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  *
  31  * lustre/obdclass/lprocfs_status.c
  32  *
  33  * Author: Hariharan Thantry <thantry@users.sourceforge.net>
  34  */
  35
  36 #define DEBUG_SUBSYSTEM S_CLASS
  37
  38 #include <obd_class.h>
  39 #include <lprocfs_status.h>
  40
  41 #ifdef CONFIG_PROC_FS
  42
  43 static int lprocfs_no_percpu_stats = 0;
  44 module_param(lprocfs_no_percpu_stats, int, 0644);
  45 MODULE_PARM_DESC(lprocfs_no_percpu_stats, "Do not alloc percpu data for lprocfs stats");
  46
  47 #define MAX_STRING_SIZE 128
  48
  49 int lprocfs_single_release(struct inode *inode, struct file *file)
  50 {
  51         return single_release(inode, file);
  52 }
  53 EXPORT_SYMBOL(lprocfs_single_release);
  54
  55 int lprocfs_seq_release(struct inode *inode, struct file *file)
  56 {
  57         return seq_release(inode, file);
  58 }
  59 EXPORT_SYMBOL(lprocfs_seq_release);
  60
  61 static umode_t default_mode(const struct proc_ops *ops)
  62 {
  63         umode_t mode = 0;
  64
  65         if (ops->proc_read)
  66                 mode = 0444;
  67         if (ops->proc_write)
  68                 mode |= 0200;
  69
  70         return mode;
  71 }
  72
  73 struct proc_dir_entry *
  74 lprocfs_add_simple(struct proc_dir_entry *root, char *name,
  75                    void *data, const struct proc_ops *fops)
  76 {
  77         struct proc_dir_entry *proc;
  78         umode_t mode;
  79
  80         if (!root || !name || !fops)
  81                 return ERR_PTR(-EINVAL);
  82
  83         mode = default_mode(fops);
  84         proc = proc_create_data(name, mode, root, fops, data);
  85         if (!proc) {
  86                 CERROR("LprocFS: No memory to create /proc entry %s\n",
  87                        name);
  88                 return ERR_PTR(-ENOMEM);
  89         }
  90         return proc;
  91 }
  92 EXPORT_SYMBOL(lprocfs_add_simple);
  93
  94 struct proc_dir_entry *lprocfs_add_symlink(const char *name,
  95                                            struct proc_dir_entry *parent,
  96                                            const char *format, ...)
  97 {
  98         struct proc_dir_entry *entry;
  99         char *dest;
 100         va_list ap;
 101
 102         if (!parent || !format)
 103                 return NULL;
 104
 105         OBD_ALLOC_WAIT(dest, MAX_STRING_SIZE + 1);
 106         if (!dest)
 107                 return NULL;
 108
 109         va_start(ap, format);
 110         vsnprintf(dest, MAX_STRING_SIZE, format, ap);
 111         va_end(ap);
 112
 113         entry = proc_symlink(name, parent, dest);
 114         if (!entry)
 115                 CERROR("LprocFS: Could not create symbolic link from "
 116                        "%s to %s\n", name, dest);
 117
 118         OBD_FREE(dest, MAX_STRING_SIZE + 1);
 119         return entry;
 120 }
 121 EXPORT_SYMBOL(lprocfs_add_symlink);
 122
 123 static const struct file_operations ldebugfs_empty_ops = { };
 124
 125 void ldebugfs_add_vars(struct dentry *parent, struct ldebugfs_vars *list,
 126                        void *data)
 127 {
 128         if (IS_ERR_OR_NULL(parent) || IS_ERR_OR_NULL(list))
 129                 return;
 130
 131         while (list->name) {
 132                 umode_t mode = 0;
 133
 134                 if (list->proc_mode != 0000) {
 135                         mode = list->proc_mode;
 136                 } else if (list->fops) {
 137                         if (list->fops->read)
 138                                 mode = 0444;
 139                         if (list->fops->write)
 140                                 mode |= 0200;
 141                 }
 142                 debugfs_create_file(list->name, mode, parent,
 143                                     list->data ? : data,
 144                                     list->fops ? : &ldebugfs_empty_ops);
 145                 list++;
 146         }
 147 }
 148 EXPORT_SYMBOL_GPL(ldebugfs_add_vars);
 149
 150 static const struct proc_ops lprocfs_empty_ops = { };
 151
 152 /**
 153  * Add /proc entries.
 154  *
 155  * \param root [in]  The parent proc entry on which new entry will be added.
 156  * \param list [in]  Array of proc entries to be added.
 157  * \param data [in]  The argument to be passed when entries read/write routines
 158  *                   are called through /proc file.
 159  *
 160  * \retval 0   on success
 161  *         < 0 on error
 162  */
 163 int
 164 lprocfs_add_vars(struct proc_dir_entry *root, struct lprocfs_vars *list,
 165                  void *data)
 166 {
 167         if (!root || !list)
 168                 return -EINVAL;
 169
 170         while (list->name) {
 171                 struct proc_dir_entry *proc;
 172                 umode_t mode = 0;
 173
 174                 if (list->proc_mode)
 175                         mode = list->proc_mode;
 176                 else if (list->fops)
 177                         mode = default_mode(list->fops);
 178                 proc = proc_create_data(list->name, mode, root,
 179                                         list->fops ?: &lprocfs_empty_ops,
 180                                         list->data ?: data);
 181                 if (!proc)
 182                         return -ENOMEM;
 183                 list++;
 184         }
 185         return 0;
 186 }
 187 EXPORT_SYMBOL(lprocfs_add_vars);
 188
 189 void lprocfs_remove(struct proc_dir_entry **rooth)
 190 {
 191         proc_remove(*rooth);
 192         *rooth = NULL;
 193 }
 194 EXPORT_SYMBOL(lprocfs_remove);
 195
 196 void lprocfs_remove_proc_entry(const char *name, struct proc_dir_entry *parent)
 197 {
 198         LASSERT(parent != NULL);
 199         remove_proc_entry(name, parent);
 200 }
 201 EXPORT_SYMBOL(lprocfs_remove_proc_entry);
 202
 203 struct proc_dir_entry *
 204 lprocfs_register(const char *name, struct proc_dir_entry *parent,
 205                  struct lprocfs_vars *list, void *data)
 206 {
 207         struct proc_dir_entry *newchild;
 208
 209         newchild = proc_mkdir(name, parent);
 210         if (!newchild)
 211                 return ERR_PTR(-ENOMEM);
 212
 213         if (list) {
 214                 int rc = lprocfs_add_vars(newchild, list, data);
 215                 if (rc) {
 216                         lprocfs_remove(&newchild);
 217                         return ERR_PTR(rc);
 218                 }
 219         }
 220         return newchild;
 221 }
 222 EXPORT_SYMBOL(lprocfs_register);
 223
 224 /* Generic callbacks */
 225 int lprocfs_uuid_seq_show(struct seq_file *m, void *data)
 226 {
 227         struct obd_device *obd = data;
 228
 229         LASSERT(obd != NULL);
 230         seq_printf(m, "%s\n", obd->obd_uuid.uuid);
 231         return 0;
 232 }
 233 EXPORT_SYMBOL(lprocfs_uuid_seq_show);
 234
 235 static ssize_t uuid_show(struct kobject *kobj, struct attribute *attr,
 236                          char *buf)
 237 {
 238         struct obd_device *obd = container_of(kobj, struct obd_device,
 239                                               obd_kset.kobj);
 240
 241         return sprintf(buf, "%s\n", obd->obd_uuid.uuid);
 242 }
 243 LUSTRE_RO_ATTR(uuid);
 244
 245 static ssize_t blocksize_show(struct kobject *kobj, struct attribute *attr,
 246                               char *buf)
 247 {
 248         struct obd_device *obd = container_of(kobj, struct obd_device,
 249                                               obd_kset.kobj);
 250         struct obd_statfs osfs;
 251         int rc;
 252
 253         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 254                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 255                         OBD_STATFS_NODELAY);
 256         if (!rc)
 257                 return sprintf(buf, "%u\n", osfs.os_bsize);
 258
 259         return rc;
 260 }
 261 LUSTRE_RO_ATTR(blocksize);
 262
 263 static ssize_t kbytestotal_show(struct kobject *kobj, struct attribute *attr,
 264                                 char *buf)
 265 {
 266         struct obd_device *obd = container_of(kobj, struct obd_device,
 267                                               obd_kset.kobj);
 268         struct obd_statfs osfs;
 269         int rc;
 270
 271         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 272                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 273                         OBD_STATFS_NODELAY);
 274         if (!rc) {
 275                 u32 blk_size = osfs.os_bsize >> 10;
 276                 u64 result = osfs.os_blocks;
 277
 278                 result *= rounddown_pow_of_two(blk_size ?: 1);
 279                 return sprintf(buf, "%llu\n", result);
 280         }
 281
 282         return rc;
 283 }
 284 LUSTRE_RO_ATTR(kbytestotal);
 285
 286 static ssize_t kbytesfree_show(struct kobject *kobj, struct attribute *attr,
 287                                char *buf)
 288 {
 289         struct obd_device *obd = container_of(kobj, struct obd_device,
 290                                               obd_kset.kobj);
 291         struct obd_statfs osfs;
 292         int rc;
 293
 294         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 295                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 296                         OBD_STATFS_NODELAY);
 297         if (!rc) {
 298                 u32 blk_size = osfs.os_bsize >> 10;
 299                 u64 result = osfs.os_bfree;
 300
 301                 while (blk_size >>= 1)
 302                         result <<= 1;
 303
 304                 return sprintf(buf, "%llu\n", result);
 305         }
 306
 307         return rc;
 308 }
 309 LUSTRE_RO_ATTR(kbytesfree);
 310
 311 static ssize_t kbytesavail_show(struct kobject *kobj, struct attribute *attr,
 312                                 char *buf)
 313 {
 314         struct obd_device *obd = container_of(kobj, struct obd_device,
 315                                               obd_kset.kobj);
 316         struct obd_statfs osfs;
 317         int rc;
 318
 319         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 320                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 321                         OBD_STATFS_NODELAY);
 322         if (!rc) {
 323                 u32 blk_size = osfs.os_bsize >> 10;
 324                 u64 result = osfs.os_bavail;
 325
 326                 while (blk_size >>= 1)
 327                         result <<= 1;
 328
 329                 return sprintf(buf, "%llu\n", result);
 330         }
 331
 332         return rc;
 333 }
 334 LUSTRE_RO_ATTR(kbytesavail);
 335
 336 static ssize_t filestotal_show(struct kobject *kobj, struct attribute *attr,
 337                                char *buf)
 338 {
 339         struct obd_device *obd = container_of(kobj, struct obd_device,
 340                                               obd_kset.kobj);
 341         struct obd_statfs osfs;
 342         int rc;
 343
 344         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 345                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 346                         OBD_STATFS_NODELAY);
 347         if (!rc)
 348                 return sprintf(buf, "%llu\n", osfs.os_files);
 349
 350         return rc;
 351 }
 352 LUSTRE_RO_ATTR(filestotal);
 353
 354 static ssize_t filesfree_show(struct kobject *kobj, struct attribute *attr,
 355                               char *buf)
 356 {
 357         struct obd_device *obd = container_of(kobj, struct obd_device,
 358                                               obd_kset.kobj);
 359         struct obd_statfs osfs;
 360         int rc;
 361
