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         /* flags names  */
 559         "read_only",
 560         "lov_index",
 561         "connect_from_mds",
 562         "write_grant",
 563         "server_lock",
 564         "version",
 565         "request_portal",
 566         "acl",
 567         "xattr",
 568         "create_on_write",
 569         "truncate_lock",
 570         "initial_transno",
 571         "inode_bit_locks",
 572         "barrier",
 573         "getattr_by_fid",
 574         "no_oh_for_devices",
 575         "remote_client",
 576         "remote_client_by_force",
 577         "max_byte_per_rpc",
 578         "64bit_qdata",
 579         "mds_capability",
 580         "oss_capability",
 581         "early_lock_cancel",
 582         "som",
 583         "adaptive_timeouts",
 584         "lru_resize",
 585         "mds_mds_connection",
 586         "real_conn",
 587         "change_qunit_size",
 588         "alt_checksum_algorithm",
 589         "fid_is_enabled",
 590         "version_recovery",
 591         "pools",
 592         "grant_shrink",
 593         "skip_orphan",
 594         "large_ea",
 595         "full20",
 596         "layout_lock",
 597         "64bithash",
 598         "object_max_bytes",
 599         "imp_recov",
 600         "jobstats",
 601         "umask",
 602         "einprogress",
 603         "grant_param",
 604         "flock_owner",
 605         "lvb_type",
 606         "nanoseconds_times",
 607         "lightweight_conn",
 608         "short_io",
 609         "pingless",
 610         "flock_deadlock",
 611         "disp_stripe",
 612         "open_by_fid",
 613         "lfsck",
 614         "unknown",
 615         "unlink_close",
 616         "multi_mod_rpcs",
 617         "dir_stripe",
 618         "subtree",
 619         "lockahead",
 620         "bulk_mbits",
 621         "compact_obdo",
 622         "second_flags",
 623         /* flags2 names */
 624         "file_secctx",  /* 0x01 */
 625         "lockaheadv2",  /* 0x02 */
 626         "dir_migrate",  /* 0x04 */
 627         "sum_statfs",   /* 0x08 */
 628         "overstriping", /* 0x10 */
 629         "flr",          /* 0x20 */
 630         "wbc",          /* 0x40 */
 631         "lock_convert",  /* 0x80 */
 632         "archive_id_array",     /* 0x100 */
 633         "increasing_xid",       /* 0x200 */
 634         "selinux_policy",       /* 0x400 */
 635         "lsom",                 /* 0x800 */
 636         "pcc",                  /* 0x1000 */
 637         "crush",                /* 0x2000 */
 638         "async_discard",        /* 0x4000 */
 639         "client_encryption",    /* 0x8000 */
 640         "fidmap",               /* 0x10000 */
 641         "getattr_pfid",         /* 0x20000 */
 642         "lseek",                /* 0x40000 */
 643         "dom_lvb",              /* 0x80000 */
 644         "reply_mbits",          /* 0x100000 */
 645         "mode_convert",         /* 0x200000 */
 646         "batch_rpc",            /* 0x400000 */
 647         "pcc_ro",               /* 0x800000 */
 648         "mne_nid_type",         /* 0x1000000 */
 649         "lock_contend",         /* 0x2000000 */
 650         "atomic_open_lock",     /* 0x4000000 */
 651         "name_encryption",      /* 0x8000000 */
 652         NULL
 653 };
 654
 655 void obd_connect_seq_flags2str(struct seq_file *m, __u64 flags, __u64 flags2,
 656                                const char *sep)
 657 {
 658         bool first = true;
 659         __u64 mask;
 660         int i;
 661
 662         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 663                 if (flags & mask) {
 664                         seq_printf(m, "%s%s",
 665                                    first ? "" : sep, obd_connect_names[i]);
 666                         first = false;
 667                 }
 668         }
 669
 670         if (flags & ~(mask - 1)) {
 671                 seq_printf(m, "%sunknown_%#llx",
 672                            first ? "" : sep, flags & ~(mask - 1));
 673                 first = false;
 674         }
 675
 676         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 677                 return;
 678
 679         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 680                 if (flags2 & mask) {
 681                         seq_printf(m, "%s%s",
 682                                    first ? "" : sep, obd_connect_names[i]);
 683                         first = false;
 684                 }
 685         }
 686
 687         if (flags2 & ~(mask - 1)) {
 688                 seq_printf(m, "%sunknown2_%#llx",
 689                            first ? "" : sep, flags2 & ~(mask - 1));
 690                 first = false;
 691         }
 692 }
 693 EXPORT_SYMBOL(obd_connect_seq_flags2str);
 694
 695 int obd_connect_flags2str(char *page, int count, __u64 flags, __u64 flags2,
 696                           const char *sep)
 697 {
 698         __u64 mask;
 699         int i, ret = 0;
 700
 701         for (i = 0, mask = 1; i < 64; i++, mask <<= 1) {
 702                 if (flags & mask)
 703                         ret += snprintf(page + ret, count - ret, "%s%s",
 704                                         ret ? sep : "", obd_connect_names[i]);
 705         }
 706
 707         if (flags & ~(mask - 1))
 708                 ret += snprintf(page + ret, count - ret,
 709                                 "%sunknown_%#llx",
 710                                 ret ? sep : "", flags & ~(mask - 1));
 711
 712         if (!(flags & OBD_CONNECT_FLAGS2) || flags2 == 0)
 713                 return ret;
 714
 715         for (i = 64, mask = 1; obd_connect_names[i] != NULL; i++, mask <<= 1) {
 716                 if (flags2 & mask)
 717                         ret += snprintf(page + ret, count - ret, "%s%s",
 718                                         ret ? sep : "", obd_connect_names[i]);
 719         }
 720
 721         if (flags2 & ~(mask - 1))
 722                 ret += snprintf(page + ret, count - ret,
 723                                 "%sunknown2_%#llx",
 724                                 ret ? sep : "", flags2 & ~(mask - 1));
 725
 726         return ret;
 727 }
 728 EXPORT_SYMBOL(obd_connect_flags2str);
 729
 730 void
 731 obd_connect_data_seqprint(struct seq_file *m, struct obd_connect_data *ocd)
 732 {
 733         __u64 flags;
 734
 735         LASSERT(ocd != NULL);
 736         flags = ocd->ocd_connect_flags;
 737
 738         seq_printf(m, "    connect_data:\n"
 739                    "       flags: %#llx\n"
 740                    "       instance: %u\n",
 741                    ocd->ocd_connect_flags,
 742                    ocd->ocd_instance);
 743         if (flags & OBD_CONNECT_VERSION)
 744                 seq_printf(m, "       target_version: %u.%u.%u.%u\n",
 745                            OBD_OCD_VERSION_MAJOR(ocd->ocd_version),
 746                            OBD_OCD_VERSION_MINOR(ocd->ocd_version),
 747                            OBD_OCD_VERSION_PATCH(ocd->ocd_version),
 748                            OBD_OCD_VERSION_FIX(ocd->ocd_version));
 749         if (flags & OBD_CONNECT_MDS)
 750                 seq_printf(m, "       mdt_index: %d\n", ocd->ocd_group);
 751         if (flags & OBD_CONNECT_GRANT)
 752                 seq_printf(m, "       initial_grant: %d\n", ocd->ocd_grant);
 753         if (flags & OBD_CONNECT_INDEX)
 754                 seq_printf(m, "       target_index: %u\n", ocd->ocd_index);
 755         if (flags & OBD_CONNECT_BRW_SIZE)
 756                 seq_printf(m, "       max_brw_size: %d\n", ocd->ocd_brw_size);
 757         if (flags & OBD_CONNECT_IBITS)
 758                 seq_printf(m, "       ibits_known: %#llx\n",
 759                            ocd->ocd_ibits_known);
 760         if (flags & OBD_CONNECT_GRANT_PARAM)
 761                 seq_printf(m, "       grant_block_size: %d\n"
 762                            "       grant_inode_size: %d\n"
 763                            "       grant_max_extent_size: %d\n"
 764                            "       grant_extent_tax: %d\n",
 765                            1 << ocd->ocd_grant_blkbits,
 766                            1 << ocd->ocd_grant_inobits,
