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         stats->ls_init = ktime_get_real();
1224         spin_lock_init(&stats->ls_lock);
1225
1226         /* alloc num of counter headers */
1227         CFS_ALLOC_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1228         if (!stats->ls_cnt_header)
1229                 goto fail;
1230
1231         if ((flags & LPROCFS_STATS_FLAG_NOPERCPU) != 0) {
1232                 /* contains only one set counters */
1233                 percpusize = lprocfs_stats_counter_size(stats);
1234                 LIBCFS_ALLOC_ATOMIC(stats->ls_percpu[0], percpusize);
1235                 if (!stats->ls_percpu[0])
1236                         goto fail;
1237                 stats->ls_biggest_alloc_num = 1;
1238         } else if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0) {
1239                 /* alloc all percpu data, currently only obd_memory use this */
1240                 for (i = 0; i < num_entry; ++i)
1241                         if (lprocfs_stats_alloc_one(stats, i) < 0)
1242                                 goto fail;
1243         }
1244
1245         return stats;
1246
1247 fail:
1248         lprocfs_free_stats(&stats);
1249         return NULL;
1250 }
1251 EXPORT_SYMBOL(lprocfs_alloc_stats);
1252
1253 void lprocfs_free_stats(struct lprocfs_stats **statsh)
1254 {
1255         struct lprocfs_stats *stats = *statsh;
1256         unsigned int num_entry;
1257         unsigned int percpusize;
1258         unsigned int i;
1259
1260         if (!stats || stats->ls_num == 0)
1261                 return;
1262         *statsh = NULL;
1263
1264         if (stats->ls_flags & LPROCFS_STATS_FLAG_NOPERCPU)
1265                 num_entry = 1;
1266         else
1267                 num_entry = num_possible_cpus();
1268
1269         percpusize = lprocfs_stats_counter_size(stats);
1270         for (i = 0; i < num_entry; i++)
1271                 if (stats->ls_percpu[i])
1272                         LIBCFS_FREE(stats->ls_percpu[i], percpusize);
1273
1274         if (stats->ls_cnt_header) {
1275                 for (i = 0; i < stats->ls_num; i++)
1276                         if (stats->ls_cnt_header[i].lc_hist != NULL)
1277                                 CFS_FREE_PTR(stats->ls_cnt_header[i].lc_hist);
1278                 CFS_FREE_PTR_ARRAY(stats->ls_cnt_header, stats->ls_num);
1279         }
1280
1281         LIBCFS_FREE(stats, offsetof(typeof(*stats), ls_percpu[num_entry]));
1282 }
1283 EXPORT_SYMBOL(lprocfs_free_stats);
1284
1285 u64 lprocfs_stats_collector(struct lprocfs_stats *stats, int idx,
1286                             enum lprocfs_fields_flags field)
1287 {
1288         unsigned long flags = 0;
1289         unsigned int num_cpu;
1290         unsigned int i;
1291         u64 ret = 0;
1292
1293         LASSERT(stats);
1294
1295         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1296         for (i = 0; i < num_cpu; i++) {
1297                 struct lprocfs_counter *cntr;
1298
1299                 if (!stats->ls_percpu[i])
1300                         continue;
1301
1302                 cntr = lprocfs_stats_counter_get(stats, i, idx);
1303                 ret += lprocfs_read_helper(cntr, &stats->ls_cnt_header[idx],
1304                                            stats->ls_flags, field);
1305         }
1306         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1307         return ret;
1308 }
1309 EXPORT_SYMBOL(lprocfs_stats_collector);
1310
1311 void lprocfs_clear_stats(struct lprocfs_stats *stats)
1312 {
1313         struct lprocfs_counter *percpu_cntr;
1314         int i, j;
1315         unsigned int num_entry;
1316         unsigned long flags = 0;
1317
1318         num_entry = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1319
1320         /* clear histogram if exists */
1321         for (j = 0; j < stats->ls_num; j++) {
1322                 struct obd_histogram *hist = stats->ls_cnt_header[j].lc_hist;
1323
1324                 if (hist != NULL)