 362         rc = obd_statfs(NULL, obd->obd_self_export, &osfs,
 363                         ktime_get_seconds() - OBD_STATFS_CACHE_SECONDS,
 364                         OBD_STATFS_NODELAY);
 365         if (!rc)
 366                 return sprintf(buf, "%llu\n", osfs.os_ffree);
 367
 368         return rc;
 369 }
 370 LUSTRE_RO_ATTR(filesfree);
 371
 372 ssize_t conn_uuid_show(struct kobject *kobj, struct attribute *attr, char *buf)
 373 {
 374         struct obd_device *obd = container_of(kobj, struct obd_device,
 375                                               obd_kset.kobj);
 376         struct obd_import *imp;
 377         struct ptlrpc_connection *conn;
 378         ssize_t count;
 379
 380         with_imp_locked(obd, imp, count) {
 381                 conn = imp->imp_connection;
 382                 if (conn)
 383                         count = sprintf(buf, "%s\n", conn->c_remote_uuid.uuid);
 384                 else
 385                         count = sprintf(buf, "%s\n", "<none>");
 386         }
 387
 388         return count;
 389 }
 390 EXPORT_SYMBOL(conn_uuid_show);
 391
 392 int lprocfs_server_uuid_seq_show(struct seq_file *m, void *data)
 393 {
 394         struct obd_device *obd = data;
 395         struct obd_import *imp;
 396         const char *imp_state_name = NULL;
 397         int rc = 0;
 398
 399         LASSERT(obd != NULL);
 400         with_imp_locked(obd, imp, rc) {
 401                 imp_state_name = ptlrpc_import_state_name(imp->imp_state);
 402                 seq_printf(m, "%s\t%s%s\n", obd2cli_tgt(obd), imp_state_name,
 403                            imp->imp_deactive ? "\tDEACTIVATED" : "");
 404         }
 405
 406         return rc;
 407 }
 408 EXPORT_SYMBOL(lprocfs_server_uuid_seq_show);
 409
 410 /** add up per-cpu counters */
 411
 412 /**
 413  * Lock statistics structure for access, possibly only on this CPU.
 414  *
 415  * The statistics struct may be allocated with per-CPU structures for
 416  * efficient concurrent update (usually only on server-wide stats), or
 417  * as a single global struct (e.g. for per-client or per-job statistics),
 418  * so the required locking depends on the type of structure allocated.
 419  *
 420  * For per-CPU statistics, pin the thread to the current cpuid so that
 421  * will only access the statistics for that CPU.  If the stats structure
 422  * for the current CPU has not been allocated (or previously freed),
 423  * allocate it now.  The per-CPU statistics do not need locking since
 424  * the thread is pinned to the CPU during update.
 425  *
 426  * For global statistics, lock the stats structure to prevent concurrent update.
 427  *
 428  * \param[in] stats     statistics structure to lock
 429  * \param[in] opc       type of operation:
 430  *                      LPROCFS_GET_SMP_ID: "lock" and return current CPU index
 431  *                              for incrementing statistics for that CPU
 432  *                      LPROCFS_GET_NUM_CPU: "lock" and return number of used
 433  *                              CPU indices to iterate over all indices
 434  * \param[out] flags    CPU interrupt saved state for IRQ-safe locking
 435  *
 436  * \retval cpuid of current thread or number of allocated structs
 437  * \retval negative on error (only for opc LPROCFS_GET_SMP_ID + per-CPU stats)
 438  */
 439 int lprocfs_stats_lock(struct lprocfs_stats *stats,
 440                        enum lprocfs_stats_lock_ops opc,
 441                        unsigned long *flags)
 442 {
 443         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 444                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 445                         spin_lock_irqsave(&stats->ls_lock, *flags);
 446                 else
 447                         spin_lock(&stats->ls_lock);
 448                 return opc == LPROCFS_GET_NUM_CPU ? 1 : 0;
 449         }
 450
 451         switch (opc) {
 452         case LPROCFS_GET_SMP_ID: {
 453                 unsigned int cpuid = get_cpu();
 454
 455                 if (unlikely(!stats->ls_percpu[cpuid])) {
 456                         int rc = lprocfs_stats_alloc_one(stats, cpuid);
 457
 458                         if (rc < 0) {
 459                                 put_cpu();
 460                                 return rc;
 461                         }
 462                 }
 463                 return cpuid;
 464         }
 465         case LPROCFS_GET_NUM_CPU:
 466                 return stats->ls_biggest_alloc_num;
 467         default:
 468                 LBUG();
 469         }
 470 }
 471
 472 /**
 473  * Unlock statistics structure after access.
 474  *
 475  * Unlock the lock acquired via lprocfs_stats_lock() for global statistics,
 476  * or unpin this thread from the current cpuid for per-CPU statistics.
 477  *
 478  * This function must be called using the same arguments as used when calling
 479  * lprocfs_stats_lock() so that the correct operation can be performed.
 480  *
 481  * \param[in] stats     statistics structure to unlock
 482  * \param[in] opc       type of operation (current cpuid or number of structs)
 483  * \param[in] flags     CPU interrupt saved state for IRQ-safe locking
 484  */
 485 void lprocfs_stats_unlock(struct lprocfs_stats *stats,
 486                           enum lprocfs_stats_lock_ops opc,
 487                           unsigned long *flags)
 488 {
 489         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) {
 490                 if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
 491                         spin_unlock_irqrestore(&stats->ls_lock, *flags);
 492                 else
 493                         spin_unlock(&stats->ls_lock);
 494         } else if (opc == LPROCFS_GET_SMP_ID) {
 495                 put_cpu();
 496         }
 497 }
 498
 499 /** add up per-cpu counters */
 500 void lprocfs_stats_collect(struct lprocfs_stats *stats, int idx,
 501                            struct lprocfs_counter *cnt)
 502 {
 503         unsigned int num_entry;
 504         struct lprocfs_counter *percpu_cntr;
 505         int i;
 506         unsigned long flags = 0;
 507
 508         memset(cnt, 0, sizeof(*cnt));
 509
 510         if (!stats) {
 511                 /* set count to 1 to avoid divide-by-zero errs in callers */
 512                 cnt->lc_count = 1;
 513                 return;
 514         }
 515
 516         cnt->lc_min = LC_MIN_INIT;
 517
 518         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
 519
 520         for (i = 0; i < num_entry; i++) {
 521                 if (!stats->ls_percpu[i])
 522                         continue;
 523                 percpu_cntr = lprocfs_stats_counter_get(stats, i, idx);
 524
 525                 cnt->lc_count += percpu_cntr->lc_count;
 526                 cnt->lc_sum += percpu_cntr->lc_sum;
 527                 if (percpu_cntr->lc_min < cnt->lc_min)
 528                         cnt->lc_min = percpu_cntr->lc_min;
 529                 if (percpu_cntr->lc_max > cnt->lc_max)
 530                         cnt->lc_max = percpu_cntr->lc_max;
 531                 cnt->lc_sumsquare += percpu_cntr->lc_sumsquare;
 532         }
 533
 534         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
 535 }
 536
 537 static void obd_import_flags2str(struct obd_import *imp, struct seq_file *m)
 538 {
 539         bool first = true;
 540
 541         if (imp->imp_obd->obd_no_recov) {
 542                 seq_printf(m, "no_recov");
 543                 first = false;
 544         }
 545
 546         flag2str(imp, invalid);
 547         flag2str(imp, deactive);
 548         flag2str(imp, replayable);
 549         flag2str(imp, delayed_recovery);
 550         flag2str(imp, vbr_failed);
 551         flag2str(imp, pingable);
 552         flag2str(imp, resend_replay);
 553         flag2str(imp, no_pinger_recover);
 554         flag2str(imp, connect_tried);
 555 }
 556
 557 static const char *const obd_connect_names[] = {
 558         "read_only",                    /* 0x01 */
 559         "lov_index",                    /* 0x02 */
 560         "connect_from_mds",             /* 0x03 */
 561         "write_grant",                  /* 0x04 */
 562         "server_lock",                  /* 0x10 */
 563         "version",                      /* 0x20 */
 564         "request_portal",               /* 0x40 */
 565         "acl",                          /* 0x80 */
 566         "xattr",                        /* 0x100 */
 567         "create_on_write",              /* 0x200 */
 568         "truncate_lock",                /* 0x400 */
 569         "initial_transno",              /* 0x800 */
 570         "inode_bit_locks",              /* 0x1000 */
 571         "barrier",                      /* 0x2000 */
 572         "getattr_by_fid",               /* 0x4000 */
 573         "no_oh_for_devices",            /* 0x8000 */
 574         "remote_client",                /* 0x10000 */
 575         "remote_client_by_force",       /* 0x20000 */
 576         "max_byte_per_rpc",             /* 0x40000 */
 577         "64bit_qdata",                  /* 0x80000 */
 578         "mds_capability",               /* 0x100000 */
 579         "oss_capability",               /* 0x200000 */
 580         "early_lock_cancel",            /* 0x400000 */
 581         "som",                          /* 0x800000 */
 582         "adaptive_timeouts",            /* 0x1000000 */
 583         "lru_resize",                   /* 0x2000000 */
 584         "mds_mds_connection",           /* 0x4000000 */
 585         "real_conn",                    /* 0x8000000 */
 586         "change_qunit_size",            /* 0x10000000 */
 587         "alt_checksum_algorithm",       /* 0x20000000 */
 588         "fid_is_enabled",               /* 0x40000000 */
 589         "version_recovery",             /* 0x80000000 */
 590         "pools",                        /* 0x100000000 */
 591         "grant_shrink",                 /* 0x200000000 */
 592         "skip_orphan",                  /* 0x400000000 */
 593         "large_ea",                     /* 0x800000000 */
 594         "full20",                       /* 0x1000000000 */
 595         "layout_lock",                  /* 0x2000000000 */
 596         "64bithash",                    /* 0x4000000000 */
 597         "object_max_bytes",             /* 0x8000000000 */
 598         "imp_recov",                    /* 0x10000000000 */
 599         "jobstats",                     /* 0x20000000000 */
 600         "umask",                        /* 0x40000000000 */
 601         "einprogress",                  /* 0x80000000000 */
 602         "grant_param",                  /* 0x100000000000 */
 603         "flock_owner",                  /* 0x200000000000 */
 604         "lvb_type",                     /* 0x400000000000 */
 605         "nanoseconds_times",            /* 0x800000000000 */
 606         "lightweight_conn",             /* 0x1000000000000 */
 607         "short_io",                     /* 0x2000000000000 */
 608         "pingless",                     /* 0x4000000000000 */
 609         "flock_deadlock",               /* 0x8000000000000 */
 610         "disp_stripe",                  /* 0x10000000000000 */
 611         "open_by_fid",                  /* 0x20000000000000 */
 612         "lfsck",                        /* 0x40000000000000 */
 613         "unknown",                      /* 0x80000000000000 */
 614         "unlink_close",                 /* 0x100000000000000 */
 615         "multi_mod_rpcs",               /* 0x200000000000000 */
 616         "dir_stripe",                   /* 0x400000000000000 */
