats);
1379                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1380                 if (stats->ls_percpu[0] == NULL)
1381                         goto fail;
1382                 stats->ls_biggest_alloc_num = 1;
1383         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1384                 /* alloc all percpu data, currently only obd_memory use this */
1385                 for (i = 0; i < num_entry; ++i)
1386                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1387                                 goto fail;
1388         }
1389
1390         return stats;
1391
1392 fail:
1393         lprocfs_free_stats(&stats);
1394         return NULL;
1395 }
1396 EXPORT_SYMBOL(lprocfs_alloc_stats);
1397
1398 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1399 {
1400         struct lprocfs_stats *stats = *statsh;
1401         unsigned int num_entry;
1402         unsigned int percpusize;
1403         unsigned int i;
1404
1405         if (stats == NULL || stats->ls_num == 0)
1406                 return;
1407         *statsh = NULL;
1408
1409         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1410                 num_entry = 1;
1411         else
1412                 num_entry = num_possible_cpus();
1413
1414         percpusize = lprocfs_stats_counter_size(stats);
1415         for (i = 0; i < num_entry; i++)
1416                 if (stats->ls_percpu[i] != NULL)
1417                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1418         if (stats->ls_cnt_header != NULL)
1419                 LIBCFS_FREE(stats->ls_cnt_header, stats->ls_num *
1420                                         sizeof(struct lprocfs_counter_header));
1421         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1422 }
1423 EXPORT_SYMBOL(lprocfs_free_stats);
1424
1425 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1426                             enum lprocfs_fields_flags field)
1427 {
1428         unsigned long flags = 0;
1429         unsigned int num_cpu;
1430         unsigned int i;
1431         u64 ret = 0;
1432
1433         LASSERT(stats);
1434
1435         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1436         for (i = 0; i < num_cpu; i++) {
1437                 struct lprocfs_counter *cntr;
1438
1439                 if (!stats->ls_percpu[i])
1440                         continue;
1441
1442                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1443                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1444                                            stats->ls_flags, field);
1445         }
1446         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1447         return ret;
1448 }
1449 EXPORT_SYMBOL(lprocfs_stats_collector);
1450
1451 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1452 {
1453         struct lprocfs_counter          *percpu_cntr;
1454         int                             i;
1455         int                             j;
1456         unsigned int                    num_entry;
1457         unsigned long                   flags = 0;
1458
1459         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1460
1461         for (i = 0; i < num_entry; i++) {
1462                 if (stats->ls_percpu[i] == NULL)
1463                         continue;
1464                 for (j = 0; j < stats->ls_num; j++) {
1465                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1466                         percpu_cntr->lc_count           = 0;
1467                         percpu_cntr->lc_min             = LC_MIN_INIT;
1468                         percpu_cntr->lc_max             = 0;
1469                         percpu_cntr->lc_sumsquare       = 0;
1470                         percpu_cntr->lc_sum             = 0;
1471                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1472                                 percpu_cntr->lc_sum_irq = 0;
1473                 }
1474         }
1475
1476         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1477 }
1478 EXPORT_SYMBOL(lprocfs_clear_stats);
1479
1480 static ssize_t lprocfs_stats_seq_write(struct file *file,
1481                                        const char __user *buf,
1482                                        size_t len, loff_t *off)
1483 {
1484         struct seq_file *seq = file->private_data;
1485         struct lprocfs_stats *stats = seq->private;
1486
1487         lprocfs_clear_stats(stats);
1488
1489         return len;
1490 }
1491
1492 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1493 {
1494         struct lprocfs_stats *stats = p->private;
1495
1496         return (*pos < stats->ls_num) ? pos : NULL;
1497 }
1498
1499 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1500 {
1501 }
1502
1503 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1504 {
1505         (*pos)++;
1506
1507         return lprocfs_stats_seq_start(p, pos);
1508 }
1509
1510 /* seq file export of one lprocfs counter */
1511 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1512 {
1513         struct lprocfs_stats            *stats  = p->private;
1514         struct lprocfs_counter_header   *hdr;
1515         struct lprocfs_counter           ctr;
1516         int                              idx    = *(loff_t *)v;
1517
1518         if (idx == 0) {
1519                 struct timespec64 now;
1520
1521                 ktime_get_real_ts64(&now);
1522                 seq_printf(p, "%-25s %llu.%09lu secs.nsecs\n",
1523                            "snapshot_time", (s64)now.tv_sec, now.tv_nsec);
1524         }
1525
1526         hdr = &stats->ls_cnt_header[idx];
1527         lprocfs_stats_collect(stats, idx, &ctr);
1528
1529         if (ctr.lc_count == 0)