1325                         lprocfs_oh_clear(hist);
1326         }
1327
1328         for (i = 0; i < num_entry; i++) {
1329                 if (!stats->ls_percpu[i])
1330                         continue;
1331                 for (j = 0; j < stats->ls_num; j++) {
1332                         percpu_cntr = lprocfs_stats_counter_get(stats, i, j);
1333                         percpu_cntr->lc_count           = 0;
1334                         percpu_cntr->lc_min             = LC_MIN_INIT;
1335                         percpu_cntr->lc_max             = 0;
1336                         percpu_cntr->lc_sumsquare       = 0;
1337                         percpu_cntr->lc_sum             = 0;
1338                         if (stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE)
1339                                 percpu_cntr->lc_sum_irq = 0;
1340                 }
1341         }
1342         stats->ls_init = ktime_get_real();
1343
1344         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1345 }
1346 EXPORT_SYMBOL(lprocfs_clear_stats);
1347
1348 static ssize_t lprocfs_stats_seq_write(struct file *file,
1349                                        const char __user *buf,
1350                                        size_t len, loff_t *off)
1351 {
1352         struct seq_file *seq = file->private_data;
1353         struct lprocfs_stats *stats = seq->private;
1354
1355         lprocfs_clear_stats(stats);
1356
1357         return len;
1358 }
1359
1360 static void *lprocfs_stats_seq_start(struct seq_file *p, loff_t *pos)
1361 {
1362         struct lprocfs_stats *stats = p->private;
1363
1364         return (*pos < stats->ls_num) ? pos : NULL;
1365 }
1366
1367 static void lprocfs_stats_seq_stop(struct seq_file *p, void *v)
1368 {
1369 }
1370
1371 static void *lprocfs_stats_seq_next(struct seq_file *p, void *v, loff_t *pos)
1372 {
1373         (*pos)++;
1374
1375         return lprocfs_stats_seq_start(p, pos);
1376 }
1377
1378 /**
1379  * print header of stats including snapshot_time, start_time and elapsed_time.
1380  *
1381  * \param seq           the file to print content to
1382  * \param now           end time to calculate elapsed_time
1383  * \param ts_init       start time to calculate elapsed_time
1384  * \param width         the width of key to align them well
1385  * \param colon         "" or ":"
1386  * \param show_units    show units or not
1387  * \param prefix        prefix (indent) before printing each line of header
1388  *                      to align them with other content
1389  */
1390 void lprocfs_stats_header(struct seq_file *seq, ktime_t now, ktime_t ts_init,
1391                           int width, const char *colon, bool show_units,
1392                           const char *prefix)
1393 {
1394         const char *units = show_units ? " secs.nsecs" : "";
1395         struct timespec64 ts;
1396         const char *field;
1397
1398         field = (colon && colon[0]) ? "snapshot_time:" : "snapshot_time";
1399         ts = ktime_to_timespec64(now);
1400         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1401                    (s64)ts.tv_sec, ts.tv_nsec, units);
1402
1403         field = (colon && colon[0]) ? "start_time:" : "start_time";
1404         ts = ktime_to_timespec64(ts_init);
1405         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1406                    (s64)ts.tv_sec, ts.tv_nsec, units);
1407
1408         field = (colon && colon[0]) ? "elapsed_time:" : "elapsed_time";
1409         ts = ktime_to_timespec64(ktime_sub(now, ts_init));
1410         seq_printf(seq, "%s%-*s %llu.%09lu%s\n", prefix, width, field,
1411                    (s64)ts.tv_sec, ts.tv_nsec, units);
1412 }
1413 EXPORT_SYMBOL(lprocfs_stats_header);
1414
1415 /* seq file export of one lprocfs counter */
1416 static int lprocfs_stats_seq_show(struct seq_file *p, void *v)
1417 {
1418         struct lprocfs_stats *stats = p->private;
1419         struct lprocfs_counter_header *hdr;
1420         struct lprocfs_counter ctr;