 617         "subtree",                      /* 0x800000000000000 */
 618         "lockahead",                    /* 0x1000000000000000 */
 619         "bulk_mbits",                   /* 0x2000000000000000 */
 620         "compact_obdo",                 /* 0x4000000000000000 */
 621         "second_flags",                 /* 0x8000000000000000 */
 622         /* ocd_connect_flags2 names */
 623         "file_secctx",                  /* 0x01 */
 624         "lockaheadv2",                  /* 0x02 */
 625         "dir_migrate",                  /* 0x04 */
 626         "sum_statfs",                   /* 0x08 */
 627         "overstriping",                 /* 0x10 */
 628         "flr",                          /* 0x20 */
 629         "wbc",                          /* 0x40 */
 630         "lock_convert",                 /* 0x80 */
 631         "archive_id_array",             /* 0x100 */
 632         "increasing_xid",               /* 0x200 */
 633         "selinux_policy",               /* 0x400 */
 634         "lsom",                         /* 0x800 */
 635         "pcc",                          /* 0x1000 */
 636         "crush",                        /* 0x2000 */
 637         "async_discard",                /* 0x4000 */
 638         "client_encryption",            /* 0x8000 */
 639         "fidmap",                       /* 0x10000 */
 640         "getattr_pfid",                 /* 0x20000 */
 641         "lseek",                        /* 0x40000 */
 642         "dom_lvb",                      /* 0x80000 */
 643         "reply_mbits",                  /* 0x100000 */
 644         "mode_convert",                 /* 0x200000 */
 645         "batch_rpc",                    /* 0x400000 */
 646         "pcc_ro",                       /* 0x800000 */
 647         "mne_nid_type",                 /* 0x1000000 */
 648         "lock_contend",                 /* 0x2000000 */
 649         "atomic_open_lock",             /* 0x4000000 */
 650         "name_encryption",              /* 0x8000000 */
 651         NULL
 652 };
 653
 654 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
 655                                const char *sep)
 656 {
 657         bool first = true;
 658         __u64 mask;
 659         int i;
 660
 661         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 662                 if (flags & mask) {
 663                         seq_printf(m, "%s%s",
 664                                    first ? "" : sep, obd_connect_names[i]);
 665                         first = false;
 666                 }
 667         }
 668
 669         if (flags & ~(mask - 1)) {
 670                 seq_printf(m, "%sunknown_%#llx",
 671                            first ? "" : sep, flags & ~(mask - 1));
 672                 first = false;
 673         }
 674
 675         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 676                 return;
 677
 678         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 679                 if (flags2 & mask) {
 680                         seq_printf(m, "%s%s",
 681                                    first ? "" : sep, obd_connect_names[i]);
 682                         first = false;
 683                 }
 684         }
 685
 686         if (flags2 & ~(mask - 1)) {
 687                 seq_printf(m, "%sunknown2_%#llx",
 688                            first ? "" : sep, flags2 & ~(mask - 1));
 689                 first = false;
 690         }
 691 }
 692 EXPORT_SYMBOL(obd_connect_seq_flags2str);
 693
 694 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 695                           const char *sep)
 696 {
 697         __u64 mask;
 698         int i, ret = 0;
 699
 700         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 701                 if (flags & mask)
 702                         ret += snprintf(page + ret, count - ret, "%s%s",
 703                                         ret ? sep : "", obd_connect_names[i]);
 704         }
 705
 706         if (flags & ~(mask - 1))
 707                 ret += snprintf(page + ret, count - ret,
 708                                 "%sunknown_%#llx",
 709                                 ret ? sep : "", flags & ~(mask - 1));
 710
 711         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 712                 return ret;
 713
 714         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 715                 if (flags2 & mask)
 716                         ret += snprintf(page + ret, count - ret, "%s%s",
 717                                         ret ? sep : "", obd_connect_names[i]);
 718         }
 719
 720         if (flags2 & ~(mask - 1))
 721                 ret += snprintf(page + ret, count - ret,
 722                                 "%sunknown2_%#llx",
 723                                 ret ? sep : "", flags2 & ~(mask - 1));
 724
 725         return ret;
 726 }
 727 EXPORT_SYMBOL(obd_connect_flags2str);
 728
 729 void
 730 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 731 {
 732         __u64 flags;
 733
 734         LASSERT(ocd != NULL);
 735         flags = ocd->ocd_connect_flags;
 736
 737         seq_printf(m, "    connect_data:\n"
 738                    "       flags: %#llx\n"
 739                    "       instance: %u\n",
 740                    ocd->ocd_connect_flags,
 741                    ocd->ocd_instance);
 742         if (flags & OBD_CONNECT_VERSION)
 743                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
 744                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
 745                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
 746                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
 747                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
 748         if (flags & OBD_CONNECT_MDS)
 749                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
 750         if (flags & OBD_CONNECT_GRANT)
 751                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
 752         if (flags & OBD_CONNECT_INDEX)
 753                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
 754         if (flags & OBD_CONNECT_BRW_SIZE)
 755                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
 756         if (flags & OBD_CONNECT_IBITS)
 757                 seq_printf(m, "       ibits_known: %#llx\n",
 758                            ocd->ocd_ibits_known);
 759         if (flags & OBD_CONNECT_GRANT_PARAM)
 760                 seq_printf(m, "       grant_block_size: %d\n"
 761                            "       grant_inode_size: %d\n"
 762                            "       grant_max_extent_size: %d\n"
 763                            "       grant_extent_tax: %d\n",
 764                            1 << ocd->ocd_grant_blkbits,
 765                            1 << ocd->ocd_grant_inobits,
 766                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
 767                            ocd->ocd_grant_tax_kb << 10);
 768         if (flags & OBD_CONNECT_TRANSNO)
 769                 seq_printf(m, "       first_transno: %#llx\n",
 770                            ocd->ocd_transno);
 771         if (flags & OBD_CONNECT_CKSUM)
 772                 seq_printf(m, "       cksum_types: %#x\n",
 773                            ocd->ocd_cksum_types);
 774         if (flags & OBD_CONNECT_MAX_EASIZE)
 775                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
 776         if (flags & OBD_CONNECT_MAXBYTES)
 777                 seq_printf(m, "       max_object_bytes: %llu\n",
 778                            ocd->ocd_maxbytes);
 779         if (flags & OBD_CONNECT_MULTIMODRPCS)
 780                 seq_printf(m, "       max_mod_rpcs: %hu\n",
 781                            ocd->ocd_maxmodrpcs);
 782 }
 783
 784 static void lprocfs_import_seq_show_locked(struct seq_file *m,
 785                                            struct obd_device *obd,
 786                                            struct obd_import *imp)
 787 {
 788         char nidstr[LNET_NIDSTR_SIZE];
 789         struct lprocfs_counter ret;
 790         struct lprocfs_counter_header *header;
 791         struct obd_import_conn *conn;
 792         struct obd_connect_data *ocd;
 793         int j;
 794         int k;
 795         int rw = 0;
 796
 797         ocd = &imp->imp_connect_data;
 798
 799         seq_printf(m, "import:\n"
 800                    "    name: %s\n"
 801                    "    target: %s\n"
 802                    "    state: %s\n"
 803                    "    connect_flags: [ ",
 804                    obd->obd_name,
 805                    obd2cli_tgt(obd),
 806                    ptlrpc_import_state_name(imp->imp_state));
 807         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
 808                                   imp->imp_connect_data.ocd_connect_flags2,
 809                                   ", ");
 810         seq_printf(m, " ]\n");
 811         obd_connect_data_seqprint(m, ocd);
 812         seq_printf(m, "    import_flags: [ ");
 813         obd_import_flags2str(imp, m);
 814
 815         seq_printf(m, " ]\n"
 816                    "    connection:\n"
 817                    "       failover_nids: [ ");
 818         spin_lock(&imp->imp_lock);
 819         j = 0;
 820         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 821                 libcfs_nidstr_r(&conn->oic_conn->c_peer.nid,
 822                                   nidstr, sizeof(nidstr));
 823                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
 824                 j++;
 825         }
 826         if (imp->imp_connection)
 827                 libcfs_nidstr_r(&imp->imp_connection->c_peer.nid,
 828                                   nidstr, sizeof(nidstr));
 829         else
 830                 strncpy(nidstr, "<none>", sizeof(nidstr));
 831         seq_printf(m, " ]\n"
 832                    "       current_connection: %s\n"
 833                    "       connection_attempts: %u\n"
 834                    "       generation: %u\n"
 835                    "       in-progress_invalidations: %u\n"
 836                    "       idle: %lld sec\n",
 837                    nidstr,
 838                    imp->imp_conn_cnt,
 839                    imp->imp_generation,
 840                    atomic_read(&imp->imp_inval_count),
 841                    ktime_get_real_seconds() - imp->imp_last_reply_time);
 842         spin_unlock(&imp->imp_lock);
 843
 844         if (!obd->obd_svc_stats)
 845                 return;
 846
 847         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
 848         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
 849         if (ret.lc_count != 0)
 850                 ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 851         else
 852                 ret.lc_sum = 0;
 853         seq_printf(m, "    rpcs:\n"
 854                    "       inflight: %u\n"
 855                    "       unregistering: %u\n"
 856                    "       timeouts: %u\n"
 857                    "       avg_waittime: %llu %s\n",
 858                    atomic_read(&imp->imp_inflight),
 859                    atomic_read(&imp->imp_unregistering),
 860                    atomic_read(&imp->imp_timeouts),
 861                    ret.lc_sum, header->lc_units);
 862
 863         k = 0;
 864         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
 865                 if (imp->imp_at.iat_portal[j] == 0)
 866                         break;
 867                 k = max_t(unsigned int, k,
 868                           at_get(&imp->imp_at.iat_service_estimate[j]));
 869         }