1530                 return 0;
1531
1532         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1533                    ctr.lc_count, hdr->lc_units);
1534
1535         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1536                 seq_printf(p, " %lld %lld %lld",
1537                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1538                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1539                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1540         }
1541         seq_putc(p, '\n');
1542         return 0;
1543 }
1544
1545 static const struct seq_operations lprocfs_stats_seq_sops = {
1546         .start  = lprocfs_stats_seq_start,
1547         .stop   = lprocfs_stats_seq_stop,
1548         .next   = lprocfs_stats_seq_next,
1549         .show   = lprocfs_stats_seq_show,
1550 };
1551
1552 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1553 {
1554         struct seq_file *seq;
1555         int rc;
1556
1557         rc = LPROCFS_ENTRY_CHECK(inode);
1558         if (rc < 0)
1559                 return rc;
1560
1561         rc = seq_open(file, &lprocfs_stats_seq_sops);
1562         if (rc)
1563                 return rc;
1564         seq = file->private_data;
1565         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1566         return 0;
1567 }
1568
1569 static const struct file_operations lprocfs_stats_seq_fops = {
1570         .owner   = THIS_MODULE,
1571         .open    = lprocfs_stats_seq_open,
1572         .read    = seq_read,
1573         .write   = lprocfs_stats_seq_write,
1574         .llseek  = seq_lseek,
1575         .release = lprocfs_seq_release,
1576 };
1577
1578 int ldebugfs_register_stats(struct dentry *parent, const char *name,
1579                             struct lprocfs_stats *stats)
1580 {
1581         struct dentry *entry;
1582
1583         LASSERT(!IS_ERR_OR_NULL(parent));
1584
1585         entry = debugfs_create_file(name, 0644, parent, stats,
1586                                     &lprocfs_stats_seq_fops);
1587         if (IS_ERR_OR_NULL(entry))
1588                 return entry ? PTR_ERR(entry) : -ENOMEM;
1589
1590         return 0;
1591 }
1592 EXPORT_SYMBOL_GPL(ldebugfs_register_stats);
1593
1594 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1595                            struct lprocfs_stats *stats)
1596 {
1597         struct proc_dir_entry *entry;
1598         LASSERT(root != NULL);
1599
1600         entry = proc_create_data(name, 0644, root,
1601                                  &lprocfs_stats_seq_fops, stats);
1602         if (entry == NULL)
1603                 return -ENOMEM;
1604         return 0;
1605 }
1606 EXPORT_SYMBOL(lprocfs_register_stats);
1607
1608 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1609                           unsigned conf, const char *name, const char *units)
1610 {
1611         struct lprocfs_counter_header   *header;
1612         struct lprocfs_counter          *percpu_cntr;
1613         unsigned long                   flags = 0;
1614         unsigned int                    i;
1615         unsigned int                    num_cpu;
1616
1617         LASSERT(stats != NULL);
1618
1619         header = &stats->ls_cnt_header[index];
1620         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1621                  index, name, units);
1622
1623         header->lc_config = conf;
1624         header->lc_name   = name;
1625         header->lc_units  = units;
1626
1627         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1628         for (i = 0; i < num_cpu; ++i) {
1629                 if (stats->ls_percpu[i] == NULL)
1630                         continue;
1631                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1632                 percpu_cntr->lc_count           = 0;
1633                 percpu_cntr->lc_min             = LC_MIN_INIT;
1634                 percpu_cntr->lc_max             = 0;
1635                 percpu_cntr->lc_sumsquare       = 0;
1636                 percpu_cntr->lc_sum             = 0;
1637                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1638                         percpu_cntr->lc_sum_irq = 0;
1639         }
1640         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1641 }
1642 EXPORT_SYMBOL(lprocfs_counter_init);
1643
1644 /* Note that we only init md counters for ops whose offset is less
1645  * than NUM_MD_STATS. This is explained in a comment in the definition
1646  * of struct md_ops. */
1647 #define LPROCFS_MD_OP_INIT(base, stats, op)                                    \
1648         do {                                                                   \
1649                 unsigned int _idx = base + MD_COUNTER_OFFSET(op);              \
1650                                                                                \
1651                 if (MD_COUNTER_OFFSET(op) < NUM_MD_STATS) {                    \
1652                         LASSERT(_idx < stats->ls_num);                         \
1653                         lprocfs_counter_init(stats, _idx, 0, #op, "reqs");     \
1654                 }                                                              \
1655         } while (0)
1656
1657 void lprocfs_init_mps_stats(int num_private_stats, struct lprocfs_stats *stats)
1658 {
1659         LPROCFS_MD_OP_INIT(num_private_stats, stats, get_root);
1660         LPROCFS_MD_OP_INIT(num_private_stats, stats, null_inode);
1661         LPROCFS_MD_OP_INIT(num_private_stats, stats, close);
1662         LPROCFS_MD_OP_INIT(num_private_stats, stats, create);
1663         LPROCFS_MD_OP_INIT(num_private_stats, stats, enqueue);
1664         LPROCFS_MD_OP_INIT(num_private_stats, stats, getattr);