1421         int idx = *(loff_t *)v;
1422
1423         if (idx == 0)
1424                 lprocfs_stats_header(p, ktime_get_real(), stats->ls_init, 25,
1425                                      "", true, "");
1426
1427         hdr = &stats->ls_cnt_header[idx];
1428         lprocfs_stats_collect(stats, idx, &ctr);
1429
1430         if (ctr.lc_count == 0)
1431                 return 0;
1432
1433         seq_printf(p, "%-25s %lld samples [%s]", hdr->lc_name,
1434                    ctr.lc_count, hdr->lc_units);
1435
1436         if ((hdr->lc_config & LPROCFS_CNTR_AVGMINMAX) && ctr.lc_count > 0) {
1437                 seq_printf(p, " %lld %lld %lld",
1438                            ctr.lc_min, ctr.lc_max, ctr.lc_sum);
1439                 if (hdr->lc_config & LPROCFS_CNTR_STDDEV)
1440                         seq_printf(p, " %llu", ctr.lc_sumsquare);
1441         }
1442         seq_putc(p, '\n');
1443         return 0;
1444 }
1445
1446 static const struct seq_operations lprocfs_stats_seq_sops = {
1447         .start  = lprocfs_stats_seq_start,
1448         .stop   = lprocfs_stats_seq_stop,
1449         .next   = lprocfs_stats_seq_next,
1450         .show   = lprocfs_stats_seq_show,
1451 };
1452
1453 static int lprocfs_stats_seq_open(struct inode *inode, struct file *file)
1454 {
1455         struct seq_file *seq;
1456         int rc;
1457
1458         rc = seq_open(file, &lprocfs_stats_seq_sops);
1459         if (rc)
1460                 return rc;
1461         seq = file->private_data;
1462         seq->private = inode->i_private ? inode->i_private : PDE_DATA(inode);
1463         return 0;
1464 }
1465
1466 const struct file_operations ldebugfs_stats_seq_fops = {
1467         .owner   = THIS_MODULE,
1468         .open    = lprocfs_stats_seq_open,
1469         .read    = seq_read,
1470         .write   = lprocfs_stats_seq_write,
1471         .llseek  = seq_lseek,
1472         .release = lprocfs_seq_release,
1473 };
1474 EXPORT_SYMBOL(ldebugfs_stats_seq_fops);
1475
1476 static const struct proc_ops lprocfs_stats_seq_fops = {
1477         PROC_OWNER(THIS_MODULE)
1478         .proc_open      = lprocfs_stats_seq_open,
1479         .proc_read      = seq_read,
1480         .proc_write     = lprocfs_stats_seq_write,
1481         .proc_lseek     = seq_lseek,
1482         .proc_release   = lprocfs_seq_release,
1483 };
1484
1485 int lprocfs_register_stats(struct proc_dir_entry *root, const char *name,
1486                            struct lprocfs_stats *stats)
1487 {
1488         struct proc_dir_entry *entry;
1489         LASSERT(root != NULL);
1490
1491         entry = proc_create_data(name, 0644, root,
1492                                  &lprocfs_stats_seq_fops, stats);
1493         if (!entry)
1494                 return -ENOMEM;
1495         return 0;
1496 }
1497 EXPORT_SYMBOL(lprocfs_register_stats);
1498
1499 static const char *lprocfs_counter_config_units(const char *name,
1500                                          enum lprocfs_counter_config config)
1501 {
1502         const char *units;
1503
1504         switch (config & LPROCFS_TYPE_MASK) {
1505         default:
1506                 units = "reqs"; break;
1507         case LPROCFS_TYPE_BYTES:
1508                 units = "bytes"; break;
1509         case LPROCFS_TYPE_PAGES:
1510                 units = "pages"; break;
1511         case LPROCFS_TYPE_LOCKS:
1512                 units = "locks"; break;
1513         case LPROCFS_TYPE_LOCKSPS:
1514                 units = "locks/s"; break;
1515         case LPROCFS_TYPE_SECS:
1516                 units = "secs"; break;
1517         case LPROCFS_TYPE_USECS:
1518                 units = "usecs"; break;
1519         }
1520
1521         return units;
1522 }
1523
1524 void lprocfs_counter_init_units(struct lprocfs_stats *stats, int index,
1525                                 enum lprocfs_counter_config config,
1526                                 const char *name, const char *units)
1527 {