 870         seq_printf(m, "    service_estimates:\n"
 871                    "       services: %u sec\n"
 872                    "       network: %d sec\n",
 873                    k,
 874                    at_get(&imp->imp_at.iat_net_latency));
 875
 876         seq_printf(m, "    transactions:\n"
 877                    "       last_replay: %llu\n"
 878                    "       peer_committed: %llu\n"
 879                    "       last_checked: %llu\n",
 880                    imp->imp_last_replay_transno,
 881                    imp->imp_peer_committed_transno,
 882                    imp->imp_last_transno_checked);
 883
 884         /* avg data rates */
 885         for (rw = 0; rw <= 1; rw++) {
 886                 lprocfs_stats_collect(obd->obd_svc_stats,
 887                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
 888                                       &ret);
 889                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
 890                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 891                         seq_printf(m, "    %s_data_averages:\n"
 892                                    "       bytes_per_rpc: %llu\n",
 893                                    rw ? "write" : "read",
 894                                    ret.lc_sum);
 895                 }
 896                 k = (int)ret.lc_sum;
 897                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
 898                 header = &obd->obd_svc_stats->ls_cnt_header[j];
 899                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
 900                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
 901                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 902                         seq_printf(m, "       %s_per_rpc: %llu\n",
 903                                    header->lc_units, ret.lc_sum);
 904                         j = (int)ret.lc_sum;
 905                         if (j > 0)
 906                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
 907                                            k / j, (100 * k / j) % 100);
 908                 }
 909         }
 910 }
 911
 912 int lprocfs_import_seq_show(struct seq_file *m, void *data)
 913 {
 914         struct obd_device *obd = (struct obd_device *)data;
 915         struct obd_import *imp;
 916         int rv;
 917
 918         LASSERT(obd != NULL);
 919         with_imp_locked(obd, imp, rv)
 920                 lprocfs_import_seq_show_locked(m, obd, imp);
 921         return rv;
 922 }
 923 EXPORT_SYMBOL(lprocfs_import_seq_show);
 924
 925 int lprocfs_state_seq_show(struct seq_file *m, void *data)
 926 {
 927         struct obd_device *obd = (struct obd_device *)data;
 928         struct obd_import *imp;
 929         int j, k;
 930         int rc;
 931
 932         LASSERT(obd != NULL);
 933         with_imp_locked(obd, imp, rc) {
 934                 seq_printf(m, "current_state: %s\n",
 935                            ptlrpc_import_state_name(imp->imp_state));
 936                 seq_printf(m, "state_history:\n");
 937                 k = imp->imp_state_hist_idx;
 938                 for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
 939                         struct import_state_hist *ish =
 940                                 &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
 941                         if (ish->ish_state == 0)
 942                                 continue;
 943                         seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
 944                                    ptlrpc_import_state_name(ish->ish_state));
 945                 }
 946         }
 947
 948         return rc;
 949 }
 950 EXPORT_SYMBOL(lprocfs_state_seq_show);
 951
 952 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
 953 {
 954         int i;
 955         for (i = 0; i < AT_BINS; i++)
 956                 seq_printf(m, "%3u ", at->at_hist[i]);
 957         seq_printf(m, "\n");
 958         return 0;
 959 }
 960 EXPORT_SYMBOL(lprocfs_at_hist_helper);
 961
 962 /* See also ptlrpc_lprocfs_timeouts_show_seq */
 963 static void lprocfs_timeouts_seq_show_locked(struct seq_file *m,
 964                                              struct obd_device *obd,
 965                                              struct obd_import *imp)
 966 {
 967         timeout_t cur_timeout, worst_timeout;
 968         time64_t now, worst_timestamp;
 969         int i;
 970
 971         LASSERT(obd != NULL);
 972
 973         now = ktime_get_real_seconds();
 974
 975         /* Some network health info for kicks */
 976         seq_printf(m, "%-10s : %lld, %llds ago\n",
 977                    "last reply", (s64)imp->imp_last_reply_time,
 978                    (s64)(now - imp->imp_last_reply_time));
 979
 980         cur_timeout = at_get(&imp->imp_at.iat_net_latency);
 981         worst_timeout = imp->imp_at.iat_net_latency.at_worst_timeout_ever;
 982         worst_timestamp = imp->imp_at.iat_net_latency.at_worst_timestamp;
 983         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
 984                    "network", cur_timeout, worst_timeout, worst_timestamp,
 985                    now - worst_timestamp);
 986         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
 987
 988         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
 989                 struct adaptive_timeout *service_est;
 990
 991                 if (imp->imp_at.iat_portal[i] == 0)
 992                         break;
 993
 994                 service_est = &imp->imp_at.iat_service_estimate[i];
 995                 cur_timeout = at_get(service_est);
 996                 worst_timeout = service_est->at_worst_timeout_ever;
 997                 worst_timestamp = service_est->at_worst_timestamp;
 998                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
 999                            imp->imp_at.iat_portal[i], cur_timeout,
1000                            worst_timeout, worst_timestamp,
1001                            now - worst_timestamp);
1002                 lprocfs_at_hist_helper(m, service_est);
1003         }
1004 }
1005
1006 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1007 {
1008         struct obd_device *obd = (struct obd_device *)data;
1009         struct obd_import *imp;
1010         int rc;
1011
1012         with_imp_locked(obd, imp, rc)
1013                 lprocfs_timeouts_seq_show_locked(m, obd, imp);
1014         return rc;
1015 }
1016 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1017
1018 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1019 {
1020         struct obd_device *obd = data;
1021         __u64 flags;
1022         __u64 flags2;
1023         struct obd_import *imp;
1024         int rc;
1025
1026         with_imp_locked(obd, imp, rc) {
1027                 flags = imp->imp_connect_data.ocd_connect_flags;
1028                 flags2 = imp->imp_connect_data.ocd_connect_flags2;
1029                 seq_printf(m, "flags=%#llx\n", flags);
1030                 seq_printf(m, "flags2=%#llx\n", flags2);
1031                 obd_connect_seq_flags2str(m, flags, flags2, "\n");
1032                 seq_printf(m, "\n");
1033         }
1034
1035         return rc;
1036 }
1037 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1038
1039 static const struct attribute *obd_def_uuid_attrs[] = {
1040         &lustre_attr_uuid.attr,
1041         NULL,
1042 };
1043
1044 static const struct attribute *obd_def_attrs[] = {
1045         &lustre_attr_blocksize.attr,
1046         &lustre_attr_kbytestotal.attr,
1047         &lustre_attr_kbytesfree.attr,
1048         &lustre_attr_kbytesavail.attr,
1049         &lustre_attr_filestotal.attr,
1050         &lustre_attr_filesfree.attr,
1051         &lustre_attr_uuid.attr,
1052         NULL,
1053 };
1054
1055 static void obd_sysfs_release(struct kobject *kobj)
1056 {
1057         struct obd_device *obd = container_of(kobj, struct obd_device,
1058                                               obd_kset.kobj);
1059
1060         complete(&obd->obd_kobj_unregister);
1061 }
1062
1063 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1064 {
1065         struct ldebugfs_vars *debugfs_vars = NULL;
1066         int rc;
1067
1068         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1069                 return -ENODEV;
1070
1071         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1072         if (rc)
1073                 return rc;
1074
1075         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1076         obd->obd_ktype.release = obd_sysfs_release;
1077
1078         obd->obd_kset.kobj.parent = &obd->obd_type->typ_kobj;
1079         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1080         init_completion(&obd->obd_kobj_unregister);
1081         rc = kset_register(&obd->obd_kset);
1082         if (rc)
1083                 return rc;
1084
1085         if (uuid_only)
1086                 obd->obd_attrs = obd_def_uuid_attrs;
1087         else
1088                 obd->obd_attrs = obd_def_attrs;
1089
1090         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1091         if (rc) {
1092                 kset_unregister(&obd->obd_kset);
1093                 return rc;
1094         }
1095
1096         if (!obd->obd_type->typ_procroot)
1097                 debugfs_vars = obd->obd_debugfs_vars;
1098         obd->obd_debugfs_entry = debugfs_create_dir(
1099                 obd->obd_name, obd->obd_type->typ_debugfs_entry);
1100         ldebugfs_add_vars(obd->obd_debugfs_entry, debugfs_vars, obd);
1101
1102         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1103                 GOTO(already_registered, rc);
1104
1105         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1106                                                obd->obd_type->typ_procroot,
1107                                                obd->obd_vars, obd);
1108         if (IS_ERR(obd->obd_proc_entry)) {
1109                 rc = PTR_ERR(obd->obd_proc_entry);
1110                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1111                 obd->obd_proc_entry = NULL;
1112
1113                 debugfs_remove_recursive(obd->obd_debugfs_entry);
1114                 obd->obd_debugfs_entry = NULL;
1115
1116                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1117                 obd->obd_attrs = NULL;
1118                 kset_unregister(&obd->obd_kset);
1119                 return rc;
1120         }
1121 already_registered:
1122         return rc;
1123 }
1124 EXPORT_SYMBOL(lprocfs_obd_setup);
1125
1126 int lprocfs_obd_cleanup(struct obd_device *obd)
1127 {
1128         if (!obd)
1129                 return -EINVAL;
1130
1131         if (obd->obd_proc_exports_entry) {
1132                 /* Should be no exports left */
1133                 lprocfs_remove(&obd->obd_proc_exports_entry);
1134                 obd->obd_proc_exports_entry = NULL;
1135         }
1136
1137         if (obd->obd_proc_entry) {
1138                 lprocfs_remove(&obd->obd_proc_entry);
1139                 obd->obd_proc_entry = NULL;
1140         }
1141
1142         debugfs_remove_recursive(obd->obd_debugfs_entry);
1143         obd->obd_debugfs_entry = NULL;
1144
1145         /* obd device never allocated a kset */
1146         if (!obd->obd_kset.kobj.state_initialized)
1147                 return 0;
1148
1149         if (obd->obd_attrs) {
1150                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1151                 obd->obd_attrs = NULL;
1152         }
1153
1154         kset_unregister(&obd->obd_kset);