1665         LPROCFS_MD_OP_INIT(num_private_stats, stats, getattr_name);
1666         LPROCFS_MD_OP_INIT(num_private_stats, stats, intent_lock);
1667         LPROCFS_MD_OP_INIT(num_private_stats, stats, link);
1668         LPROCFS_MD_OP_INIT(num_private_stats, stats, rename);
1669         LPROCFS_MD_OP_INIT(num_private_stats, stats, setattr);
1670         LPROCFS_MD_OP_INIT(num_private_stats, stats, fsync);
1671         LPROCFS_MD_OP_INIT(num_private_stats, stats, read_page);
1672         LPROCFS_MD_OP_INIT(num_private_stats, stats, unlink);
1673         LPROCFS_MD_OP_INIT(num_private_stats, stats, setxattr);
1674         LPROCFS_MD_OP_INIT(num_private_stats, stats, getxattr);
1675         LPROCFS_MD_OP_INIT(num_private_stats, stats, init_ea_size);
1676         LPROCFS_MD_OP_INIT(num_private_stats, stats, get_lustre_md);
1677         LPROCFS_MD_OP_INIT(num_private_stats, stats, free_lustre_md);
1678         LPROCFS_MD_OP_INIT(num_private_stats, stats, merge_attr);
1679         LPROCFS_MD_OP_INIT(num_private_stats, stats, set_open_replay_data);
1680         LPROCFS_MD_OP_INIT(num_private_stats, stats, clear_open_replay_data);
1681         LPROCFS_MD_OP_INIT(num_private_stats, stats, set_lock_data);
1682         LPROCFS_MD_OP_INIT(num_private_stats, stats, lock_match);
1683         LPROCFS_MD_OP_INIT(num_private_stats, stats, cancel_unused);
1684         LPROCFS_MD_OP_INIT(num_private_stats, stats, intent_getattr_async);
1685         LPROCFS_MD_OP_INIT(num_private_stats, stats, revalidate_lock);
1686 }
1687
1688 int lprocfs_alloc_md_stats(struct obd_device *obd,
1689                            unsigned int num_private_stats)
1690 {
1691         struct lprocfs_stats *stats;
1692         unsigned int num_stats;
1693         int rc, i;
1694
1695         CLASSERT(offsetof(struct md_ops, MD_STATS_FIRST_OP) == 0);
1696         CLASSERT(_MD_COUNTER_OFFSET(MD_STATS_FIRST_OP) == 0);
1697         CLASSERT(_MD_COUNTER_OFFSET(MD_STATS_LAST_OP) > 0);
1698
1699         /* TODO Ensure that this function is only used where
1700          * appropriate by adding an assertion to the effect that
1701          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1702          * because mdt_procfs_init() uses this function to allocate
1703          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1704          * mdt layer does not use the md_ops interface. This is
1705          * confusing and a waste of memory. See LU-2484.
1706          */
1707         LASSERT(obd->obd_proc_entry != NULL);
1708         LASSERT(obd->obd_md_stats == NULL);
1709         LASSERT(obd->obd_md_cntr_base == 0);
1710
1711         num_stats = NUM_MD_STATS + num_private_stats;
1712         stats = lprocfs_alloc_stats(num_stats, 0);
1713         if (stats == NULL)
1714                 return -ENOMEM;
1715
1716         lprocfs_init_mps_stats(num_private_stats, stats);
1717
1718         for (i = num_private_stats; i < num_stats; i++) {
1719                 if (stats->ls_cnt_header[i].lc_name == NULL) {
1720                         CERROR("Missing md_stat initializer md_op "
1721                                "operation at offset %d. Aborting.\n",
1722                                i - num_private_stats);
1723                         LBUG();
1724                 }
1725         }
1726
1727         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1728         if (rc < 0) {
1729                 lprocfs_free_stats(&stats);
1730         } else {
1731                 obd->obd_md_stats = stats;
1732                 obd->obd_md_cntr_base = num_private_stats;
1733         }
1734
1735         return rc;
1736 }
1737 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1738
1739 void lprocfs_free_md_stats(struct obd_device *obd)
1740 {
1741         struct lprocfs_stats *stats = obd->obd_md_stats;
1742
1743         if (stats != NULL) {
1744                 obd->obd_md_stats = NULL;
1745                 obd->obd_md_cntr_base = 0;
1746                 lprocfs_free_stats(&stats);
1747         }
1748 }
1749 EXPORT_SYMBOL(lprocfs_free_md_stats);
1750
1751 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1752 {
1753         lprocfs_counter_init(ldlm_stats,
1754                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1755                              0, "ldlm_enqueue", "reqs");
1756         lprocfs_counter_init(ldlm_stats,
1757                              LDLM_CONVERT - LDLM_FIRST_OPC,
1758                              0, "ldlm_convert", "reqs");
1759         lprocfs_counter_init(ldlm_stats,
1760                              LDLM_CANCEL - LDLM_FIRST_OPC,
1761                              0, "ldlm_cancel", "reqs");
1762         lprocfs_counter_init(ldlm_stats,
1763                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1764                              0, "ldlm_bl_callback", "reqs");
1765         lprocfs_counter_init(ldlm_stats,
1766                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1767                              0, "ldlm_cp_callback", "reqs");
1768         lprocfs_counter_init(ldlm_stats,
1769                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1770                              0, "ldlm_gl_callback", "reqs");
1771 }
1772 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1773
1774 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1775                           struct lprocfs_counter_header *header,
1776                           enum lprocfs_stats_flags flags,
1777                           enum lprocfs_fields_flags field)
1778 {
1779         __s64 ret = 0;
1780
1781         if (lc == NULL || header == NULL)
1782                 RETURN(0);
1783
1784         switch (field) {