1528         struct lprocfs_counter_header *header;
1529         struct lprocfs_counter *percpu_cntr;
1530         unsigned long flags = 0;
1531         unsigned int i;
1532         unsigned int num_cpu;
1533
1534         LASSERT(stats != NULL);
1535
1536         header = &stats->ls_cnt_header[index];
1537         LASSERTF(header != NULL, "Failed to allocate stats header:[%d]%s/%s\n",
1538                  index, name, units);
1539
1540         header->lc_config = config;
1541         header->lc_name = name;
1542         header->lc_units = units;
1543
1544         if (config & LPROCFS_CNTR_HISTOGRAM) {
1545                 CFS_ALLOC_PTR(stats->ls_cnt_header[index].lc_hist);
1546                 if (stats->ls_cnt_header[index].lc_hist == NULL)
1547                         CERROR("LprocFS: Failed to allocate histogram:[%d]%s/%s\n",
1548                                index, name, units);
1549                 else
1550                         spin_lock_init(&stats->ls_cnt_header[index].lc_hist->oh_lock);
1551         }
1552         num_cpu = lprocfs_stats_lock(stats, LPROCFS_GET_NUM_CPU, &flags);
1553         for (i = 0; i < num_cpu; ++i) {
1554                 if (!stats->ls_percpu[i])
1555                         continue;
1556                 percpu_cntr = lprocfs_stats_counter_get(stats, i, index);
1557                 percpu_cntr->lc_count           = 0;
1558                 percpu_cntr->lc_min             = LC_MIN_INIT;
1559                 percpu_cntr->lc_max             = 0;
1560                 percpu_cntr->lc_sumsquare       = 0;
1561                 percpu_cntr->lc_sum             = 0;
1562                 if ((stats->ls_flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1563                         percpu_cntr->lc_sum_irq = 0;
1564         }
1565         lprocfs_stats_unlock(stats, LPROCFS_GET_NUM_CPU, &flags);
1566 }
1567 EXPORT_SYMBOL(lprocfs_counter_init_units);
1568
1569 void lprocfs_counter_init(struct lprocfs_stats *stats, int index,
1570                           enum lprocfs_counter_config config,
1571                           const char *name)
1572 {
1573         lprocfs_counter_init_units(stats, index, config, name,
1574                                    lprocfs_counter_config_units(name, config));
1575 }
1576 EXPORT_SYMBOL(lprocfs_counter_init);
1577
1578 static const char * const mps_stats[] = {
1579         [LPROC_MD_CLOSE]                = "close",
1580         [LPROC_MD_CREATE]               = "create",
1581         [LPROC_MD_ENQUEUE]              = "enqueue",
1582         [LPROC_MD_GETATTR]              = "getattr",
1583         [LPROC_MD_INTENT_LOCK]          = "intent_lock",
1584         [LPROC_MD_LINK]                 = "link",
1585         [LPROC_MD_RENAME]               = "rename",
1586         [LPROC_MD_SETATTR]              = "setattr",
1587         [LPROC_MD_FSYNC]                = "fsync",
1588         [LPROC_MD_READ_PAGE]            = "read_page",
1589         [LPROC_MD_UNLINK]               = "unlink",
1590         [LPROC_MD_SETXATTR]             = "setxattr",
1591         [LPROC_MD_GETXATTR]             = "getxattr",
1592         [LPROC_MD_INTENT_GETATTR_ASYNC] = "intent_getattr_async",
1593         [LPROC_MD_REVALIDATE_LOCK]      = "revalidate_lock",
1594 };
1595
1596 int lprocfs_alloc_md_stats(struct obd_device *obd,
1597                            unsigned int num_private_stats)
1598 {
1599         struct lprocfs_stats *stats;
1600         unsigned int num_stats;
1601         int rc, i;
1602
1603         /*
1604          * TODO Ensure that this function is only used where
1605          * appropriate by adding an assertion to the effect that
1606          * obd->obd_type->typ_md_ops is not NULL. We can't do this now
1607          * because mdt_procfs_init() uses this function to allocate
1608          * the stats backing /proc/fs/lustre/mdt/.../md_stats but the
1609          * mdt layer does not use the md_ops interface. This is
1610          * confusing and a waste of memory. See LU-2484.