1155         wait_for_completion(&obd->obd_kobj_unregister);
1156         return 0;
1157 }
1158 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1159
1160 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1161 {
1162         struct lprocfs_counter *cntr;
1163         unsigned int percpusize;
1164         int rc = -ENOMEM;
1165         unsigned long flags = 0;
1166         int i;
1167
1168         LASSERT(stats->ls_percpu[cpuid] == NULL);
1169         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1170
1171         percpusize = lprocfs_stats_counter_size(stats);
1172         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1173         if (stats->ls_percpu[cpuid]) {
1174                 rc = 0;
1175                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1176                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1177                                 spin_lock_irqsave(&stats->ls_lock, flags);
1178                         else
1179                                 spin_lock(&stats->ls_lock);
1180                         if (stats->ls_biggest_alloc_num <= cpuid)
1181                                 stats->ls_biggest_alloc_num = cpuid + 1;
1182                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1183                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1184                         } else {
1185                                 spin_unlock(&stats->ls_lock);
1186                         }
1187                 }
1188                 /* initialize the ls_percpu[cpuid] non-zero counter */
1189                 for (i = 0; i < stats->ls_num; ++i) {
1190                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1191                         cntr->lc_min = LC_MIN_INIT;
1192                 }
1193         }
1194         return rc;
1195 }
1196
1197 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1198                                           enum lprocfs_stats_flags flags)
1199 {
1200         struct lprocfs_stats *stats;
1201         unsigned int num_entry;
1202         unsigned int percpusize = 0;
1203         int i;
1204
1205         if (num == 0)
1206                 return NULL;
1207
1208         if (lprocfs_no_percpu_stats != 0)
1209                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1210
1211         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1212                 num_entry = 1;
1213         else
1214                 num_entry = num_possible_cpus();
1215
1216         /* alloc percpu pointers for all possible cpu slots */
1217         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1218         if (!stats)
1219                 return NULL;
1220
1221         stats->ls_num = num;
1222         stats->ls_flags = flags;
1223         spin_lock_init(&stats->ls_lock);
1224
1225         /* alloc num of counter headers */
1226         CFS_ALLOC_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1227         if (!stats->ls_cnt_header)
1228                 goto fail;
1229
1230         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1231                 /* contains only one set counters */
1232                 percpusize = lprocfs_stats_counter_size(stats);
1233                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1234                 if (!stats->ls_percpu[0])
1235                         goto fail;
1236                 stats->ls_biggest_alloc_num = 1;
1237         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1238                 /* alloc all percpu data, currently only obd_memory use this */
1239                 for (i = 0; i < num_entry; ++i)
1240                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1241                                 goto fail;
1242         }
1243
1244         return stats;
1245
1246 fail:
1247         lprocfs_free_stats(&stats);
1248         return NULL;
1249 }
1250 EXPORT_SYMBOL(lprocfs_alloc_stats);
1251
1252 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1253 {
1254         struct lprocfs_stats *stats = *statsh;
1255         unsigned int num_entry;
1256         unsigned int percpusize;
1257         unsigned int i;
1258
1259         if (!stats || stats->ls_num == 0)
1260                 return;
1261         *statsh = NULL;
1262
1263         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1264                 num_entry = 1;
1265         else
1266                 num_entry = num_possible_cpus();
1267
1268         percpusize = lprocfs_stats_counter_size(stats);
1269         for (i = 0; i < num_entry; i++)
1270                 if (stats->ls_percpu[i])
1271                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1272         if (stats->ls_cnt_header)
1273                 CFS_FREE_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1274         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1275 }
1276 EXPORT_SYMBOL(lprocfs_free_stats);
1277
1278 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1279                             enum lprocfs_fields_flags field)
1280 {
1281         unsigned long flags = 0;
1282         unsigned int num_cpu;
1283         unsigned int i;
1284         u64 ret = 0;
1285
1286         LASSERT(stats);
1287
1288         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1289         for (i = 0; i < num_cpu; i++) {
1290                 struct lprocfs_counter *cntr;
1291
1292                 if (!stats->ls_percpu[i])
1293                         continue;
1294
1295                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1296                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1297                                            stats->ls_flags, field);
1298         }
1299         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1300         return ret;
1301 }
1302 EXPORT_SYMBOL(lprocfs_stats_collector);
1303
1304 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1305 {
1306         struct lprocfs_counter *percpu_cntr;
1307         int i;
1308         int j;
1309         unsigned int num_entry;
1310         unsigned long flags = 0;
1311
1312         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1313
1314         for (i = 0; i < num_entry; i++) {
1315                 if (!stats->ls_percpu[i])
1316                         continue;
1317                 for (j = 0; j < stats->ls_num; j++) {
1318                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1319                         percpu_cntr->lc_count           = 0;
1320                         percpu_cntr->lc_min             = LC_MIN_INIT;
1321                         percpu_cntr->lc_max             = 0;
1322                         percpu_cntr->lc_sumsquare       = 0;
1323                         percpu_cntr->lc_sum             = 0;
1324                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1325                                 percpu_cntr->lc_sum_irq = 0;
1326                 }
1327         }
1328
1329         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1330 }
1331 EXPORT_SYMBOL(lprocfs_clear_stats);
1332
1333 static ssize_t lprocfs_stats_seq_write(struct file *file,
1334                                        const char __user *buf,
1335                                        size_t len, loff_t *off)
1336 {
1337         struct seq_file *seq = file->private_data;
1338         struct lprocfs_stats *stats = seq->private;
1339
1340         lprocfs_clear_stats(stats);
1341
1342         return len;
1343 }
1344
1345 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1346 {
1347         struct lprocfs_stats *stats = p->private;
1348
1349         return (*pos < stats->ls_num) ? pos : NULL;
1350 }
1351
1352 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1353 {
1354 }
1355
1356 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1357 {
1358         (*pos)++;
1359
1360         return lprocfs_stats_seq_start(p, pos);
1361 }
1362
1363 void lprocfs_stats_header(struct seq_file *seq, ktime_t now, ktime_t ts_init,
1364                           int width, const char *colon, bool show_units)
1365 {
1366         const char *units = show_units ? " secs.nsecs" : "";
1367         struct timespec64 ts;
1368
1369         ts = ktime_to_timespec64(now);
1370         seq_printf(seq, "%-*s%s %llu.%09lu%s\n", width,
1371                    "snapshot_time", colon, (s64)ts.tv_sec, ts.tv_nsec, units);
1372         ts = ktime_to_timespec64(ts_init);
1373         seq_printf(seq, "%-*s%s %llu.%09lu%s\n", width,
1374                    "start_time", colon, (s64)ts.tv_sec, ts.tv_nsec, units);
1375         ts = ktime_to_timespec64(ktime_sub(now, ts_init));
1376         seq_printf(seq, "%-*s%s %llu.%09lu%s\n", width,
1377                    "elapsed_time", colon, (s64)ts.tv_sec, ts.tv_nsec, units);
1378 }
1379 EXPORT_SYMBOL(lprocfs_stats_header);
1380
1381 /* seq file export of one lprocfs counter */
1382 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1383 {
1384         struct lprocfs_stats *stats = p->private;
1385         struct lprocfs_counter_header *hdr;
1386         struct lprocfs_counter ctr;
1387         int idx = *(loff_t *)v;
1388
1389         if (idx == 0)
1390                 lprocfs_stats_header(p, ktime_get(), stats->ls_init, 25, "", 1);
1391
1392         hdr = &stats->ls_cnt_header[idx];
1393         lprocfs_stats_collect(stats, idx, &ctr);
1394
1395         if (ctr.lc_count == 0)
1396                 return 0;
1397
1398         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1399                    ctr.lc_count, hdr->lc_units);
1400
1401         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1402                 seq_printf(p, " %lld %lld %lld",
1403                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1404                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1405                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1406         }
1407         seq_putc(p, '\n');
1408         return 0;
1409 }
1410
1411 static const struct seq_operations lprocfs_stats_seq_sops = {
1412         .start  = lprocfs_stats_seq_start,
1413         .stop   = lprocfs_stats_seq_stop,
1414         .next   = lprocfs_stats_seq_next,
1415         .show   = lprocfs_stats_seq_show,
1416 };
1417
1418 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1419 {
1420         struct seq_file *seq;
1421         int rc;
1422
1423         rc = seq_open(file, &lprocfs_stats_seq_sops);
1424         if (rc)
1425                 return rc;
1426         seq = file->private_data;
1427         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1428         return 0;
1429 }
1430
1431 const struct file_operations ldebugfs_stats_seq_fops = {
1432         .owner   = THIS_MODULE,
1433         .open    = lprocfs_stats_seq_open,
1434         .read    = seq_read,
1435         .write   = lprocfs_stats_seq_write,
1436         .llseek  = seq_lseek,
1437         .release = lprocfs_seq_release,
1438 };
1439 EXPORT_SYMBOL(ldebugfs_stats_seq_fops);
1440
1441 static const struct proc_ops lprocfs_stats_seq_fops = {
1442         PROC_OWNER(THIS_MODULE)
1443         .proc_open      = lprocfs_stats_seq_open,
1444         .proc_read      = seq_read,
1445         .proc_write     = lprocfs_stats_seq_write,
1446         .proc_lseek     = seq_lseek,
1447         .proc_release   = lprocfs_seq_release,
1448 };
1449
1450 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1451                            struct lprocfs_stats *stats)
1452 {
1453         struct proc_dir_entry *entry;
1454         LASSERT(root != NULL);
1455
1456         entry = proc_create_data(name, 0644, root,
1457                                  &lprocfs_stats_seq_fops, stats);