 767                            ocd->ocd_grant_max_blks << ocd->ocd_grant_blkbits,
 768                            ocd->ocd_grant_tax_kb << 10);
 769         if (flags & OBD_CONNECT_TRANSNO)
 770                 seq_printf(m, "       first_transno: %#llx\n",
 771                            ocd->ocd_transno);
 772         if (flags & OBD_CONNECT_CKSUM)
 773                 seq_printf(m, "       cksum_types: %#x\n",
 774                            ocd->ocd_cksum_types);
 775         if (flags & OBD_CONNECT_MAX_EASIZE)
 776                 seq_printf(m, "       max_easize: %d\n", ocd->ocd_max_easize);
 777         if (flags & OBD_CONNECT_MAXBYTES)
 778                 seq_printf(m, "       max_object_bytes: %llu\n",
 779                            ocd->ocd_maxbytes);
 780         if (flags & OBD_CONNECT_MULTIMODRPCS)
 781                 seq_printf(m, "       max_mod_rpcs: %hu\n",
 782                            ocd->ocd_maxmodrpcs);
 783 }
 784
 785 static void lprocfs_import_seq_show_locked(struct seq_file *m,
 786                                            struct obd_device *obd,
 787                                            struct obd_import *imp)
 788 {
 789         char nidstr[LNET_NIDSTR_SIZE];
 790         struct lprocfs_counter ret;
 791         struct lprocfs_counter_header *header;
 792         struct obd_import_conn *conn;
 793         struct obd_connect_data *ocd;
 794         int j;
 795         int k;
 796         int rw = 0;
 797
 798         ocd = &imp->imp_connect_data;
 799
 800         seq_printf(m, "import:\n"
 801                    "    name: %s\n"
 802                    "    target: %s\n"
 803                    "    state: %s\n"
 804                    "    connect_flags: [ ",
 805                    obd->obd_name,
 806                    obd2cli_tgt(obd),
 807                    ptlrpc_import_state_name(imp->imp_state));
 808         obd_connect_seq_flags2str(m, imp->imp_connect_data.ocd_connect_flags,
 809                                   imp->imp_connect_data.ocd_connect_flags2,
 810                                   ", ");
 811         seq_printf(m, " ]\n");
 812         obd_connect_data_seqprint(m, ocd);
 813         seq_printf(m, "    import_flags: [ ");
 814         obd_import_flags2str(imp, m);
 815
 816         seq_printf(m, " ]\n"
 817                    "    connection:\n"
 818                    "       failover_nids: [ ");
 819         spin_lock(&imp->imp_lock);
 820         j = 0;
 821         list_for_each_entry(conn, &imp->imp_conn_list, oic_item) {
 822                 libcfs_nidstr_r(&conn->oic_conn->c_peer.nid,
 823                                   nidstr, sizeof(nidstr));
 824                 seq_printf(m, "%s%s", j ? ", " : "", nidstr);
 825                 j++;
 826         }
 827         if (imp->imp_connection)
 828                 libcfs_nidstr_r(&imp->imp_connection->c_peer.nid,
 829                                   nidstr, sizeof(nidstr));
 830         else
 831                 strncpy(nidstr, "<none>", sizeof(nidstr));
 832         seq_printf(m, " ]\n"
 833                    "       current_connection: %s\n"
 834                    "       connection_attempts: %u\n"
 835                    "       generation: %u\n"
 836                    "       in-progress_invalidations: %u\n"
 837                    "       idle: %lld sec\n",
 838                    nidstr,
 839                    imp->imp_conn_cnt,
 840                    imp->imp_generation,
 841                    atomic_read(&imp->imp_inval_count),
 842                    ktime_get_real_seconds() - imp->imp_last_reply_time);
 843         spin_unlock(&imp->imp_lock);
 844
 845         if (!obd->obd_svc_stats)
 846                 return;
 847
 848         header = &obd->obd_svc_stats->ls_cnt_header[PTLRPC_REQWAIT_CNTR];
 849         lprocfs_stats_collect(obd->obd_svc_stats, PTLRPC_REQWAIT_CNTR, &ret);
 850         if (ret.lc_count != 0)
 851                 ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 852         else
 853                 ret.lc_sum = 0;
 854         seq_printf(m, "    rpcs:\n"
 855                    "       inflight: %u\n"
 856                    "       unregistering: %u\n"
 857                    "       timeouts: %u\n"
 858                    "       avg_waittime: %llu %s\n",
 859                    atomic_read(&imp->imp_inflight),
 860                    atomic_read(&imp->imp_unregistering),
 861                    atomic_read(&imp->imp_timeouts),
 862                    ret.lc_sum, header->lc_units);
 863
 864         k = 0;
 865         for(j = 0; j < IMP_AT_MAX_PORTALS; j++) {
 866                 if (imp->imp_at.iat_portal[j] == 0)
 867                         break;
 868                 k = max_t(unsigned int, k,
 869                           at_get(&imp->imp_at.iat_service_estimate[j]));
 870         }
 871         seq_printf(m, "    service_estimates:\n"
 872                    "       services: %u sec\n"
 873                    "       network: %d sec\n",
 874                    k,
 875                    at_get(&imp->imp_at.iat_net_latency));
 876
 877         seq_printf(m, "    transactions:\n"
 878                    "       last_replay: %llu\n"
 879                    "       peer_committed: %llu\n"
 880                    "       last_checked: %llu\n",
 881                    imp->imp_last_replay_transno,
 882                    imp->imp_peer_committed_transno,
 883                    imp->imp_last_transno_checked);
 884
 885         /* avg data rates */
 886         for (rw = 0; rw <= 1; rw++) {
 887                 lprocfs_stats_collect(obd->obd_svc_stats,
 888                                       PTLRPC_LAST_CNTR + BRW_READ_BYTES + rw,
 889                                       &ret);
 890                 if (ret.lc_sum > 0 && ret.lc_count > 0) {
 891                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 892                         seq_printf(m, "    %s_data_averages:\n"
 893                                    "       bytes_per_rpc: %llu\n",
 894                                    rw ? "write" : "read",
 895                                    ret.lc_sum);
 896                 }
 897                 k = (int)ret.lc_sum;
 898                 j = opcode_offset(OST_READ + rw) + EXTRA_MAX_OPCODES;
 899                 header = &obd->obd_svc_stats->ls_cnt_header[j];
 900                 lprocfs_stats_collect(obd->obd_svc_stats, j, &ret);
 901                 if (ret.lc_sum > 0 && ret.lc_count != 0) {
 902                         ret.lc_sum = div64_s64(ret.lc_sum, ret.lc_count);
 903                         seq_printf(m, "       %s_per_rpc: %llu\n",
 904                                    header->lc_units, ret.lc_sum);
 905                         j = (int)ret.lc_sum;
 906                         if (j > 0)
 907                                 seq_printf(m, "       MB_per_sec: %u.%.02u\n",
 908                                            k / j, (100 * k / j) % 100);
 909                 }
 910         }
 911 }
 912
 913 int lprocfs_import_seq_show(struct seq_file *m, void *data)
 914 {
 915         struct obd_device *obd = (struct obd_device *)data;
 916         struct obd_import *imp;
 917         int rv;
 918
 919         LASSERT(obd != NULL);
 920         with_imp_locked(obd, imp, rv)
 921                 lprocfs_import_seq_show_locked(m, obd, imp);
 922         return rv;
 923 }
 924 EXPORT_SYMBOL(lprocfs_import_seq_show);
 925
 926 int lprocfs_state_seq_show(struct seq_file *m, void *data)
 927 {
 928         struct obd_device *obd = (struct obd_device *)data;
 929         struct obd_import *imp;
 930         int j, k;
 931         int rc;
 932
 933         LASSERT(obd != NULL);
 934         with_imp_locked(obd, imp, rc) {
 935                 seq_printf(m, "current_state: %s\n",
 936                            ptlrpc_import_state_name(imp->imp_state));
 937                 seq_printf(m, "state_history:\n");
 938                 k = imp->imp_state_hist_idx;
 939                 for (j = 0; j < IMP_STATE_HIST_LEN; j++) {
 940                         struct import_state_hist *ish =
 941                                 &imp->imp_state_hist[(k + j) % IMP_STATE_HIST_LEN];
 942                         if (ish->ish_state == 0)