1785                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1786                         ret = header->lc_config;
1787                         break;
1788                 case LPROCFS_FIELDS_FLAGS_SUM:
1789                         ret = lc->lc_sum;
1790                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1791                                 ret += lc->lc_sum_irq;
1792                         break;
1793                 case LPROCFS_FIELDS_FLAGS_MIN:
1794                         ret = lc->lc_min;
1795                         break;
1796                 case LPROCFS_FIELDS_FLAGS_MAX:
1797                         ret = lc->lc_max;
1798                         break;
1799                 case LPROCFS_FIELDS_FLAGS_AVG:
1800                         ret = (lc->lc_max - lc->lc_min) / 2;
1801                         break;
1802                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1803                         ret = lc->lc_sumsquare;
1804                         break;
1805                 case LPROCFS_FIELDS_FLAGS_COUNT:
1806                         ret = lc->lc_count;
1807                         break;
1808                 default:
1809                         break;
1810         };
1811         RETURN(ret);
1812 }
1813 EXPORT_SYMBOL(lprocfs_read_helper);
1814
1815 int lprocfs_read_frac_helper(char *buffer, unsigned long count, long val,
1816                              int mult)
1817 {
1818         long decimal_val, frac_val;
1819         int prtn;
1820
1821         if (count < 10)
1822                 return -EINVAL;
1823
1824         decimal_val = val / mult;
1825         prtn = snprintf(buffer, count, "%ld", decimal_val);
1826         frac_val = val % mult;
1827
1828         if (prtn < (count - 4) && frac_val > 0) {
1829                 long temp_frac;
1830                 int i, temp_mult = 1, frac_bits = 0;
1831
1832                 temp_frac = frac_val * 10;
1833                 buffer[prtn++] = '.';
1834                 while (frac_bits < 2 && (temp_frac / mult) < 1 ) {
1835                         /* only reserved 2 bits fraction */
1836                         buffer[prtn++] ='0';
1837                         temp_frac *= 10;
1838                         frac_bits++;
1839                 }
1840                 /*
1841                  * Need to think these cases :
1842                  *      1. #echo x.00 > /proc/xxx       output result : x
1843                  *      2. #echo x.0x > /proc/xxx       output result : x.0x
1844                  *      3. #echo x.x0 > /proc/xxx       output result : x.x
1845                  *      4. #echo x.xx > /proc/xxx       output result : x.xx
1846                  *      Only reserved 2 bits fraction.
1847                  */
1848                 for (i = 0; i < (5 - prtn); i++)
1849                         temp_mult *= 10;
1850
1851                 frac_bits = min((int)count - prtn, 3 - frac_bits);
1852                 prtn += snprintf(buffer + prtn, frac_bits, "%ld",
1853                                  frac_val * temp_mult / mult);
1854
1855                 prtn--;
1856                 while(buffer[prtn] < '1' || buffer[prtn] > '9') {
1857                         prtn--;
1858                         if (buffer[prtn] == '.') {
1859                                 prtn--;
1860                                 break;
1861                         }
1862                 }
1863                 prtn++;
1864         }
1865         buffer[prtn++] ='\n';
1866         return prtn;
1867 }
1868 EXPORT_SYMBOL(lprocfs_read_frac_helper);
1869
1870 int lprocfs_seq_read_frac_helper(struct seq_file *m, long val, int mult)
1871 {
1872         long decimal_val, frac_val;
1873
1874         decimal_val = val / mult;
1875         seq_printf(m, "%ld", decimal_val);
1876         frac_val = val % mult;
1877
1878         if (frac_val > 0) {
1879                 frac_val *= 100;
1880                 frac_val /= mult;
1881         }
1882         if (frac_val > 0) {
1883                 /* Three cases: x0, xx, 0x */
1884                 if ((frac_val % 10) != 0)
1885                         seq_printf(m, ".%ld", frac_val);
1886                 else
1887                         seq_printf(m, ".%ld", frac_val / 10);
1888         }
1889
1890         seq_printf(m, "\n");
1891         return 0;
1892 }
1893 EXPORT_SYMBOL(lprocfs_seq_read_frac_helper);
1894
1895 /* Obtains the conversion factor for the unit specified */
1896 static int get_mult(char unit, __u64 *mult)
1897 {
1898         __u64 units = 1;
1899
1900         switch (unit) {
1901         /* peta, tera, giga, mega, and kilo */
1902         case 'p':
1903         case 'P':
1904                 units <<= 10;
1905         case 't':
1906         case 'T':
1907                 units <<= 10;
1908         case 'g':
1909         case 'G':
1910                 units <<= 10;
1911         case 'm':
1912         case 'M':
1913                 units <<= 10;
1914         case 'k':
1915         case 'K':
1916                 units <<= 10;
1917                 break;
1918         /* some tests expect % to be accepted */
1919         case '%':
1920                 units = 1;
1921                 break;
1922         default:
1923                 return -EINVAL;
1924         }
1925
1926         *mult = units;
1927
1928         return 0;
1929 }
1930
1931 /*
1932  * Ensures the numeric string is valid. The function provides the final
1933  * multiplier in the case a unit exists at the end of the string. It also
1934  * locates the start of the whole and fractional parts (if any). This
1935  * function modifies the string so kstrtoull can be used to parse both
1936  * the whole and fraction portions. This function also figures out
1937  * the base of the number.