1611          */
1612         LASSERT(obd->obd_proc_entry != NULL);
1613         LASSERT(obd->obd_md_stats == NULL);
1614
1615         num_stats = ARRAY_SIZE(mps_stats) + num_private_stats;
1616         stats = lprocfs_alloc_stats(num_stats, 0);
1617         if (!stats)
1618                 return -ENOMEM;
1619
1620         for (i = 0; i < ARRAY_SIZE(mps_stats); i++) {
1621                 lprocfs_counter_init(stats, i, LPROCFS_TYPE_REQS,
1622                                      mps_stats[i]);
1623                 if (!stats->ls_cnt_header[i].lc_name) {
1624                         CERROR("Missing md_stat initializer md_op operation at offset %d. Aborting.\n",
1625                                i);
1626                         LBUG();
1627                 }
1628         }
1629
1630         rc = lprocfs_register_stats(obd->obd_proc_entry, "md_stats", stats);
1631         if (rc < 0) {
1632                 lprocfs_free_stats(&stats);
1633         } else {
1634                 obd->obd_md_stats = stats;
1635         }
1636
1637         return rc;
1638 }
1639 EXPORT_SYMBOL(lprocfs_alloc_md_stats);
1640
1641 void lprocfs_free_md_stats(struct obd_device *obd)
1642 {
1643         struct lprocfs_stats *stats = obd->obd_md_stats;
1644
1645         if (stats) {
1646                 obd->obd_md_stats = NULL;
1647                 lprocfs_free_stats(&stats);
1648         }
1649 }
1650 EXPORT_SYMBOL(lprocfs_free_md_stats);
1651
1652 void lprocfs_init_ldlm_stats(struct lprocfs_stats *ldlm_stats)
1653 {
1654         lprocfs_counter_init(ldlm_stats, LDLM_ENQUEUE - LDLM_FIRST_OPC,
1655                              LPROCFS_TYPE_REQS, "ldlm_enqueue");
1656         lprocfs_counter_init(ldlm_stats, LDLM_CONVERT - LDLM_FIRST_OPC,
1657                              LPROCFS_TYPE_REQS, "ldlm_convert");
1658         lprocfs_counter_init(ldlm_stats, LDLM_CANCEL - LDLM_FIRST_OPC,
1659                              LPROCFS_TYPE_REQS, "ldlm_cancel");
1660         lprocfs_counter_init(ldlm_stats, LDLM_BL_CALLBACK - LDLM_FIRST_OPC,
1661                              LPROCFS_TYPE_REQS, "ldlm_bl_callback");
1662         lprocfs_counter_init(ldlm_stats, LDLM_CP_CALLBACK - LDLM_FIRST_OPC,
1663                              LPROCFS_TYPE_REQS, "ldlm_cp_callback");
1664         lprocfs_counter_init(ldlm_stats, LDLM_GL_CALLBACK - LDLM_FIRST_OPC,
1665                              LPROCFS_TYPE_REQS, "ldlm_gl_callback");
1666 }
1667 EXPORT_SYMBOL(lprocfs_init_ldlm_stats);
1668
1669 __s64 lprocfs_read_helper(struct lprocfs_counter *lc,
1670                           struct lprocfs_counter_header *header,
1671                           enum lprocfs_stats_flags flags,
1672                           enum lprocfs_fields_flags field)
1673 {
1674         __s64 ret = 0;
1675
1676         if (!lc || !header)
1677                 RETURN(0);
1678
1679         switch (field) {