 943                                 continue;
 944                         seq_printf(m, " - [ %lld, %s ]\n", (s64)ish->ish_time,
 945                                    ptlrpc_import_state_name(ish->ish_state));
 946                 }
 947         }
 948
 949         return rc;
 950 }
 951 EXPORT_SYMBOL(lprocfs_state_seq_show);
 952
 953 int lprocfs_at_hist_helper(struct seq_file *m, struct adaptive_timeout *at)
 954 {
 955         int i;
 956         for (i = 0; i < AT_BINS; i++)
 957                 seq_printf(m, "%3u ", at->at_hist[i]);
 958         seq_printf(m, "\n");
 959         return 0;
 960 }
 961 EXPORT_SYMBOL(lprocfs_at_hist_helper);
 962
 963 /* See also ptlrpc_lprocfs_timeouts_show_seq */
 964 static void lprocfs_timeouts_seq_show_locked(struct seq_file *m,
 965                                              struct obd_device *obd,
 966                                              struct obd_import *imp)
 967 {
 968         timeout_t cur_timeout, worst_timeout;
 969         time64_t now, worst_timestamp;
 970         int i;
 971
 972         LASSERT(obd != NULL);
 973
 974         now = ktime_get_real_seconds();
 975
 976         /* Some network health info for kicks */
 977         seq_printf(m, "%-10s : %lld, %llds ago\n",
 978                    "last reply", (s64)imp->imp_last_reply_time,
 979                    (s64)(now - imp->imp_last_reply_time));
 980
 981         cur_timeout = at_get(&imp->imp_at.iat_net_latency);
 982         worst_timeout = imp->imp_at.iat_net_latency.at_worst_timeout_ever;
 983         worst_timestamp = imp->imp_at.iat_net_latency.at_worst_timestamp;
 984         seq_printf(m, "%-10s : cur %3u  worst %3u (at %lld, %llds ago) ",
 985                    "network", cur_timeout, worst_timeout, worst_timestamp,
 986                    now - worst_timestamp);
 987         lprocfs_at_hist_helper(m, &imp->imp_at.iat_net_latency);
 988
 989         for(i = 0; i < IMP_AT_MAX_PORTALS; i++) {
 990                 struct adaptive_timeout *service_est;
 991
 992                 if (imp->imp_at.iat_portal[i] == 0)
 993                         break;
 994
 995                 service_est = &imp->imp_at.iat_service_estimate[i];
 996                 cur_timeout = at_get(service_est);
 997                 worst_timeout = service_est->at_worst_timeout_ever;
 998                 worst_timestamp = service_est->at_worst_timestamp;
 999                 seq_printf(m, "portal %-2d  : cur %3u  worst %3u (at %lld, %llds ago) ",
1000                            imp->imp_at.iat_portal[i], cur_timeout,
1001                            worst_timeout, worst_timestamp,
1002                            now - worst_timestamp);
1003                 lprocfs_at_hist_helper(m, service_est);
1004         }
1005 }
1006
1007 int lprocfs_timeouts_seq_show(struct seq_file *m, void *data)
1008 {
1009         struct obd_device *obd = (struct obd_device *)data;
1010         struct obd_import *imp;
1011         int rc;
1012
1013         with_imp_locked(obd, imp, rc)
1014                 lprocfs_timeouts_seq_show_locked(m, obd, imp);
1015         return rc;
1016 }
1017 EXPORT_SYMBOL(lprocfs_timeouts_seq_show);
1018
1019 int lprocfs_connect_flags_seq_show(struct seq_file *m, void *data)
1020 {
1021         struct obd_device *obd = data;
1022         __u64 flags;
1023         __u64 flags2;
1024         struct obd_import *imp;
1025         int rc;
1026
1027         with_imp_locked(obd, imp, rc) {
1028                 flags = imp->imp_connect_data.ocd_connect_flags;
1029                 flags2 = imp->imp_connect_data.ocd_connect_flags2;
1030                 seq_printf(m, "flags=%#llx\n", flags);
1031                 seq_printf(m, "flags2=%#llx\n", flags2);
1032                 obd_connect_seq_flags2str(m, flags, flags2, "\n");
1033                 seq_printf(m, "\n");
1034         }
1035
1036         return rc;
1037 }
1038 EXPORT_SYMBOL(lprocfs_connect_flags_seq_show);
1039
1040 static const struct attribute *obd_def_uuid_attrs[] = {
1041         &lustre_attr_uuid.attr,
1042         NULL,
1043 };
1044
1045 static const struct attribute *obd_def_attrs[] = {
1046         &lustre_attr_blocksize.attr,
1047         &lustre_attr_kbytestotal.attr,
1048         &lustre_attr_kbytesfree.attr,
1049         &lustre_attr_kbytesavail.attr,
1050         &lustre_attr_filestotal.attr,
1051         &lustre_attr_filesfree.attr,
1052         &lustre_attr_uuid.attr,
1053         NULL,
1054 };
1055
1056 static void obd_sysfs_release(struct kobject *kobj)
1057 {
1058         struct obd_device *obd = container_of(kobj, struct obd_device,
1059                                               obd_kset.kobj);
1060
1061         complete(&obd->obd_kobj_unregister);
1062 }
1063
1064 int lprocfs_obd_setup(struct obd_device *obd, bool uuid_only)
1065 {
1066         struct ldebugfs_vars *debugfs_vars = NULL;
1067         int rc;
1068
1069         if (!obd || obd->obd_magic != OBD_DEVICE_MAGIC)
1070                 return -ENODEV;
1071
1072         rc = kobject_set_name(&obd->obd_kset.kobj, "%s", obd->obd_name);
1073         if (rc)
1074                 return rc;
1075
1076         obd->obd_ktype.sysfs_ops = &lustre_sysfs_ops;
1077         obd->obd_ktype.release = obd_sysfs_release;
1078
1079         obd->obd_kset.kobj.parent = &obd->obd_type->typ_kobj;
1080         obd->obd_kset.kobj.ktype = &obd->obd_ktype;
1081         init_completion(&obd->obd_kobj_unregister);
1082         rc = kset_register(&obd->obd_kset);
1083         if (rc)
1084                 return rc;
1085
1086         if (uuid_only)
1087                 obd->obd_attrs = obd_def_uuid_attrs;
1088         else
1089                 obd->obd_attrs = obd_def_attrs;
1090
1091         rc = sysfs_create_files(&obd->obd_kset.kobj, obd->obd_attrs);
1092         if (rc) {
1093                 kset_unregister(&obd->obd_kset);
1094                 return rc;
1095         }
1096
1097         if (!obd->obd_type->typ_procroot)
1098                 debugfs_vars = obd->obd_debugfs_vars;
1099         obd->obd_debugfs_entry = debugfs_create_dir(
1100                 obd->obd_name, obd->obd_type->typ_debugfs_entry);
1101         ldebugfs_add_vars(obd->obd_debugfs_entry, debugfs_vars, obd);
1102
1103         if (obd->obd_proc_entry || !obd->obd_type->typ_procroot)
1104                 GOTO(already_registered, rc);
1105
1106         obd->obd_proc_entry = lprocfs_register(obd->obd_name,
1107                                                obd->obd_type->typ_procroot,
1108                                                obd->obd_vars, obd);
1109         if (IS_ERR(obd->obd_proc_entry)) {
1110                 rc = PTR_ERR(obd->obd_proc_entry);
1111                 CERROR("error %d setting up lprocfs for %s\n",rc,obd->obd_name);
1112                 obd->obd_proc_entry = NULL;
1113
1114                 debugfs_remove_recursive(obd->obd_debugfs_entry);
1115                 obd->obd_debugfs_entry = NULL;
1116
1117                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1118                 obd->obd_attrs = NULL;
1119                 kset_unregister(&obd->obd_kset);
1120                 return rc;
1121         }
1122 already_registered:
1123         return rc;
1124 }
1125 EXPORT_SYMBOL(lprocfs_obd_setup);
1126
1127 int lprocfs_obd_cleanup(struct obd_device *obd)
1128 {
1129         if (!obd)
1130                 return -EINVAL;
1131
1132         if (obd->obd_proc_exports_entry) {
1133                 /* Should be no exports left */
1134                 lprocfs_remove(&obd->obd_proc_exports_entry);
1135                 obd->obd_proc_exports_entry = NULL;
1136         }
1137
1138         if (obd->obd_proc_entry) {
1139                 lprocfs_remove(&obd->obd_proc_entry);
1140                 obd->obd_proc_entry = NULL;
1141         }
1142
1143         debugfs_remove_recursive(obd->obd_debugfs_entry);
1144         obd->obd_debugfs_entry = NULL;
1145
1146         /* obd device never allocated a kset */
1147         if (!obd->obd_kset.kobj.state_initialized)
1148                 return 0;
1149
1150         if (obd->obd_attrs) {
1151                 sysfs_remove_files(&obd->obd_kset.kobj, obd->obd_attrs);
1152                 obd->obd_attrs = NULL;
1153         }