1938  */
1939 static int preprocess_numeric_str(char *buffer, __u64 *mult, __u64 def_mult,
1940                                   bool allow_units, char **whole, char **frac,
1941                                   unsigned int *base)
1942 {
1943         bool hit_decimal = false;
1944         bool hit_unit = false;
1945         int rc = 0;
1946         char *start;
1947         *mult = def_mult;
1948         *whole = NULL;
1949         *frac = NULL;
1950         *base = 10;
1951
1952         /* a hex string if it starts with "0x" */
1953         if (buffer[0] == '0' && tolower(buffer[1]) == 'x') {
1954                 *base = 16;
1955                 buffer += 2;
1956         }
1957
1958         start = buffer;
1959
1960         while (*buffer) {
1961                 /* allow for a single new line before the null terminator */
1962                 if (*buffer == '\n') {
1963                         *buffer = '\0';
1964                         buffer++;
1965
1966                         if (*buffer)
1967                                 return -EINVAL;
1968
1969                         break;
1970                 }
1971
1972                 /* any chars after our unit indicates a malformed string */
1973                 if (hit_unit)
1974                         return -EINVAL;
1975
1976                 /* ensure we only hit one decimal */
1977                 if (*buffer == '.') {
1978                         if (hit_decimal)
1979                                 return -EINVAL;
1980
1981                         /* if past start, there's a whole part */
1982                         if (start != buffer)
1983                                 *whole = start;
1984
1985                         *buffer = '\0';
1986                         start = buffer + 1;
1987                         hit_decimal = true;
1988                 } else if (!isdigit(*buffer) &&
1989                            !(*base == 16 && isxdigit(*buffer))) {
1990                         if (allow_units) {
1991                                 /* if we allow units, attempt to get mult */
1992                                 hit_unit = true;
1993                                 rc = get_mult(*buffer, mult);
1994                                 if (rc)
1995                                         return rc;
1996
1997                                 /* string stops here, but keep processing */
1998                                 *buffer = '\0';
1999                         } else {
2000                                 /* bad string */
2001                                 return -EINVAL;
2002                         }
2003                 }
2004
2005                 buffer++;
2006         }
2007
2008         if (hit_decimal) {
2009                 /* hit a decimal, make sure there's a fractional part */
2010                 if (!*start)
2011                         return -EINVAL;
2012
2013                 *frac = start;
2014         } else {
2015                 /* didn't hit a decimal, but may have a whole part */
2016                 if (start != buffer && *start)
2017                         *whole = start;
2018         }
2019
2020         /* malformed string if we didn't get anything */
2021         if (!*frac && !*whole)
2022                 return -EINVAL;
2023
2024         return 0;
2025 }
2026
2027 /*
2028  * Parses a numeric string which can contain a whole and fraction portion
2029  * into a __u64. Accepts a multiplier to apply to the value parsed. Also
2030  * allows the string to have a unit at the end. The function handles
2031  * wrapping of the final unsigned value.
2032  */
2033 static int str_to_u64_parse(char *buffer, unsigned long count,
2034                             __u64 *val, __u64 def_mult, bool allow_units)
2035 {
2036         __u64 whole = 0;
2037         __u64 frac = 0;
2038         unsigned int frac_d = 1;
2039         __u64 wrap_indicator = ULLONG_MAX;
2040         int rc = 0;
2041         __u64 mult;
2042         char *strwhole;
2043         char *strfrac;
2044         unsigned int base = 10;
2045
2046         rc = preprocess_numeric_str(buffer, &mult, def_mult, allow_units,
2047                                     &strwhole, &strfrac, &base);
2048
2049         if (rc)
2050                 return rc;
2051
2052         if (mult == 0) {
2053                 *val = 0;
2054                 return 0;
2055         }
2056
2057         /* the multiplier limits how large the value can be */
2058         wrap_indicator = div64_u64(wrap_indicator, mult);
2059
2060         if (strwhole) {
2061                 rc = kstrtoull(strwhole, base, &whole);
2062                 if (rc)
2063                         return rc;
2064
2065                 if (whole > wrap_indicator)
2066                         return -ERANGE;
2067
2068                 whole *= mult;
2069         }
2070
2071         if (strfrac) {
2072                 if (strlen(strfrac) > 10)
2073                         strfrac[10] = '\0';
2074
2075                 rc = kstrtoull(strfrac, base, &frac);
2076                 if (rc)
2077                         return rc;
2078
2079                 /* determine power of fractional portion */
2080                 while (*strfrac) {
2081                         frac_d *= base;
2082                         strfrac++;
2083                 }
2084
2085                 /* fractional portion is too large to perform calculation */
2086                 if (frac > wrap_indicator)
2087                         return -ERANGE;
2088
2089                 frac *= mult;
2090                 do_div(frac, frac_d);
2091         }
2092
2093         /* check that the sum of whole and fraction fits in u64 */
2094         if (whole > (ULLONG_MAX - frac))
2095                 return -ERANGE;
2096
2097         *val = whole + frac;
2098
2099         return 0;
2100 }
2101
2102 /*
2103  * This function parses numeric/hex strings into __s64. It accepts a multiplier
2104  * which will apply to the value parsed. It also can allow the string to
2105  * have a unit as the last character. The function handles overflow/underflow
2106  * of the signed integer.