1680                 case LPROCFS_FIELDS_FLAGS_CONFIG:
1681                         ret = header->lc_config;
1682                         break;
1683                 case LPROCFS_FIELDS_FLAGS_SUM:
1684                         ret = lc->lc_sum;
1685                         if ((flags & LPROCFS_STATS_FLAG_IRQ_SAFE) != 0)
1686                                 ret += lc->lc_sum_irq;
1687                         break;
1688                 case LPROCFS_FIELDS_FLAGS_MIN:
1689                         ret = lc->lc_min;
1690                         break;
1691                 case LPROCFS_FIELDS_FLAGS_MAX:
1692                         ret = lc->lc_max;
1693                         break;
1694                 case LPROCFS_FIELDS_FLAGS_AVG:
1695                         ret = div64_u64((flags & LPROCFS_STATS_FLAG_IRQ_SAFE ?
1696                                          lc->lc_sum_irq : 0) + lc->lc_sum,
1697                                         lc->lc_count);
1698                         break;
1699                 case LPROCFS_FIELDS_FLAGS_SUMSQUARE:
1700                         ret = lc->lc_sumsquare;
1701                         break;
1702                 case LPROCFS_FIELDS_FLAGS_COUNT:
1703                         ret = lc->lc_count;
1704                         break;
1705                 default:
1706                         break;
1707         };
1708         RETURN(ret);
1709 }
1710 EXPORT_SYMBOL(lprocfs_read_helper);
1711
1712 /**
1713  * string_to_size - convert ASCII string representing a numerical
1714  *                  value with optional units to 64-bit binary value
1715  *
1716  * @size:       The numerical value extract out of @buffer
1717  * @buffer:     passed in string to parse
1718  * @count:      length of the @buffer
1719  *
1720  * This function returns a 64-bit binary value if @buffer contains a valid
1721  * numerical string. The string is parsed to 3 significant figures after
1722  * the decimal point. Support the string containing an optional units at
1723  * the end which can be base 2 or base 10 in value. If no units are given
1724  * the string is assumed to just a numerical value.
1725  *
1726  * Returns:     @count if the string is successfully parsed,
1727  *              -errno on invalid input strings. Error values:
1728  *
1729  *  - ``-EINVAL``: @buffer is not a proper numerical string
1730  *  - ``-EOVERFLOW``: results does not fit into 64 bits.
1731  *  - ``-E2BIG ``: @buffer is too large (not a valid number)
1732  */
1733 int string_to_size(u64 *size, const char *buffer, size_t count)
1734 {
1735         /* For string_get_size() it can support values above exabytes,
1736          * (ZiB, YiB) due to breaking the return value into a size and
1737          * bulk size to avoid 64 bit overflow. We don't break the size
1738          * up into block size units so we don't support ZiB or YiB.