1458         if (!entry)
1459                 return -ENOMEM;
1460         return 0;
1461 }
1462 EXPORT_SYMBOL(lprocfs_register_stats);
1463
1464 static const char *lprocfs_counter_config_units(const char *name,
1465                                          enum lprocfs_counter_config config)
1466 {
1467         const char *units;
1468
1469         switch (config & LPROCFS_TYPE_MASK) {
1470         default:
1471                 units = "reqs"; break;
1472         case LPROCFS_TYPE_BYTES:
1473                 units = "bytes"; break;
1474         case LPROCFS_TYPE_PAGES:
1475                 units = "pages"; break;
1476         case LPROCFS_TYPE_LOCKS:
1477                 units = "locks"; break;
1478         case LPROCFS_TYPE_LOCKSPS:
1479                 units = "locks/s"; break;
1480         case LPROCFS_TYPE_SECS:
1481                 units = "secs"; break;
1482         case LPROCFS_TYPE_USECS:
1483                 units = "usecs"; break;
1484         }
1485
1486         return units;
1487 }
1488
1489 void lprocfs_counter_init_units(struct lprocfs_stats *stats, int index,
1490                                 enum lprocfs_counter_config config,
1491                                 const char *name, const char *units)
1492 {
1493         struct lprocfs_counter_header *header;
1494         struct lprocfs_counter *percpu_cntr;
1495         unsigned long flags = 0;
1496         unsigned int i;
1497         unsigned int num_cpu;
1498
1499         LASSERT(stats != NULL);
1500
1501         header = &stats->ls_cnt_header[index];
1502         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1503                  index, name, units);
1504
1505         header->lc_config = config;
1506         header->lc_name = name;
1507         header->lc_units = units;
1508
1509         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1510         for (i = 0; i < num_cpu; ++i) {
1511                 if (!stats->ls_percpu[i])
1512                         continue;
1513                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1514                 percpu_cntr->lc_count           = 0;
1515                 percpu_cntr->lc_min             = LC_MIN_INIT;
1516                 percpu_cntr->lc_max             = 0;
1517                 percpu_cntr->lc_sumsquare       = 0;
1518                 percpu_cntr->lc_sum             = 0;
1519                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1520                         percpu_cntr->lc_sum_irq = 0;
1521         }
1522         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1523 }
1524 EXPORT_SYMBOL(lprocfs_counter_init_units);
1525
1526 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1527                           enum lprocfs_counter_config config,
1528                           const char *name)
1529 {
1530         lprocfs_counter_init_units(stats, index, config, name,
1531                                    lprocfs_counter_config_units(name, config));
1532 }
1533 EXPORT_SYMBOL(lprocfs_counter_init);
1534
1535 static const char * const mps_stats[] = {
1536         [LPROC_MD_CLOSE]                = "close",
1537         [LPROC_MD_CREATE]               = "create",
1538         [LPROC_MD_ENQUEUE]              = "enqueue",
1539         [LPROC_MD_GETATTR]              = "getattr",
1540         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1541         [LPROC_MD_LINK]                 = "link",
1542         [LPROC_MD_RENAME]               = "rename",
1543         [LPROC_MD_SETATTR]              = "setattr",
1544         [LPROC_MD_FSYNC]                = "fsync",
1545         [LPROC_MD_READ_PAGE]            = "read_page",
1546         [LPROC_MD_UNLINK]               = "unlink",
1547         [LPROC_MD_SETXATTR]             = "setxattr",
1548         [LPROC_MD_GETXATTR]             = "getxattr",
1549         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1550         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1551 };
1552
1553 int lprocfs_alloc_md_stats(struct obd_device *obd,
1554                            unsigned int num_private_stats)
1555 {
1556         struct lprocfs_stats *stats;
1557         unsigned int num_stats;
1558         int rc, i;
1559
1560         /*
1561          * TODO Ensure that this function is only used where
1562          * appropriate by adding an assertion to the effect that
1563          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1564          * because mdt_procfs_init() uses this function to allocate
1565          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1566          * mdt layer does not use the md_ops interface. This is
1567          * confusing and a waste of memory. See LU-2484.
1568          */
1569         LASSERT(obd->obd_proc_entry != NULL);
1570         LASSERT(obd->obd_md_stats == NULL);
1571
1572         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1573         stats = lprocfs_alloc_stats(num_stats, 0);
1574         if (!stats)
1575                 return -ENOMEM;
1576
1577         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1578                 lprocfs_counter_init(stats, i, LPROCFS_TYPE_REQS,
1579                                      mps_stats[i]);
1580                 if (!stats->ls_cnt_header[i].lc_name) {
1581                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1582                                i);
1583                         LBUG();
1584                 }
1585         }
1586
1587         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1588         if (rc < 0) {
1589                 lprocfs_free_stats(&stats);
1590         } else {
1591                 obd->obd_md_stats = stats;
1592         }
1593
1594         return rc;
1595 }
1596 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1597
1598 void lprocfs_free_md_stats(struct obd_device *obd)
1599 {
1600         struct lprocfs_stats *stats = obd->obd_md_stats;
1601
1602         if (stats) {
1603                 obd->obd_md_stats = NULL;
1604                 lprocfs_free_stats(&stats);
1605         }
1606 }
1607 EXPORT_SYMBOL(lprocfs_free_md_stats);
1608
1609 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1610 {
1611         lprocfs_counter_init(ldlm_stats, LDLM_ENQUEUE - LDLM_FIRST_OPC,
1612                              LPROCFS_TYPE_REQS, "ldlm_enqueue");
1613         lprocfs_counter_init(ldlm_stats, LDLM_CONVERT - LDLM_FIRST_OPC,
1614                              LPROCFS_TYPE_REQS, "ldlm_convert");
1615         lprocfs_counter_init(ldlm_stats, LDLM_CANCEL - LDLM_FIRST_OPC,
1616                              LPROCFS_TYPE_REQS, "ldlm_cancel");
1617         lprocfs_counter_init(ldlm_stats, LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1618                              LPROCFS_TYPE_REQS, "ldlm_bl_callback");
1619         lprocfs_counter_init(ldlm_stats, LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1620                              LPROCFS_TYPE_REQS, "ldlm_cp_callback");
1621         lprocfs_counter_init(ldlm_stats, LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1622                              LPROCFS_TYPE_REQS, "ldlm_gl_callback");
1623 }
1624 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1625
1626 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1627                           struct lprocfs_counter_header *header,
1628                           enum lprocfs_stats_flags flags,
1629                           enum lprocfs_fields_flags field)
1630 {
1631         __s64 ret = 0;
1632
1633         if (!lc || !header)
1634                 RETURN(0);
1635
1636         switch (field) {
1637                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1638                         ret = header->lc_config;
1639                         break;
1640                 case LPROCFS_FIELDS_FLAGS_SUM:
1641                         ret = lc->lc_sum;
1642                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1643                                 ret += lc->lc_sum_irq;
1644                         break;
1645                 case LPROCFS_FIELDS_FLAGS_MIN:
1646                         ret = lc->lc_min;
1647                         break;
1648                 case LPROCFS_FIELDS_FLAGS_MAX:
1649                         ret = lc->lc_max;
1650                         break;
1651                 case LPROCFS_FIELDS_FLAGS_AVG:
1652                         ret = div64_u64((flags & LPROCFS_STATS_FLAG_IRQ_SAFE ?
1653                                          lc->lc_sum_irq : 0) + lc->lc_sum,
1654                                         lc->lc_count);
1655                         break;
1656                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1657                         ret = lc->lc_sumsquare;
1658                         break;
1659                 case LPROCFS_FIELDS_FLAGS_COUNT:
1660                         ret = lc->lc_count;
1661                         break;
1662                 default:
1663                         break;
1664         };
1665         RETURN(ret);
1666 }
1667 EXPORT_SYMBOL(lprocfs_read_helper);
1668
1669 /**
1670  * string_to_size - convert ASCII string representing a numerical
1671  *                  value with optional units to 64-bit binary value
1672  *
1673  * @size:       The numerical value extract out of @buffer
1674  * @buffer:     passed in string to parse
1675  * @count:      length of the @buffer
1676  *
1677  * This function returns a 64-bit binary value if @buffer contains a valid
1678  * numerical string. The string is parsed to 3 significant figures after
1679  * the decimal point. Support the string containing an optional units at
1680  * the end which can be base 2 or base 10 in value. If no units are given
1681  * the string is assumed to just a numerical value.
1682  *
1683  * Returns:     @count if the string is successfully parsed,
1684  *              -errno on invalid input strings. Error values:
1685  *
1686  *  - ``-EINVAL``: @buffer is not a proper numerical string
1687  *  - ``-EOVERFLOW``: results does not fit into 64 bits.
1688  *  - ``-E2BIG ``: @buffer is too large (not a valid number)
1689  */
1690 int string_to_size(u64 *size, const char *buffer, size_t count)
1691 {
1692         /* For string_get_size() it can support values above exabytes,
1693          * (ZiB, YiB) due to breaking the return value into a size and
1694          * bulk size to avoid 64 bit overflow. We don't break the size
1695          * up into block size units so we don't support ZiB or YiB.
1696          */
1697         static const char *const units_10[] = {
1698                 "kB", "MB", "GB", "TB", "PB", "EB",
1699         };
1700         static const char *const units_2[] = {
1701                 "K",  "M",  "G",  "T",  "P",  "E",
1702         };
1703         static const char *const *const units_str[] = {
1704                 [STRING_UNITS_2] = units_2,
1705                 [STRING_UNITS_10] = units_10,
1706         };
1707         static const unsigned int coeff[] = {
1708                 [STRING_UNITS_10] = 1000,
1709                 [STRING_UNITS_2] = 1024,
1710         };
1711         enum string_size_units unit = STRING_UNITS_2;
1712         u64 whole, blk_size = 1;
1713         char kernbuf[22], *end;
1714         size_t len = count;
1715         int rc;
1716         int i;
1717
1718         if (count >= sizeof(kernbuf)) {
1719                 CERROR("count %zd > buffer %zd\n", count, sizeof(kernbuf));
1720                 return -E2BIG;
1721         }
1722
1723         *size = 0;
1724         /* The "iB" suffix is optionally allowed for indicating base-2 numbers.