2107  */
2108 static int str_to_s64_internal(const char __user *buffer, unsigned long count,
2109                                __s64 *val, __u64 def_mult, bool allow_units)
2110 {
2111         char kernbuf[22];
2112         __u64 tmp;
2113         unsigned int offset = 0;
2114         int signed sign = 1;
2115         __u64 max = LLONG_MAX;
2116         int rc = 0;
2117
2118         if (count > (sizeof(kernbuf) - 1))
2119                 return -EINVAL;
2120
2121         if (copy_from_user(kernbuf, buffer, count))
2122                 return -EFAULT;
2123
2124         kernbuf[count] = '\0';
2125
2126         /* keep track of our sign */
2127         if (*kernbuf == '-') {
2128                 sign = -1;
2129                 offset++;
2130                 /* equivalent to max = -LLONG_MIN, avoids overflow */
2131                 max++;
2132         }
2133
2134         rc = str_to_u64_parse(kernbuf + offset, count - offset,
2135                               &tmp, def_mult, allow_units);
2136         if (rc)
2137                 return rc;
2138
2139         /* check for overflow/underflow */
2140         if (max < tmp)
2141                 return -ERANGE;
2142
2143         *val = (__s64)tmp * sign;
2144
2145         return 0;
2146 }
2147
2148 /**
2149  * Convert a user string into a signed 64 bit number. This function produces
2150  * an error when the value parsed from the string times multiplier underflows or
2151  * overflows. This function only accepts strings that contains digits, an
2152  * optional decimal, and a char representing a unit at the end. If a unit is
2153  * specified in the string, the multiplier provided by the caller is ignored.
2154  * This function can also accept hexadecimal strings which are prefixed with
2155  * "0x".
2156  *
2157  * \param[in] buffer    string consisting of numbers, a decimal, and a unit
2158  * \param[in] count     buffer length
2159  * \param[in] val       if successful, the value represented by the string
2160  * \param[in] defunit   default unit if string doesn't contain one
2161  *
2162  * \retval              0 on success
2163  * \retval              negative number on error
2164  */
2165 int lprocfs_str_with_units_to_s64(const char __user *buffer,
2166                                   unsigned long count, __s64 *val, char defunit)
2167 {
2168         __u64 mult = 1;
2169         int rc;
2170
2171         if (defunit != '1') {
2172                 rc = get_mult(defunit, &mult);
2173                 if (rc)
2174                         return rc;
2175         }
2176
2177         return str_to_s64_internal(buffer, count, val, mult, true);
2178 }
2179 EXPORT_SYMBOL(lprocfs_str_with_units_to_s64);
2180
2181 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
2182 {
2183         size_t l2;
2184
2185         l2 = strlen(s2);
2186         if (!l2)
2187                 return (char *)s1;
2188         while (len >= l2) {
2189                 len--;
2190                 if (!memcmp(s1, s2, l2))
2191                         return (char *)s1;
2192                 s1++;
2193         }
2194         return NULL;
2195 }
2196 EXPORT_SYMBOL(lprocfs_strnstr);
2197
2198 /**
2199  * Find the string \a name in the input \a buffer, and return a pointer to the
2200  * value immediately following \a name, reducing \a count appropriately.
2201  * If \a name is not found the original \a buffer is returned.
2202  */
2203 char *lprocfs_find_named_value(const char *buffer, const char *name,
2204                                 size_t *count)
2205 {
2206         char *val;
2207         size_t buflen = *count;
2208
2209         /* there is no strnstr() in rhel5 and ubuntu kernels */
2210         val = lprocfs_strnstr(buffer, name, buflen);
2211         if (val == NULL)
2212                 return (char *)buffer;
2213
2214         val += strlen(name);                             /* skip prefix */
2215         while (val < buffer + buflen && isspace(*val)) /* skip separator */
2216                 val++;
2217
2218         *count = 0;
2219         while (val < buffer + buflen && isalnum(*val)) {
2220                 ++*count;
2221                 ++val;
2222         }
2223
2224         return val - *count;
2225 }
2226 EXPORT_SYMBOL(lprocfs_find_named_value);
2227
2228 int ldebugfs_seq_create(struct dentry *parent, const char *name, umode_t mode,
2229                         const struct file_operations *seq_fops, void *data)
2230 {
2231         struct dentry *entry;
2232
2233         /* Disallow secretly (un)writable entries. */
2234         LASSERT((!seq_fops->write) == (!(mode & 0222)));
2235
2236         entry = debugfs_create_file(name, mode, parent, data, seq_fops);
2237         if (IS_ERR_OR_NULL(entry))
2238                 return entry ? PTR_ERR(entry) : -ENOMEM;
2239
2240         return 0;
2241 }
2242 EXPORT_SYMBOL_GPL(ldebugfs_seq_create);
2243
2244 int lprocfs_seq_create(struct proc_dir_entry *parent,
2245                        const char *name,
2246                        mode_t mode,
2247                        const struct file_operations *seq_fops,
2248                        void *data)
2249 {
2250         struct proc_dir_entry *entry;
2251         ENTRY;
2252
2253         /* Disallow secretly (un)writable entries. */
2254         LASSERT((seq_fops->write == NULL) == ((mode & 0222) == 0));
2255
2256         entry = proc_create_data(name, mode, parent, seq_fops, data);
2257
2258         if (entry == NULL)
2259                 RETURN(-ENOMEM);
2260
2261         RETURN(0);
2262 }
2263 EXPORT_SYMBOL(lprocfs_seq_create);
2264