1154
1155         kset_unregister(&obd->obd_kset);
1156         wait_for_completion(&obd->obd_kobj_unregister);
1157         return 0;
1158 }
1159 EXPORT_SYMBOL(lprocfs_obd_cleanup);
1160
1161 int lprocfs_stats_alloc_one(struct lprocfs_stats *stats, unsigned int cpuid)
1162 {
1163         struct lprocfs_counter *cntr;
1164         unsigned int percpusize;
1165         int rc = -ENOMEM;
1166         unsigned long flags = 0;
1167         int i;
1168
1169         LASSERT(stats->ls_percpu[cpuid] == NULL);
1170         LASSERT((stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU) == 0);
1171
1172         percpusize = lprocfs_stats_counter_size(stats);
1173         LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[cpuid], percpusize);
1174         if (stats->ls_percpu[cpuid]) {
1175                 rc = 0;
1176                 if (unlikely(stats->ls_biggest_alloc_num <= cpuid)) {
1177                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1178                                 spin_lock_irqsave(&stats->ls_lock, flags);
1179                         else
1180                                 spin_lock(&stats->ls_lock);
1181                         if (stats->ls_biggest_alloc_num <= cpuid)
1182                                 stats->ls_biggest_alloc_num = cpuid + 1;
1183                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) {
1184                                 spin_unlock_irqrestore(&stats->ls_lock, flags);
1185                         } else {
1186                                 spin_unlock(&stats->ls_lock);
1187                         }
1188                 }
1189                 /* initialize the ls_percpu[cpuid] non-zero counter */
1190                 for (i = 0; i < stats->ls_num; ++i) {
1191                         cntr = lprocfs_stats_counter_get(stats, cpuid, i);
1192                         cntr->lc_min = LC_MIN_INIT;
1193                 }
1194         }
1195         return rc;
1196 }
1197
1198 struct lprocfs_stats *lprocfs_alloc_stats(unsigned int num,
1199                                           enum lprocfs_stats_flags flags)
1200 {
1201         struct lprocfs_stats *stats;
1202         unsigned int num_entry;
1203         unsigned int percpusize = 0;
1204         int i;
1205
1206         if (num == 0)
1207                 return NULL;
1208
1209         if (lprocfs_no_percpu_stats != 0)
1210                 flags |= LPROCFS_STATS_FLAG_NOPERCPU;
1211
1212         if (flags & LPROCFS_STATS_FLAG_NOPERCPU)
1213                 num_entry = 1;
1214         else
1215                 num_entry = num_possible_cpus();
1216
1217         /* alloc percpu pointers for all possible cpu slots */
1218         LIBCFS_ALLOC(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1219         if (!stats)
1220                 return NULL;
1221
1222         stats->ls_num = num;
1223         stats->ls_flags = flags;
1224         stats->ls_init = ktime_get_real();
1225         spin_lock_init(&stats->ls_lock);
1226
1227         /* alloc num of counter headers */
1228         CFS_ALLOC_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1229         if (!stats->ls_cnt_header)
1230                 goto fail;
1231
1232         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1233                 /* contains only one set counters */
1234                 percpusize = lprocfs_stats_counter_size(stats);
1235                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1236                 if (!stats->ls_percpu[0])
1237                         goto fail;
1238                 stats->ls_biggest_alloc_num = 1;
1239         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1240                 /* alloc all percpu data, currently only obd_memory use this */
1241                 for (i = 0; i < num_entry; ++i)
1242                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1243                                 goto fail;
1244         }
1245
1246         return stats;
1247
1248 fail:
1249         lprocfs_free_stats(&stats);
1250         return NULL;
1251 }
1252 EXPORT_SYMBOL(lprocfs_alloc_stats);
1253
1254 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1255 {
1256         struct lprocfs_stats *stats = *statsh;
1257         unsigned int num_entry;
1258         unsigned int percpusize;
1259         unsigned int i;
1260
1261         if (!stats || stats->ls_num == 0)
1262                 return;
1263         *statsh = NULL;
1264
1265         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1266                 num_entry = 1;
1267         else
1268                 num_entry = num_possible_cpus();
1269
1270         percpusize = lprocfs_stats_counter_size(stats);
1271         for (i = 0; i < num_entry; i++)
1272                 if (stats->ls_percpu[i])
1273                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1274         if (stats->ls_cnt_header)
1275                 CFS_FREE_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1276         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1277 }
1278 EXPORT_SYMBOL(lprocfs_free_stats);
1279
1280 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1281                             enum lprocfs_fields_flags field)
1282 {
1283         unsigned long flags = 0;
1284         unsigned int num_cpu;
1285         unsigned int i;
1286         u64 ret = 0;
1287
1288         LASSERT(stats);
1289
1290         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1291         for (i = 0; i < num_cpu; i++) {
1292                 struct lprocfs_counter *cntr;
1293
1294                 if (!stats->ls_percpu[i])
1295                         continue;
1296
1297                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1298                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1299                                            stats->ls_flags, field);
1300         }
1301         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1302         return ret;
1303 }
1304 EXPORT_SYMBOL(lprocfs_stats_collector);
1305
1306 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1307 {
1308         struct lprocfs_counter *percpu_cntr;
1309         int i;
1310         int j;
1311         unsigned int num_entry;
1312         unsigned long flags = 0;
1313
1314         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1315
1316         for (i = 0; i < num_entry; i++) {
1317                 if (!stats->ls_percpu[i])
1318                         continue;
1319                 for (j = 0; j < stats->ls_num; j++) {
1320                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1321                         percpu_cntr->lc_count           = 0;
1322                         percpu_cntr->lc_min             = LC_MIN_INIT;
1323                         percpu_cntr->lc_max             = 0;
1324                         percpu_cntr->lc_sumsquare       = 0;
1325                         percpu_cntr->lc_sum             = 0;
1326                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1327                                 percpu_cntr->lc_sum_irq = 0;
1328                 }
1329         }
1330         stats->ls_init = ktime_get_real();
1331
1332         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1333 }
1334 EXPORT_SYMBOL(lprocfs_clear_stats);
1335
1336 static ssize_t lprocfs_stats_seq_write(struct file *file,
1337                                        const char __user *buf,
1338                                        size_t len, loff_t *off)
1339 {
1340         struct seq_file *seq = file->private_data;
1341         struct lprocfs_stats *stats = seq->private;
1342
1343         lprocfs_clear_stats(stats);
1344
1345         return len;
1346 }
1347
1348 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1349 {
1350         struct lprocfs_stats *stats = p->private;
1351
1352         return (*pos < stats->ls_num) ? pos : NULL;
1353 }
1354
1355 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1356 {
1357 }
1358
1359 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1360 {
1361         (*pos)++;
1362
1363         return lprocfs_stats_seq_start(p, pos);
1364 }
1365
1366 /**
1367  * print header of stats including snapshot_time, start_time and elapsed_time.