1725          * If suffix is only "B" and not "iB" then we treat it as base-10.
1726          */
1727         end = strstr(buffer, "B");
1728         if (end && *(end - 1) != 'i')
1729                 unit = STRING_UNITS_10;
1730
1731         i = unit == STRING_UNITS_2 ? ARRAY_SIZE(units_2) - 1 :
1732                                      ARRAY_SIZE(units_10) - 1;
1733         do {
1734                 end = strnstr(buffer, units_str[unit][i], count);
1735                 if (end) {
1736                         for (; i >= 0; i--)
1737                                 blk_size *= coeff[unit];
1738                         len = end - buffer;
1739                         break;
1740                 }
1741         } while (i--);
1742
1743         /* as 'B' is a substring of all units, we need to handle it
1744          * separately.
1745          */
1746         if (!end) {
1747                 /* 'B' is only acceptable letter at this point */
1748                 end = strnchr(buffer, count, 'B');
1749                 if (end) {
1750                         len = end - buffer;
1751
1752                         if (count - len > 2 ||
1753                             (count - len == 2 && strcmp(end, "B\n") != 0)) {
1754                                 CDEBUG(D_INFO, "unknown suffix '%s'\n", buffer);
1755                                 return -EINVAL;
1756                         }
1757                 }
1758                 /* kstrtoull will error out if it has non digits */
1759                 goto numbers_only;
1760         }
1761
1762         end = strnchr(buffer, count, '.');
1763         if (end) {
1764                 /* need to limit 3 decimal places */
1765                 char rem[4] = "000";
1766                 u64 frac = 0;
1767                 size_t off;
1768
1769                 len = end - buffer;
1770                 end++;
1771
1772                 /* limit to 3 decimal points */
1773                 off = min_t(size_t, 3, strspn(end, "0123456789"));
1774                 /* need to limit frac_d to a u32 */
1775                 memcpy(rem, end, off);
1776                 rc = kstrtoull(rem, 10, &frac);
1777                 if (rc)
1778                         return rc;
1779
1780                 if (fls64(frac) + fls64(blk_size) - 1 > 64)
1781                         return -EOVERFLOW;
1782
1783                 frac *= blk_size;
1784                 do_div(frac, 1000);
1785                 *size += frac;
1786         }
1787 numbers_only:
1788         snprintf(kernbuf, sizeof(kernbuf), "%.*s", (int)len, buffer);
1789         rc = kstrtoull(kernbuf, 10, &whole);
1790         if (rc)
1791                 return rc;
1792
1793         if (whole != 0 && fls64(whole) + fls64(blk_size) - 1 > 64)
1794                 return -EOVERFLOW;
1795
1796         *size += whole * blk_size;
1797
1798         return count;
1799 }
1800 EXPORT_SYMBOL(string_to_size);
1801
1802 /**
1803  * sysfs_memparse - parse a ASCII string to 64-bit binary value,
1804  *                  with optional units
1805  *
1806  * @buffer:     kernel pointer to input string
1807  * @count:      number of bytes in the input @buffer
1808  * @val:        (output) binary value returned to caller
1809  * @defunit:    default unit suffix to use if none is provided
1810  *
1811  * Parses a string into a number. The number stored at @buffer is
1812  * potentially suffixed with K, M, G, T, P, E. Besides these other
1813  * valid suffix units are shown in the string_to_size() function.
1814  * If the string lacks a suffix then the defunit is used. The defunit
1815  * should be given as a binary unit (e.g. MiB) as that is the standard
1816  * for tunables in Lustre. If no unit suffix is given (e.g. 'G'), then
1817  * it is assumed to be in binary units.
1818  *
1819  * Returns:     0 on success or -errno on failure.
1820  */
1821 int sysfs_memparse(const char *buffer, size_t count, u64 *val,
1822                    const char *defunit)
1823 {
1824         const char *param = buffer;
1825         char tmp_buf[23];
1826         int rc;
1827
1828         count = strlen(buffer);
1829         while (count > 0 && isspace(buffer[count - 1]))
1830                 count--;
1831
1832         if (!count)
1833                 RETURN(-EINVAL);
1834
1835         /* If there isn't already a unit on this value, append @defunit.
1836          * Units of 'B' don't affect the value, so don't bother adding.
1837          */
1838         if (!isalpha(buffer[count - 1]) && defunit[0] != 'B') {
1839                 if (count + 3 >= sizeof(tmp_buf)) {
1840                         CERROR("count %zd > size %zd\n", count, sizeof(param));
1841                         RETURN(-E2BIG);
1842                 }
1843
1844                 scnprintf(tmp_buf, sizeof(tmp_buf), "%.*s%s", (int)count,
1845                           buffer, defunit);
1846                 param = tmp_buf;
1847                 count = strlen(param);
1848         }
1849
1850         rc = string_to_size(val, param, count);
1851
1852         return rc < 0 ? rc : 0;
1853 }
1854 EXPORT_SYMBOL(sysfs_memparse);
1855
1856 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
1857 {
1858         size_t l2;
1859
1860         l2 = strlen(s2);
1861         if (!l2)
1862                 return (char *)s1;
1863         while (len >= l2) {
1864                 len--;
1865                 if (!memcmp(s1, s2, l2))
1866                         return (char *)s1;
1867                 s1++;
1868         }
1869         return NULL;
1870 }
1871 EXPORT_SYMBOL(lprocfs_strnstr);
1872
1873 /**
1874  * Find the string \a name in the input \a buffer, and return a pointer to the
1875  * value immediately following \a name, reducing \a count appropriately.
1876  * If \a name is not found the original \a buffer is returned.
1877  */
1878 char *lprocfs_find_named_value(const char *buffer, const char *name,
1879                                 size_t *count)
1880 {
1881         char *val;
1882         size_t buflen = *count;
1883
1884         /* there is no strnstr() in rhel5 and ubuntu kernels */
1885         val = lprocfs_strnstr(buffer, name, buflen);
1886         if (!val)
1887                 return (char *)buffer;
1888
1889         val += strlen(name);                             /* skip prefix */
1890         while (val < buffer + buflen && isspace(*val)) /* skip separator */
1891                 val++;
1892
1893         *count = 0;
1894         while (val < buffer + buflen && isalnum(*val)) {
1895                 ++*count;
1896                 ++val;
1897         }
1898
1899         return val - *count;
1900 }
1901 EXPORT_SYMBOL(lprocfs_find_named_value);
1902
1903 int lprocfs_seq_create(struct proc_dir_entry *parent,
1904                        const char *name,
1905                        mode_t mode,
1906                        const struct proc_ops *seq_fops,
1907                        void *data)
1908 {
1909         struct proc_dir_entry *entry;
1910         ENTRY;
1911
1912         /* Disallow secretly (un)writable entries. */
1913         LASSERT(!seq_fops->proc_write == !(mode & 0222));
1914
1915         entry = proc_create_data(name, mode, parent, seq_fops, data);
1916
1917         if (!entry)
1918                 RETURN(-ENOMEM);
1919
1920         RETURN(0);
1921 }
1922 EXPORT_SYMBOL(lprocfs_seq_create);
1923
1924 int lprocfs_obd_seq_create(struct obd_device *obd,
1925                            const char *name,
1926                            mode_t mode,
1927                            const struct proc_ops *seq_fops,
1928                            void *data)
1929 {
1930         return lprocfs_seq_create(obd->obd_proc_entry, name,
1931                                   mode, seq_fops, data);
1932 }
1933 EXPORT_SYMBOL(lprocfs_obd_seq_create);
1934
1935 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
1936 {
1937         if (value >= OBD_HIST_MAX)
1938                 value = OBD_HIST_MAX - 1;
1939
1940         spin_lock(&oh->oh_lock);
1941         oh->oh_buckets[value]++;
1942         spin_unlock(&oh->oh_lock);
1943 }
1944 EXPORT_SYMBOL(lprocfs_oh_tally);
1945
1946 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
1947 {
1948         unsigned int val = 0;
1949
1950         if (likely(value != 0))
1951                 val = min(fls(value - 1), OBD_HIST_MAX);
1952
1953         lprocfs_oh_tally(oh, val);
1954 }
1955 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
1956
1957 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
1958 {
1959         unsigned long ret = 0;
1960         int i;
1961
1962         for (i = 0; i < OBD_HIST_MAX; i++)
1963                 ret +=  oh->oh_buckets[i];
1964         return ret;
1965 }
1966 EXPORT_SYMBOL(lprocfs_oh_sum);
1967
1968 void lprocfs_oh_clear(struct obd_histogram *oh)
1969 {
1970         spin_lock(&oh->oh_lock);
1971         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
1972         spin_unlock(&oh->oh_lock);
1973 }
1974 EXPORT_SYMBOL(lprocfs_oh_clear);
1975
1976 void lprocfs_oh_tally_pcpu(struct obd_hist_pcpu *oh,
1977                            unsigned int value)
1978 {
1979         if (value >= OBD_HIST_MAX)
1980                 value = OBD_HIST_MAX - 1;
1981
1982         percpu_counter_inc(&oh->oh_pc_buckets[value]);
1983 }
1984 EXPORT_SYMBOL(lprocfs_oh_tally_pcpu);
1985
1986 void lprocfs_oh_tally_log2_pcpu(struct obd_hist_pcpu *oh,
1987                                 unsigned int value)
1988 {
1989         unsigned int val = 0;
1990
1991         if (likely(value != 0))
1992                 val = min(fls(value - 1), OBD_HIST_MAX);
1993
1994         lprocfs_oh_tally_pcpu(oh, val);
1995 }
1996 EXPORT_SYMBOL(lprocfs_oh_tally_log2_pcpu);
1997
1998 unsigned long lprocfs_oh_counter_pcpu(struct obd_hist_pcpu *oh,
1999                                       unsigned int value)
2000 {
2001         return percpu_counter_sum(&oh->oh_pc_buckets[value]);
2002 }
2003 EXPORT_SYMBOL(lprocfs_oh_counter_pcpu);
2004
2005 unsigned long lprocfs_oh_sum_pcpu(struct obd_hist_pcpu *oh)
2006 {
2007         unsigned long ret = 0;
2008         int i;
2009
2010         for (i = 0; i < OBD_HIST_MAX; i++)
2011                 ret += percpu_counter_sum(&oh->oh_pc_buckets[i]);
2012
2013         return ret;
2014 }
2015 EXPORT_SYMBOL(lprocfs_oh_sum_pcpu);
2016
2017 int lprocfs_oh_alloc_pcpu(struct obd_hist_pcpu *oh)
2018 {
2019         int i, rc;
2020
2021         if (oh->oh_initialized)
2022                 return 0;
2023
2024         for (i = 0; i < OBD_HIST_MAX; i++) {
2025                 rc = percpu_counter_init(&oh->oh_pc_buckets[i], 0, GFP_KERNEL);
2026                 if (rc)
2027                         goto out;
2028         }
2029
2030         oh->oh_initialized = true;