2265 int lprocfs_obd_seq_create(struct obd_device *dev,
2266                            const char *name,
2267                            mode_t mode,
2268                            const struct file_operations *seq_fops,
2269                            void *data)
2270 {
2271         return (lprocfs_seq_create(dev->obd_proc_entry, name,
2272                                    mode, seq_fops, data));
2273 }
2274 EXPORT_SYMBOL(lprocfs_obd_seq_create);
2275
2276 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
2277 {
2278         if (value >= OBD_HIST_MAX)
2279                 value = OBD_HIST_MAX - 1;
2280
2281         spin_lock(&oh->oh_lock);
2282         oh->oh_buckets[value]++;
2283         spin_unlock(&oh->oh_lock);
2284 }
2285 EXPORT_SYMBOL(lprocfs_oh_tally);
2286
2287 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
2288 {
2289         unsigned int val = 0;
2290
2291         if (likely(value != 0))
2292                 val = min(fls(value - 1), OBD_HIST_MAX);
2293
2294         lprocfs_oh_tally(oh, val);
2295 }
2296 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
2297
2298 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2299 {
2300         unsigned long ret = 0;
2301         int i;
2302
2303         for (i = 0; i < OBD_HIST_MAX; i++)
2304                 ret +=  oh->oh_buckets[i];
2305         return ret;
2306 }
2307 EXPORT_SYMBOL(lprocfs_oh_sum);
2308
2309 void lprocfs_oh_clear(struct obd_histogram *oh)
2310 {
2311         spin_lock(&oh->oh_lock);
2312         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2313         spin_unlock(&oh->oh_lock);
2314 }
2315 EXPORT_SYMBOL(lprocfs_oh_clear);
2316
2317 ssize_t lustre_attr_show(struct kobject *kobj,
2318                          struct attribute *attr, char *buf)
2319 {
2320         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2321
2322         return a->show ? a->show(kobj, attr, buf) : 0;
2323 }
2324 EXPORT_SYMBOL_GPL(lustre_attr_show);
2325
2326 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2327                           const char *buf, size_t len)
2328 {
2329         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2330
2331         return a->store ? a->store(kobj, attr, buf, len) : len;
2332 }
2333 EXPORT_SYMBOL_GPL(lustre_attr_store);
2334
2335 const struct sysfs_ops lustre_sysfs_ops = {
2336         .show  = lustre_attr_show,
2337         .store = lustre_attr_store,
2338 };
2339 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2340
2341 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2342 {
2343         struct obd_device *dev = data;
2344         struct client_obd *cli = &dev->u.cli;
2345
2346         spin_lock(&cli->cl_loi_list_lock);
2347         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2348         spin_unlock(&cli->cl_loi_list_lock);
2349         return 0;
2350 }
2351 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2352
2353 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2354                                                 const char __user *buffer,
2355                                                 size_t count, loff_t *off)
2356 {
2357         struct obd_device *dev =
2358                 ((struct seq_file *)file->private_data)->private;
2359         struct client_obd *cli = &dev->u.cli;
2360         struct obd_connect_data *ocd = &cli->cl_import->imp_connect_data;
2361         int chunk_mask, rc;
2362         s64 val;
2363
2364         rc = lprocfs_str_with_units_to_s64(buffer, count, &val, '1');
2365         if (rc)
2366                 return rc;
2367         if (val < 0)
2368                 return -ERANGE;
2369
2370         /* if the max_pages is specified in bytes, convert to pages */
2371         if (val >= ONE_MB_BRW_SIZE)
2372                 val >>= PAGE_SHIFT;
2373
2374         LPROCFS_CLIMP_CHECK(dev);
2375
2376         chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2377         /* max_pages_per_rpc must be chunk aligned */
2378         val = (val + ~chunk_mask) & chunk_mask;
2379         if (val == 0 || (ocd->ocd_brw_size != 0 &&
2380                          val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2381                 LPROCFS_CLIMP_EXIT(dev);
2382                 return -ERANGE;
2383         }
2384         spin_lock(&cli->cl_loi_list_lock);
2385         cli->cl_max_pages_per_rpc = val;
2386         client_adjust_max_dirty(cli);
2387         spin_unlock(&cli->cl_loi_list_lock);
2388
2389         LPROCFS_CLIMP_EXIT(dev);
2390         return count;
2391 }
2392 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2393
2394 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2395                             char *buf)
2396 {
2397         struct obd_device *dev = container_of(kobj, struct obd_device,
2398                                               obd_kset.kobj);
2399         struct client_obd *cli = &dev->u.cli;
2400         int rc;
2401
2402         spin_lock(&cli->cl_loi_list_lock);
2403         rc = sprintf(buf, "%d\n", cli->cl_short_io_bytes);
2404         spin_unlock(&cli->cl_loi_list_lock);
2405         return rc;
2406 }
2407 EXPORT_SYMBOL(short_io_bytes_show);
2408
2409 /* Used to catch people who think they're specifying pages. */
2410 #define MIN_SHORT_IO_BYTES 64
2411
2412 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2413                              const char *buffer, size_t count)
2414 {
2415         struct obd_device *dev = container_of(kobj, struct obd_device,
2416                                               obd_kset.kobj);
2417         struct client_obd *cli = &dev->u.cli;
2418         u32 val;