1368  *
1369  * \param seq           the file to print content to
1370  * \param now           end time to calculate elapsed_time
1371  * \param ts_init       start time to calculate elapsed_time
1372  * \param width         the width of key to align them well
1373  * \param colon         "" or ":"
1374  * \param show_units    show units or not
1375  * \param prefix        prefix (indent) before printing each line of header
1376  *                      to align them with other content
1377  */
1378 void lprocfs_stats_header(struct seq_file *seq, ktime_t now, ktime_t ts_init,
1379                           int width, const char *colon, bool show_units,
1380                           const char *prefix)
1381 {
1382         const char *units = show_units ? " secs.nsecs" : "";
1383         struct timespec64 ts;
1384         const char *field;
1385
1386         field = (colon && colon[0]) ? "snapshot_time:" : "snapshot_time";
1387         ts = ktime_to_timespec64(now);
1388         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1389                    (s64)ts.tv_sec, ts.tv_nsec, units);
1390
1391         field = (colon && colon[0]) ? "start_time:" : "start_time";
1392         ts = ktime_to_timespec64(ts_init);
1393         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1394                    (s64)ts.tv_sec, ts.tv_nsec, units);
1395
1396         field = (colon && colon[0]) ? "elapsed_time:" : "elapsed_time";
1397         ts = ktime_to_timespec64(ktime_sub(now, ts_init));
1398         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1399                    (s64)ts.tv_sec, ts.tv_nsec, units);
1400 }
1401 EXPORT_SYMBOL(lprocfs_stats_header);
1402
1403 /* seq file export of one lprocfs counter */
1404 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1405 {
1406         struct lprocfs_stats *stats = p->private;
1407         struct lprocfs_counter_header *hdr;
1408         struct lprocfs_counter ctr;
1409         int idx = *(loff_t *)v;
1410
1411         if (idx == 0)
1412                 lprocfs_stats_header(p, ktime_get_real(), stats->ls_init, 25,
1413                                      "", true, "");
1414
1415         hdr = &stats->ls_cnt_header[idx];
1416         lprocfs_stats_collect(stats, idx, &ctr);
1417
1418         if (ctr.lc_count == 0)
1419                 return 0;
1420
1421         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1422                    ctr.lc_count, hdr->lc_units);
1423
1424         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1425                 seq_printf(p, " %lld %lld %lld",
1426                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1427                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1428                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1429         }
1430         seq_putc(p, '\n');
1431         return 0;
1432 }
1433
1434 static const struct seq_operations lprocfs_stats_seq_sops = {
1435         .start  = lprocfs_stats_seq_start,
1436         .stop   = lprocfs_stats_seq_stop,
1437         .next   = lprocfs_stats_seq_next,
1438         .show   = lprocfs_stats_seq_show,
1439 };
1440
1441 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1442 {
1443         struct seq_file *seq;
1444         int rc;
1445
1446         rc = seq_open(file, &lprocfs_stats_seq_sops);
1447         if (rc)
1448                 return rc;
1449         seq = file->private_data;
1450         seq->private = inode->i_private ? inode->i_private : pde_data(inode);
1451         return 0;
1452 }
1453
1454 const struct file_operations ldebugfs_stats_seq_fops = {
1455         .owner   = THIS_MODULE,
1456         .open    = lprocfs_stats_seq_open,
1457         .read    = seq_read,
1458         .write   = lprocfs_stats_seq_write,
1459         .llseek  = seq_lseek,
1460         .release = lprocfs_seq_release,
1461 };
1462 EXPORT_SYMBOL(ldebugfs_stats_seq_fops);
1463
1464 static const struct proc_ops lprocfs_stats_seq_fops = {
1465         PROC_OWNER(THIS_MODULE)
1466         .proc_open      = lprocfs_stats_seq_open,
1467         .proc_read      = seq_read,
1468         .proc_write     = lprocfs_stats_seq_write,
1469         .proc_lseek     = seq_lseek,
1470         .proc_release   = lprocfs_seq_release,
1471 };
1472
1473 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1474                            struct lprocfs_stats *stats)
1475 {
1476         struct proc_dir_entry *entry;
1477         LASSERT(root != NULL);
1478
1479         entry = proc_create_data(name, 0644, root,
1480                                  &lprocfs_stats_seq_fops, stats);
1481         if (!entry)
1482                 return -ENOMEM;
1483         return 0;
1484 }
1485 EXPORT_SYMBOL(lprocfs_register_stats);
1486
1487 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1488                           unsigned conf, const char *name, const char *units)
1489 {
1490         struct lprocfs_counter_header *header;
1491         struct lprocfs_counter *percpu_cntr;
1492         unsigned long flags = 0;
1493         unsigned int i;
1494         unsigned int num_cpu;
1495
1496         LASSERT(stats != NULL);
1497
1498         header = &stats->ls_cnt_header[index];
1499         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1500                  index, name, units);
1501
1502         header->lc_config = conf;
1503         header->lc_name   = name;
1504         header->lc_units  = units;
1505
1506         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1507         for (i = 0; i < num_cpu; ++i) {
1508                 if (!stats->ls_percpu[i])
1509                         continue;
1510                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1511                 percpu_cntr->lc_count           = 0;
1512                 percpu_cntr->lc_min             = LC_MIN_INIT;
1513                 percpu_cntr->lc_max             = 0;
1514                 percpu_cntr->lc_sumsquare       = 0;
1515                 percpu_cntr->lc_sum             = 0;
1516                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1517                         percpu_cntr->lc_sum_irq = 0;
1518         }
1519         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1520 }
1521 EXPORT_SYMBOL(lprocfs_counter_init);
1522
1523 static const char * const mps_stats[] = {
1524         [LPROC_MD_CLOSE]                = "close",
1525         [LPROC_MD_CREATE]               = "create",
1526         [LPROC_MD_ENQUEUE]              = "enqueue",
1527         [LPROC_MD_GETATTR]              = "getattr",
1528         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1529         [LPROC_MD_LINK]                 = "link",
1530         [LPROC_MD_RENAME]               = "rename",
1531         [LPROC_MD_SETATTR]              = "setattr",
1532         [LPROC_MD_FSYNC]                = "fsync",
1533         [LPROC_MD_READ_PAGE]            = "read_page",
1534         [LPROC_MD_UNLINK]               = "unlink",
1535         [LPROC_MD_SETXATTR]             = "setxattr",
1536         [LPROC_MD_GETXATTR]             = "getxattr",
1537         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1538         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1539 };
1540
1541 int lprocfs_alloc_md_stats(struct obd_device *obd,
1542                            unsigned int num_private_stats)
1543 {
1544         struct lprocfs_stats *stats;
1545         unsigned int num_stats;
1546         int rc, i;
1547
1548         /*
1549          * TODO Ensure that this function is only used where
1550          * appropriate by adding an assertion to the effect that
1551          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1552          * because mdt_procfs_init() uses this function to allocate
1553          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1554          * mdt layer does not use the md_ops interface. This is
1555          * confusing and a waste of memory. See LU-2484.