2031
2032         return 0;
2033
2034 out:
2035         for (i--; i >= 0; i--)
2036                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2037
2038         return rc;
2039 }
2040 EXPORT_SYMBOL(lprocfs_oh_alloc_pcpu);
2041
2042 void lprocfs_oh_clear_pcpu(struct obd_hist_pcpu *oh)
2043 {
2044         int i;
2045
2046         for (i = 0; i < OBD_HIST_MAX; i++)
2047                 percpu_counter_set(&oh->oh_pc_buckets[i], 0);
2048 }
2049 EXPORT_SYMBOL(lprocfs_oh_clear_pcpu);
2050
2051 void lprocfs_oh_release_pcpu(struct obd_hist_pcpu *oh)
2052 {
2053         int i;
2054
2055         if (!oh->oh_initialized)
2056                 return;
2057
2058         for (i = 0; i < OBD_HIST_MAX; i++)
2059                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2060
2061         oh->oh_initialized = false;
2062 }
2063 EXPORT_SYMBOL(lprocfs_oh_release_pcpu);
2064
2065 ssize_t lustre_attr_show(struct kobject *kobj,
2066                          struct attribute *attr, char *buf)
2067 {
2068         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2069
2070         return a->show ? a->show(kobj, attr, buf) : 0;
2071 }
2072 EXPORT_SYMBOL_GPL(lustre_attr_show);
2073
2074 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2075                           const char *buf, size_t len)
2076 {
2077         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2078
2079         return a->store ? a->store(kobj, attr, buf, len) : len;
2080 }
2081 EXPORT_SYMBOL_GPL(lustre_attr_store);
2082
2083 const struct sysfs_ops lustre_sysfs_ops = {
2084         .show  = lustre_attr_show,
2085         .store = lustre_attr_store,
2086 };
2087 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2088
2089 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2090 {
2091         struct obd_device *obd = data;
2092         struct client_obd *cli = &obd->u.cli;
2093
2094         spin_lock(&cli->cl_loi_list_lock);
2095         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2096         spin_unlock(&cli->cl_loi_list_lock);
2097         return 0;
2098 }
2099 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2100
2101 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2102                                                 const char __user *buffer,
2103                                                 size_t count, loff_t *off)
2104 {
2105         struct seq_file *m = file->private_data;
2106         struct obd_device *obd = m->private;
2107         struct client_obd *cli = &obd->u.cli;
2108         struct obd_import *imp;
2109         struct obd_connect_data *ocd;
2110         int chunk_mask, rc;
2111         char kernbuf[22];
2112         u64 val;
2113
2114         if (count > sizeof(kernbuf) - 1)
2115                 return -EINVAL;
2116
2117         if (copy_from_user(kernbuf, buffer, count))
2118                 return -EFAULT;
2119
2120         kernbuf[count] = '\0';
2121
2122         rc = sysfs_memparse(kernbuf, count, &val, "B");
2123         if (rc)
2124                 return rc;
2125
2126         /* if the max_pages is specified in bytes, convert to pages */
2127         if (val >= ONE_MB_BRW_SIZE)
2128                 val >>= PAGE_SHIFT;
2129
2130         with_imp_locked(obd, imp, rc) {
2131                 ocd = &imp->imp_connect_data;
2132                 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2133                 /* max_pages_per_rpc must be chunk aligned */
2134                 val = (val + ~chunk_mask) & chunk_mask;
2135                 if (val == 0 || (ocd->ocd_brw_size != 0 &&
2136                                  val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2137                         rc = -ERANGE;
2138                 } else {
2139                         spin_lock(&cli->cl_loi_list_lock);
2140                         cli->cl_max_pages_per_rpc = val;
2141                         client_adjust_max_dirty(cli);
2142                         spin_unlock(&cli->cl_loi_list_lock);
2143                 }
2144         }
2145
2146         return rc ?: count;
2147 }
2148 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2149
2150 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2151                             char *buf)
2152 {
2153         struct obd_device *obd = container_of(kobj, struct obd_device,
2154                                               obd_kset.kobj);
2155         struct client_obd *cli = &obd->u.cli;
2156         int rc;
2157
2158         spin_lock(&cli->cl_loi_list_lock);
2159         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2160         spin_unlock(&cli->cl_loi_list_lock);
2161         return rc;
2162 }
2163 EXPORT_SYMBOL(short_io_bytes_show);
2164
2165 /* Used to catch people who think they're specifying pages. */
2166 #define MIN_SHORT_IO_BYTES 64U
2167
2168 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2169                              const char *buffer, size_t count)
2170 {
2171         struct obd_device *obd = container_of(kobj, struct obd_device,
2172                                               obd_kset.kobj);
2173         struct client_obd *cli = &obd->u.cli;
2174         u64 val;
2175         int rc;
2176
2177         if (strcmp(buffer, "-1") == 0) {
2178                 val = OBD_DEF_SHORT_IO_BYTES;
2179         } else {
2180                 rc = sysfs_memparse(buffer, count, &val, "B");
2181                 if (rc)
2182                         GOTO(out, rc);
2183         }
2184
2185         if (val && (val < MIN_SHORT_IO_BYTES || val > LNET_MTU))
2186                 GOTO(out, rc = -ERANGE);
2187
2188         rc = count;
2189
2190         spin_lock(&cli->cl_loi_list_lock);
2191         cli->cl_max_short_io_bytes = min_t(u64, val, OST_MAX_SHORT_IO_BYTES);
2192         spin_unlock(&cli->cl_loi_list_lock);
2193
2194 out:
2195         return rc;
2196 }
2197 EXPORT_SYMBOL(short_io_bytes_store);
2198
2199 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2200                            struct root_squash_info *squash, char *name)
2201 {
2202         int rc;
2203         char kernbuf[64], *tmp, *errmsg;
2204         unsigned long uid, gid;
2205         ENTRY;
2206
2207         if (count >= sizeof(kernbuf)) {
2208                 errmsg = "string too long";
2209                 GOTO(failed_noprint, rc = -EINVAL);
2210         }
2211         if (copy_from_user(kernbuf, buffer, count)) {
2212                 errmsg = "bad address";
2213                 GOTO(failed_noprint, rc = -EFAULT);
2214         }
2215         kernbuf[count] = '\0';
2216
2217         /* look for uid gid separator */
2218         tmp = strchr(kernbuf, ':');
2219         if (!tmp) {
2220                 errmsg = "needs uid:gid format";
2221                 GOTO(failed, rc = -EINVAL);
2222         }
2223         *tmp = '\0';
2224         tmp++;
2225
2226         /* parse uid */
2227         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2228                 errmsg = "bad uid";
2229                 GOTO(failed, rc = -EINVAL);
2230         }
2231
2232         /* parse gid */
2233         if (kstrtoul(tmp, 0, &gid) != 0) {
2234                 errmsg = "bad gid";
2235                 GOTO(failed, rc = -EINVAL);
2236         }
2237
2238         squash->rsi_uid = uid;
2239         squash->rsi_gid = gid;
2240
2241         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2242                       name, squash->rsi_uid, squash->rsi_gid);
2243         RETURN(count);
2244
2245 failed:
2246         if (tmp) {
2247                 tmp--;
2248                 *tmp = ':';
2249         }
2250         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2251               name, kernbuf, errmsg, rc);
2252         RETURN(rc);
2253 failed_noprint:
2254         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2255               name, errmsg, rc);
2256         RETURN(rc);
2257 }
2258 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2259
2260
2261 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2262                              struct root_squash_info *squash, char *name)
2263 {
2264         int rc;
2265         char *kernbuf = NULL;
2266         char *errmsg;
2267         LIST_HEAD(tmp);
2268         int len = count;
2269         ENTRY;
2270
2271         if (count > 4096) {
2272                 errmsg = "string too long";
2273                 GOTO(failed, rc = -EINVAL);
2274         }
2275
2276         OBD_ALLOC(kernbuf, count + 1);
2277         if (!kernbuf) {
2278                 errmsg = "no memory";
2279                 GOTO(failed, rc = -ENOMEM);
2280         }
2281         if (copy_from_user(kernbuf, buffer, count)) {
2282                 errmsg = "bad address";
2283                 GOTO(failed, rc = -EFAULT);
2284         }
2285         kernbuf[count] = '\0';
2286
2287         if (count > 0 && kernbuf[count - 1] == '\n')
2288                 len = count - 1;
2289
2290         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2291             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2292                 /* empty string is special case */
2293                 spin_lock(&squash->rsi_lock);
2294                 if (!list_empty(&squash->rsi_nosquash_nids))
2295                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2296                 spin_unlock(&squash->rsi_lock);
2297                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2298                 OBD_FREE(kernbuf, count + 1);
2299                 RETURN(count);
2300         }
2301
2302         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2303                 errmsg = "can't parse";
2304                 GOTO(failed, rc = -EINVAL);
2305         }
2306         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2307                       name, kernbuf);
2308         OBD_FREE(kernbuf, count + 1);
2309         kernbuf = NULL;
2310
2311         spin_lock(&squash->rsi_lock);
2312         if (!list_empty(&squash->rsi_nosquash_nids))
2313                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2314         list_splice(&tmp, &squash->rsi_nosquash_nids);
2315         spin_unlock(&squash->rsi_lock);
2316
2317         RETURN(count);
2318
2319 failed:
2320         if (kernbuf) {
2321                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2322                       name, kernbuf, errmsg, rc);
2323                 OBD_FREE(kernbuf, count + 1);
2324         } else {
2325                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2326                       name, errmsg, rc);
2327         }
2328         RETURN(rc);
2329 }
2330 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2331
2332 #endif /* CONFIG_PROC_FS*/