2419         int rc;
2420
2421         LPROCFS_CLIMP_CHECK(dev);
2422
2423         rc = kstrtouint(buffer, 0, &val);
2424         if (rc)
2425                 GOTO(out, rc);
2426
2427         if (val < MIN_SHORT_IO_BYTES || val > OBD_MAX_SHORT_IO_BYTES)
2428                 GOTO(out, rc = -ERANGE);
2429
2430         rc = count;
2431
2432         spin_lock(&cli->cl_loi_list_lock);
2433         if (val > (cli->cl_max_pages_per_rpc << PAGE_SHIFT))
2434                 rc = -ERANGE;
2435         else
2436                 cli->cl_short_io_bytes = val;
2437         spin_unlock(&cli->cl_loi_list_lock);
2438
2439 out:
2440         LPROCFS_CLIMP_EXIT(dev);
2441         return rc;
2442 }
2443 EXPORT_SYMBOL(short_io_bytes_store);
2444
2445 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2446                            struct root_squash_info *squash, char *name)
2447 {
2448         int rc;
2449         char kernbuf[64], *tmp, *errmsg;
2450         unsigned long uid, gid;
2451         ENTRY;
2452
2453         if (count >= sizeof(kernbuf)) {
2454                 errmsg = "string too long";
2455                 GOTO(failed_noprint, rc = -EINVAL);
2456         }
2457         if (copy_from_user(kernbuf, buffer, count)) {
2458                 errmsg = "bad address";
2459                 GOTO(failed_noprint, rc = -EFAULT);
2460         }
2461         kernbuf[count] = '\0';
2462
2463         /* look for uid gid separator */
2464         tmp = strchr(kernbuf, ':');
2465         if (tmp == NULL) {
2466                 errmsg = "needs uid:gid format";
2467                 GOTO(failed, rc = -EINVAL);
2468         }
2469         *tmp = '\0';
2470         tmp++;
2471
2472         /* parse uid */
2473         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2474                 errmsg = "bad uid";
2475                 GOTO(failed, rc = -EINVAL);
2476         }
2477
2478         /* parse gid */
2479         if (kstrtoul(tmp, 0, &gid) != 0) {
2480                 errmsg = "bad gid";
2481                 GOTO(failed, rc = -EINVAL);
2482         }
2483
2484         squash->rsi_uid = uid;
2485         squash->rsi_gid = gid;
2486
2487         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2488                       name, squash->rsi_uid, squash->rsi_gid);
2489         RETURN(count);
2490
2491 failed:
2492         if (tmp != NULL) {
2493                 tmp--;
2494                 *tmp = ':';
2495         }
2496         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2497               name, kernbuf, errmsg, rc);
2498         RETURN(rc);
2499 failed_noprint:
2500         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2501               name, errmsg, rc);
2502         RETURN(rc);
2503 }
2504 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2505
2506
2507 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2508                              struct root_squash_info *squash, char *name)
2509 {
2510         int rc;
2511         char *kernbuf = NULL;
2512         char *errmsg;
2513         struct list_head tmp;
2514         int len = count;
2515         ENTRY;
2516
2517         if (count > 4096) {
2518                 errmsg = "string too long";
2519                 GOTO(failed, rc = -EINVAL);
2520         }
2521
2522         OBD_ALLOC(kernbuf, count + 1);
2523         if (kernbuf == NULL) {
2524                 errmsg = "no memory";
2525                 GOTO(failed, rc = -ENOMEM);
2526         }
2527         if (copy_from_user(kernbuf, buffer, count)) {
2528                 errmsg = "bad address";
2529                 GOTO(failed, rc = -EFAULT);
2530         }
2531         kernbuf[count] = '\0';
2532
2533         if (count > 0 && kernbuf[count - 1] == '\n')
2534                 len = count - 1;
2535
2536         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2537             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2538                 /* empty string is special case */
2539                 down_write(&squash->rsi_sem);
2540                 if (!list_empty(&squash->rsi_nosquash_nids))
2541                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2542                 up_write(&squash->rsi_sem);
2543                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2544                 OBD_FREE(kernbuf, count + 1);
2545                 RETURN(count);
2546         }
2547
2548         INIT_LIST_HEAD(&tmp);
2549         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2550                 errmsg = "can't parse";
2551                 GOTO(failed, rc = -EINVAL);
2552         }
2553         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2554                       name, kernbuf);
2555         OBD_FREE(kernbuf, count + 1);
2556         kernbuf = NULL;
2557
2558         down_write(&squash->rsi_sem);
2559         if (!list_empty(&squash->rsi_nosquash_nids))
2560                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2561         list_splice(&tmp, &squash->rsi_nosquash_nids);
2562         up_write(&squash->rsi_sem);
2563
2564         RETURN(count);
2565
2566 failed:
2567         if (kernbuf) {
2568                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2569                       name, kernbuf, errmsg, rc);
2570                 OBD_FREE(kernbuf, count + 1);
2571         } else {
2572                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2573                       name, errmsg, rc);
2574         }
2575         RETURN(rc);
2576 }
2577 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2578
2579 #endif /* CONFIG_PROC_FS*/