1556          */
1557         LASSERT(obd->obd_proc_entry != NULL);
1558         LASSERT(obd->obd_md_stats == NULL);
1559
1560         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1561         stats = lprocfs_alloc_stats(num_stats, 0);
1562         if (!stats)
1563                 return -ENOMEM;
1564
1565         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1566                 lprocfs_counter_init(stats, i, 0, mps_stats[i], "reqs");
1567                 if (!stats->ls_cnt_header[i].lc_name) {
1568                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1569                                i);
1570                         LBUG();
1571                 }
1572         }
1573
1574         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1575         if (rc < 0) {
1576                 lprocfs_free_stats(&stats);
1577         } else {
1578                 obd->obd_md_stats = stats;
1579         }
1580
1581         return rc;
1582 }
1583 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1584
1585 void lprocfs_free_md_stats(struct obd_device *obd)
1586 {
1587         struct lprocfs_stats *stats = obd->obd_md_stats;
1588
1589         if (stats) {
1590                 obd->obd_md_stats = NULL;
1591                 lprocfs_free_stats(&stats);
1592         }
1593 }
1594 EXPORT_SYMBOL(lprocfs_free_md_stats);
1595
1596 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1597 {
1598         lprocfs_counter_init(ldlm_stats,
1599                              LDLM_ENQUEUE - LDLM_FIRST_OPC,
1600                              0, "ldlm_enqueue", "reqs");
1601         lprocfs_counter_init(ldlm_stats,
1602                              LDLM_CONVERT - LDLM_FIRST_OPC,
1603                              0, "ldlm_convert", "reqs");
1604         lprocfs_counter_init(ldlm_stats,
1605                              LDLM_CANCEL - LDLM_FIRST_OPC,
1606                              0, "ldlm_cancel", "reqs");
1607         lprocfs_counter_init(ldlm_stats,
1608                              LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1609                              0, "ldlm_bl_callback", "reqs");
1610         lprocfs_counter_init(ldlm_stats,
1611                              LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1612                              0, "ldlm_cp_callback", "reqs");
1613         lprocfs_counter_init(ldlm_stats,
1614                              LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1615                              0, "ldlm_gl_callback", "reqs");
1616 }
1617 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1618
1619 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1620                           struct lprocfs_counter_header *header,
1621                           enum lprocfs_stats_flags flags,
1622                           enum lprocfs_fields_flags field)
1623 {
1624         __s64 ret = 0;
1625
1626         if (!lc || !header)
1627                 RETURN(0);
1628
1629         switch (field) {
1630                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1631                         ret = header->lc_config;
1632                         break;
1633                 case LPROCFS_FIELDS_FLAGS_SUM:
1634                         ret = lc->lc_sum;
1635                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1636                                 ret += lc->lc_sum_irq;
1637                         break;
1638                 case LPROCFS_FIELDS_FLAGS_MIN:
1639                         ret = lc->lc_min;
1640                         break;
1641                 case LPROCFS_FIELDS_FLAGS_MAX:
1642                         ret = lc->lc_max;
1643                         break;
1644                 case LPROCFS_FIELDS_FLAGS_AVG:
1645                         ret = (lc->lc_max - lc->lc_min) / 2;
1646                         break;
1647                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1648                         ret = lc->lc_sumsquare;
1649                         break;
1650                 case LPROCFS_FIELDS_FLAGS_COUNT:
1651                         ret = lc->lc_count;
1652                         break;
1653                 default:
1654                         break;
1655         };
1656         RETURN(ret);
1657 }
1658 EXPORT_SYMBOL(lprocfs_read_helper);
1659
1660 /**
1661  * string_to_size - convert ASCII string representing a numerical
1662  *                  value with optional units to 64-bit binary value
1663  *
1664  * @size:       The numerical value extract out of @buffer
1665  * @buffer:     passed in string to parse
1666  * @count:      length of the @buffer
1667  *
1668  * This function returns a 64-bit binary value if @buffer contains a valid
1669  * numerical string. The string is parsed to 3 significant figures after
1670  * the decimal point. Support the string containing an optional units at
1671  * the end which can be base 2 or base 10 in value. If no units are given
1672  * the string is assumed to just a numerical value.
1673  *
1674  * Returns:     @count if the string is successfully parsed,
1675  *              -errno on invalid input strings. Error values:
1676  *
1677  *  - ``-EINVAL``: @buffer is not a proper numerical string
1678  *  - ``-EOVERFLOW``: results does not fit into 64 bits.
1679  *  - ``-E2BIG ``: @buffer is too large (not a valid number)
1680  */
1681 int string_to_size(u64 *size, const char *buffer, size_t count)
1682 {
1683         /* For string_get_size() it can support values above exabytes,
1684          * (ZiB, YiB) due to breaking the return value into a size and
1685          * bulk size to avoid 64 bit overflow. We don't break the size
1686          * up into block size units so we don't support ZiB or YiB.
1687          */
1688         static const char *const units_10[] = {
1689                 "kB", "MB", "GB", "TB", "PB", "EB",
1690         };
1691         static const char *const units_2[] = {
1692                 "K",  "M",  "G",  "T",  "P",  "E",
1693         };
1694         static const char *const *const units_str[] = {
1695                 [STRING_UNITS_2] = units_2,
1696                 [STRING_UNITS_10] = units_10,
1697         };
1698         static const unsigned int coeff[] = {
1699                 [STRING_UNITS_10] = 1000,
1700                 [STRING_UNITS_2] = 1024,
1701         };
1702         enum string_size_units unit = STRING_UNITS_2;
1703         u64 whole, blk_size = 1;
1704         char kernbuf[22], *end;
1705         size_t len = count;
1706         int rc;
1707         int i;
1708
1709         if (count >= sizeof(kernbuf)) {
1710                 CERROR("count %zd > buffer %zd\n", count, sizeof(kernbuf));
1711                 return -E2BIG;
1712         }
1713
1714         *size = 0;
1715         /* The "iB" suffix is optionally allowed for indicating base-2 numbers.
1716          * If suffix is only "B" and not "iB" then we treat it as base-10.
1717          */
1718         end = strstr(buffer, "B");
1719         if (end && *(end - 1) != 'i')
1720                 unit = STRING_UNITS_10;
1721
1722         i = unit == STRING_UNITS_2 ? ARRAY_SIZE(units_2) - 1 :
1723                                      ARRAY_SIZE(units_10) - 1;
1724         do {
1725                 end = strnstr(buffer, units_str[unit][i], count);
1726                 if (end) {
1727                         for (; i >= 0; i--)
1728                                 blk_size *= coeff[unit];
1729                         len = end - buffer;
1730                         break;
1731                 }
1732         } while (i--);
1733
1734         /* as 'B' is a substring of all units, we need to handle it
1735          * separately.
1736          */
1737         if (!end) {
1738                 /* 'B' is only acceptable letter at this point */
1739                 end = strnchr(buffer, count, 'B');
1740                 if (end) {
1741                         len = end - buffer;
1742
1743                         if (count - len > 2 ||
1744                             (count - len == 2 && strcmp(end, "B\n") != 0)) {
1745                                 CDEBUG(D_INFO, "unknown suffix '%s'\n", buffer);
1746                                 return -EINVAL;
1747                         }
1748                 }
1749                 /* kstrtoull will error out if it has non digits */
1750                 goto numbers_only;
1751         }
1752
1753         end = strnchr(buffer, count, '.');
1754         if (end) {
1755                 /* need to limit 3 decimal places */
1756                 char rem[4] = "000";
1757                 u64 frac = 0;
1758                 size_t off;
1759
1760                 len = end - buffer;
1761                 end++;
1762
1763                 /* limit to 3 decimal points */
1764                 off = min_t(size_t, 3, strspn(end, "0123456789"));
1765                 /* need to limit frac_d to a u32 */
1766                 memcpy(rem, end, off);
1767                 rc = kstrtoull(rem, 10, &frac);
1768                 if (rc)
1769                         return rc;
1770
1771                 if (fls64(frac) + fls64(blk_size) - 1 > 64)
1772                         return -EOVERFLOW;
1773
1774                 frac *= blk_size;
1775                 do_div(frac, 1000);
1776                 *size += frac;
1777         }
1778 numbers_only:
1779         snprintf(kernbuf, sizeof(kernbuf), "%.*s", (int)len, buffer);
1780         rc = kstrtoull(kernbuf, 10, &whole);
1781         if (rc)
1782                 return rc;
1783
1784         if (whole != 0 && fls64(whole) + fls64(blk_size) - 1 > 64)
1785                 return -EOVERFLOW;
1786
1787         *size += whole * blk_size;
1788
1789         return count;
1790 }
1791 EXPORT_SYMBOL(string_to_size);
1792
1793 /**
1794  * sysfs_memparse - parse a ASCII string to 64-bit binary value,
1795  *                  with optional units
1796  *
1797  * @buffer:     kernel pointer to input string
1798  * @count:      number of bytes in the input @buffer
1799  * @val:        (output) binary value returned to caller
1800  * @defunit:    default unit suffix to use if none is provided
1801  *
1802  * Parses a string into a number. The number stored at @buffer is
1803  * potentially suffixed with K, M, G, T, P, E. Besides these other
1804  * valid suffix units are shown in the string_to_size() function.
1805  * If the string lacks a suffix then the defunit is used. The defunit
1806  * should be given as a binary unit (e.g. MiB) as that is the standard
1807  * for tunables in Lustre. If no unit suffix is given (e.g. 'G'), then
1808  * it is assumed to be in binary units.
1809  *
1810  * Returns:     0 on success or -errno on failure.
1811  */
1812 int sysfs_memparse(const char *buffer, size_t count, u64 *val,
1813                    const char *defunit)
1814 {
1815         const char *param = buffer;
1816         char tmp_buf[23];
1817         int rc;
1818
1819         count = strlen(buffer);
1820         while (count > 0 && isspace(buffer[count - 1]))
1821                 count--;
1822
1823         if (!count)
1824                 RETURN(-EINVAL);
1825
1826         /* If there isn't already a unit on this value, append @defunit.
1827          * Units of 'B' don't affect the value, so don't bother adding.
1828          */
1829         if (!isalpha(buffer[count - 1]) && defunit[0] != 'B') {
1830                 if (count + 3 >= sizeof(tmp_buf)) {
1831                         CERROR("count %zd > size %zd\n", count, sizeof(param));
1832                         RETURN(-E2BIG);
1833                 }
1834
1835                 scnprintf(tmp_buf, sizeof(tmp_buf), "%.*s%s", (int)count,
1836                           buffer, defunit);
1837                 param = tmp_buf;
1838                 count = strlen(param);
1839         }
1840
1841         rc = string_to_size(val, param, count);
1842
1843         return rc < 0 ? rc : 0;
1844 }
1845 EXPORT_SYMBOL(sysfs_memparse);
1846
1847 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
1848 {
1849         size_t l2;
1850
1851         l2 = strlen(s2);
1852         if (!l2)
1853                 return (char *)s1;
1854         while (len >= l2) {
1855                 len--;
1856                 if (!memcmp(s1, s2, l2))
1857                         return (char *)s1;
1858                 s1++;
1859         }
1860         return NULL;
1861 }
1862 EXPORT_SYMBOL(lprocfs_strnstr);
1863
1864 /**
1865  * Find the string \a name in the input \a buffer, and return a pointer to the
1866  * value immediately following \a name, reducing \a count appropriately.
1867  * If \a name is not found the original \a buffer is returned.
1868  */
1869 char *lprocfs_find_named_value(const char *buffer, const char *name,
1870                                 size_t *count)
1871 {
1872         char *val;
1873         size_t buflen = *count;
1874
1875         /* there is no strnstr() in rhel5 and ubuntu kernels */
1876         val = lprocfs_strnstr(buffer, name, buflen);
1877         if (!val)
1878                 return (char *)buffer;
1879
1880         val += strlen(name);                             /* skip prefix */
1881         while (val < buffer + buflen && isspace(*val)) /* skip separator */
1882                 val++;
1883
1884         *count = 0;
1885         while (val < buffer + buflen && isalnum(*val)) {
1886                 ++*count;
1887                 ++val;
1888         }
1889
1890         return val - *count;
1891 }
1892 EXPORT_SYMBOL(lprocfs_find_named_value);
1893
1894 int lprocfs_seq_create(struct proc_dir_entry *parent,
1895                        const char *name,
1896                        mode_t mode,
1897                        const struct proc_ops *seq_fops,
1898                        void *data)
1899 {
1900         struct proc_dir_entry *entry;
1901         ENTRY;
1902
1903         /* Disallow secretly (un)writable entries. */
1904         LASSERT(!seq_fops->proc_write == !(mode & 0222));
1905
1906         entry = proc_create_data(name, mode, parent, seq_fops, data);
1907
1908         if (!entry)
1909                 RETURN(-ENOMEM);
1910
1911         RETURN(0);
1912 }
1913 EXPORT_SYMBOL(lprocfs_seq_create);
1914
1915 int lprocfs_obd_seq_create(struct obd_device *obd,
1916                            const char *name,
1917                            mode_t mode,
1918                            const struct proc_ops *seq_fops,
1919                            void *data)
1920 {
1921         return lprocfs_seq_create(obd->obd_proc_entry, name,
1922                                   mode, seq_fops, data);
1923 }
1924 EXPORT_SYMBOL(lprocfs_obd_seq_create);
1925
1926 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
1927 {
1928         if (value >= OBD_HIST_MAX)
1929                 value = OBD_HIST_MAX - 1;
1930
1931         spin_lock(&oh->oh_lock);
1932         oh->oh_buckets[value]++;
1933         spin_unlock(&oh->oh_lock);
1934 }
1935 EXPORT_SYMBOL(lprocfs_oh_tally);
1936
1937 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
1938 {
1939         unsigned int val = 0;
1940
1941         if (likely(value != 0))
1942                 val = min(fls(value - 1), OBD_HIST_MAX);
1943
1944         lprocfs_oh_tally(oh, val);
1945 }
1946 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
1947
1948 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
1949 {
1950         unsigned long ret = 0;
1951         int i;
1952
1953         for (i = 0; i < OBD_HIST_MAX; i++)
1954                 ret +=  oh->oh_buckets[i];
1955         return ret;
1956 }
1957 EXPORT_SYMBOL(lprocfs_oh_sum);
1958
1959 void lprocfs_oh_clear(struct obd_histogram *oh)
1960 {
1961         spin_lock(&oh->oh_lock);
1962         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
1963         spin_unlock(&oh->oh_lock);
1964 }
1965 EXPORT_SYMBOL(lprocfs_oh_clear);
1966
1967 void lprocfs_oh_tally_pcpu(struct obd_hist_pcpu *oh,
1968                            unsigned int value)
1969 {
1970         if (value >= OBD_HIST_MAX)
1971                 value = OBD_HIST_MAX - 1;
1972
1973         percpu_counter_inc(&oh->oh_pc_buckets[value]);
1974 }
1975 EXPORT_SYMBOL(lprocfs_oh_tally_pcpu);
1976
1977 void lprocfs_oh_tally_log2_pcpu(struct obd_hist_pcpu *oh,
1978                                 unsigned int value)
1979 {
1980         unsigned int val = 0;
1981
1982         if (likely(value != 0))
1983                 val = min(fls(value - 1), OBD_HIST_MAX);
1984
1985         lprocfs_oh_tally_pcpu(oh, val);
1986 }
1987 EXPORT_SYMBOL(lprocfs_oh_tally_log2_pcpu);
1988
1989 unsigned long lprocfs_oh_counter_pcpu(struct obd_hist_pcpu *oh,
1990                                       unsigned int value)
1991 {
1992         return percpu_counter_sum(&oh->oh_pc_buckets[value]);
1993 }
1994 EXPORT_SYMBOL(lprocfs_oh_counter_pcpu);
1995
1996 unsigned long lprocfs_oh_sum_pcpu(struct obd_hist_pcpu *oh)
1997 {
1998         unsigned long ret = 0;
1999         int i;
2000
2001         for (i = 0; i < OBD_HIST_MAX; i++)
2002                 ret += percpu_counter_sum(&oh->oh_pc_buckets[i]);
2003
2004         return ret;
2005 }
2006 EXPORT_SYMBOL(lprocfs_oh_sum_pcpu);
2007
2008 int lprocfs_oh_alloc_pcpu(struct obd_hist_pcpu *oh)
2009 {
2010         int i, rc;
2011
2012         if (oh->oh_initialized)
2013                 return 0;
2014
2015         for (i = 0; i < OBD_HIST_MAX; i++) {
2016                 rc = percpu_counter_init(&oh->oh_pc_buckets[i], 0, GFP_KERNEL);
2017                 if (rc)
2018                         goto out;
2019         }
2020
2021         oh->oh_initialized = true;
2022
2023         return 0;
2024
2025 out:
2026         for (i--; i >= 0; i--)
2027                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2028
2029         return rc;
2030 }
2031 EXPORT_SYMBOL(lprocfs_oh_alloc_pcpu);
2032
2033 void lprocfs_oh_clear_pcpu(struct obd_hist_pcpu *oh)
2034 {
2035         int i;
2036
2037         for (i = 0; i < OBD_HIST_MAX; i++)
2038                 percpu_counter_set(&oh->oh_pc_buckets[i], 0);
2039 }
2040 EXPORT_SYMBOL(lprocfs_oh_clear_pcpu);
2041
2042 void lprocfs_oh_release_pcpu(struct obd_hist_pcpu *oh)
2043 {
2044         int i;
2045
2046         if (!oh->oh_initialized)
2047                 return;
2048
2049         for (i = 0; i < OBD_HIST_MAX; i++)
2050                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2051
2052         oh->oh_initialized = false;
2053 }
2054 EXPORT_SYMBOL(lprocfs_oh_release_pcpu);
2055
2056 ssize_t lustre_attr_show(struct kobject *kobj,
2057                          struct attribute *attr, char *buf)
2058 {
2059         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2060
2061         return a->show ? a->show(kobj, attr, buf) : 0;
2062 }
2063 EXPORT_SYMBOL_GPL(lustre_attr_show);
2064
2065 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2066                           const char *buf, size_t len)
2067 {
2068         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2069
2070         return a->store ? a->store(kobj, attr, buf, len) : len;
2071 }
2072 EXPORT_SYMBOL_GPL(lustre_attr_store);
2073
2074 const struct sysfs_ops lustre_sysfs_ops = {
2075         .show  = lustre_attr_show,
2076         .store = lustre_attr_store,
2077 };
2078 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2079
2080 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2081 {
2082         struct obd_device *obd = data;
2083         struct client_obd *cli = &obd->u.cli;
2084
2085         spin_lock(&cli->cl_loi_list_lock);
2086         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2087         spin_unlock(&cli->cl_loi_list_lock);
2088         return 0;
2089 }
2090 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2091
2092 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2093                                                 const char __user *buffer,
2094                                                 size_t count, loff_t *off)
2095 {
2096         struct seq_file *m = file->private_data;
2097         struct obd_device *obd = m->private;
2098         struct client_obd *cli = &obd->u.cli;
2099         struct obd_import *imp;
2100         struct obd_connect_data *ocd;
2101         int chunk_mask, rc;
2102         char kernbuf[22];
2103         u64 val;
2104
2105         if (count > sizeof(kernbuf) - 1)
2106                 return -EINVAL;
2107
2108         if (copy_from_user(kernbuf, buffer, count))
2109                 return -EFAULT;
2110
2111         kernbuf[count] = '\0';
2112
2113         rc = sysfs_memparse(kernbuf, count, &val, "B");
2114         if (rc)
2115                 return rc;
2116
2117         /* if the max_pages is specified in bytes, convert to pages */
2118         if (val >= ONE_MB_BRW_SIZE)
2119                 val >>= PAGE_SHIFT;
2120
2121         with_imp_locked(obd, imp, rc) {
2122                 ocd = &imp->imp_connect_data;
2123                 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2124                 /* max_pages_per_rpc must be chunk aligned */
2125                 val = (val + ~chunk_mask) & chunk_mask;
2126                 if (val == 0 || (ocd->ocd_brw_size != 0 &&
2127                                  val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2128                         rc = -ERANGE;
2129                 } else {
2130                         spin_lock(&cli->cl_loi_list_lock);
2131                         cli->cl_max_pages_per_rpc = val;
2132                         client_adjust_max_dirty(cli);
2133                         spin_unlock(&cli->cl_loi_list_lock);
2134                 }
2135         }
2136
2137         return rc ?: count;
2138 }
2139 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2140
2141 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2142                             char *buf)
2143 {
2144         struct obd_device *obd = container_of(kobj, struct obd_device,
2145                                               obd_kset.kobj);
2146         struct client_obd *cli = &obd->u.cli;
2147         int rc;
2148
2149         spin_lock(&cli->cl_loi_list_lock);
2150         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2151         spin_unlock(&cli->cl_loi_list_lock);
2152         return rc;
2153 }
2154 EXPORT_SYMBOL(short_io_bytes_show);
2155
2156 /* Used to catch people who think they're specifying pages. */
2157 #define MIN_SHORT_IO_BYTES 64U
2158
2159 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2160                              const char *buffer, size_t count)
2161 {
2162         struct obd_device *obd = container_of(kobj, struct obd_device,
2163                                               obd_kset.kobj);
2164         struct client_obd *cli = &obd->u.cli;
2165         u64 val;
2166         int rc;
2167
2168         if (strcmp(buffer, "-1") == 0) {
2169                 val = OBD_DEF_SHORT_IO_BYTES;
2170         } else {
2171                 rc = sysfs_memparse(buffer, count, &val, "B");
2172                 if (rc)
2173                         GOTO(out, rc);
2174         }
2175
2176         if (val && (val < MIN_SHORT_IO_BYTES || val > LNET_MTU))
2177                 GOTO(out, rc = -ERANGE);
2178
2179         rc = count;
2180
2181         spin_lock(&cli->cl_loi_list_lock);
2182         cli->cl_max_short_io_bytes = min_t(u64, val, OST_MAX_SHORT_IO_BYTES);
2183         spin_unlock(&cli->cl_loi_list_lock);
2184
2185 out:
2186         return rc;
2187 }
2188 EXPORT_SYMBOL(short_io_bytes_store);
2189
2190 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2191                            struct root_squash_info *squash, char *name)
2192 {
2193         int rc;
2194         char kernbuf[64], *tmp, *errmsg;
2195         unsigned long uid, gid;
2196         ENTRY;
2197
2198         if (count >= sizeof(kernbuf)) {
2199                 errmsg = "string too long";
2200                 GOTO(failed_noprint, rc = -EINVAL);
2201         }
2202         if (copy_from_user(kernbuf, buffer, count)) {
2203                 errmsg = "bad address";
2204                 GOTO(failed_noprint, rc = -EFAULT);
2205         }
2206         kernbuf[count] = '\0';
2207
2208         /* look for uid gid separator */
2209         tmp = strchr(kernbuf, ':');
2210         if (!tmp) {
2211                 errmsg = "needs uid:gid format";
2212                 GOTO(failed, rc = -EINVAL);
2213         }
2214         *tmp = '\0';
2215         tmp++;
2216
2217         /* parse uid */
2218         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2219                 errmsg = "bad uid";
2220                 GOTO(failed, rc = -EINVAL);
2221         }
2222
2223         /* parse gid */
2224         if (kstrtoul(tmp, 0, &gid) != 0) {
2225                 errmsg = "bad gid";
2226                 GOTO(failed, rc = -EINVAL);
2227         }
2228
2229         squash->rsi_uid = uid;
2230         squash->rsi_gid = gid;
2231
2232         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2233                       name, squash->rsi_uid, squash->rsi_gid);
2234         RETURN(count);
2235
2236 failed:
2237         if (tmp) {
2238                 tmp--;
2239                 *tmp = ':';
2240         }
2241         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2242               name, kernbuf, errmsg, rc);
2243         RETURN(rc);
2244 failed_noprint:
2245         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2246               name, errmsg, rc);
2247         RETURN(rc);
2248 }
2249 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2250
2251
2252 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2253                              struct root_squash_info *squash, char *name)
2254 {
2255         int rc;
2256         char *kernbuf = NULL;
2257         char *errmsg;
2258         LIST_HEAD(tmp);
2259         int len = count;
2260         ENTRY;
2261
2262         if (count > 4096) {
2263                 errmsg = "string too long";
2264                 GOTO(failed, rc = -EINVAL);
2265         }
2266
2267         OBD_ALLOC(kernbuf, count + 1);
2268         if (!kernbuf) {
2269                 errmsg = "no memory";
2270                 GOTO(failed, rc = -ENOMEM);
2271         }
2272         if (copy_from_user(kernbuf, buffer, count)) {
2273                 errmsg = "bad address";
2274                 GOTO(failed, rc = -EFAULT);
2275         }
2276         kernbuf[count] = '\0';
2277
2278         if (count > 0 && kernbuf[count - 1] == '\n')
2279                 len = count - 1;
2280
2281         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2282             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2283                 /* empty string is special case */
2284                 spin_lock(&squash->rsi_lock);
2285                 if (!list_empty(&squash->rsi_nosquash_nids))
2286                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2287                 spin_unlock(&squash->rsi_lock);
2288                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2289                 OBD_FREE(kernbuf, count + 1);
2290                 RETURN(count);
2291         }
2292
2293         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2294                 errmsg = "can't parse";
2295                 GOTO(failed, rc = -EINVAL);
2296         }
2297         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2298                       name, kernbuf);
2299         OBD_FREE(kernbuf, count + 1);
2300         kernbuf = NULL;
2301
2302         spin_lock(&squash->rsi_lock);
2303         if (!list_empty(&squash->rsi_nosquash_nids))
2304                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2305         list_splice(&tmp, &squash->rsi_nosquash_nids);
2306         spin_unlock(&squash->rsi_lock);
2307
2308         RETURN(count);
2309
2310 failed:
2311         if (kernbuf) {
2312                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2313                       name, kernbuf, errmsg, rc);
2314                 OBD_FREE(kernbuf, count + 1);
2315         } else {
2316                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2317                       name, errmsg, rc);
2318         }
2319         RETURN(rc);
2320 }
2321 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2322
2323 #endif /* CONFIG_PROC_FS*/