1739          */
1740         static const char *const units_10[] = {
1741                 "kB", "MB", "GB", "TB", "PB", "EB",
1742         };
1743         static const char *const units_2[] = {
1744                 "K",  "M",  "G",  "T",  "P",  "E",
1745         };
1746         static const char *const *const units_str[] = {
1747                 [STRING_UNITS_2] = units_2,
1748                 [STRING_UNITS_10] = units_10,
1749         };
1750         static const unsigned int coeff[] = {
1751                 [STRING_UNITS_10] = 1000,
1752                 [STRING_UNITS_2] = 1024,
1753         };
1754         enum string_size_units unit = STRING_UNITS_2;
1755         u64 whole, blk_size = 1;
1756         char kernbuf[22], *end;
1757         size_t len = count;
1758         int rc;
1759         int i;
1760
1761         if (count >= sizeof(kernbuf)) {
1762                 CERROR("count %zd > buffer %zd\n", count, sizeof(kernbuf));
1763                 return -E2BIG;
1764         }
1765
1766         *size = 0;
1767         /* The "iB" suffix is optionally allowed for indicating base-2 numbers.
1768          * If suffix is only "B" and not "iB" then we treat it as base-10.
1769          */
1770         end = strstr(buffer, "B");
1771         if (end && *(end - 1) != 'i')
1772                 unit = STRING_UNITS_10;
1773
1774         i = unit == STRING_UNITS_2 ? ARRAY_SIZE(units_2) - 1 :
1775                                      ARRAY_SIZE(units_10) - 1;
1776         do {
1777                 end = strnstr(buffer, units_str[unit][i], count);
1778                 if (end) {
1779                         for (; i >= 0; i--)
1780                                 blk_size *= coeff[unit];
1781                         len = end - buffer;
1782                         break;
1783                 }
1784         } while (i--);
1785
1786         /* as 'B' is a substring of all units, we need to handle it
1787          * separately.
1788          */
1789         if (!end) {
1790                 /* 'B' is only acceptable letter at this point */
1791                 end = strnchr(buffer, count, 'B');
1792                 if (end) {
1793                         len = end - buffer;
1794
1795                         if (count - len > 2 ||
1796                             (count - len == 2 && strcmp(end, "B\n") != 0)) {
1797                                 CDEBUG(D_INFO, "unknown suffix '%s'\n", buffer);
1798                                 return -EINVAL;
1799                         }
1800                 }
1801                 /* kstrtoull will error out if it has non digits */
1802                 goto numbers_only;
1803         }
1804
1805         end = strnchr(buffer, count, '.');
1806         if (end) {
1807                 /* need to limit 3 decimal places */
1808                 char rem[4] = "000";
1809                 u64 frac = 0;
1810                 size_t off;
1811
1812                 len = end - buffer;
1813                 end++;
1814
1815                 /* limit to 3 decimal points */
1816                 off = min_t(size_t, 3, strspn(end, "0123456789"));
1817                 /* need to limit frac_d to a u32 */
1818                 memcpy(rem, end, off);
1819                 rc = kstrtoull(rem, 10, &frac);
1820                 if (rc)
1821                         return rc;
1822
1823                 if (fls64(frac) + fls64(blk_size) - 1 > 64)
1824                         return -EOVERFLOW;
1825
1826                 frac *= blk_size;
1827                 do_div(frac, 1000);
1828                 *size += frac;
1829         }
1830 numbers_only:
1831         snprintf(kernbuf, sizeof(kernbuf), "%.*s", (int)len, buffer);
1832         rc = kstrtoull(kernbuf, 10, &whole);
1833         if (rc)
1834                 return rc;
1835
1836         if (whole != 0 && fls64(whole) + fls64(blk_size) - 1 > 64)
1837                 return -EOVERFLOW;
1838
1839         *size += whole * blk_size;
1840
1841         return count;
1842 }
1843 EXPORT_SYMBOL(string_to_size);
1844
1845 /**
1846  * sysfs_memparse - parse a ASCII string to 64-bit binary value,
1847  *                  with optional units
1848  *
1849  * @buffer:     kernel pointer to input string
1850  * @count:      number of bytes in the input @buffer
1851  * @val:        (output) binary value returned to caller
1852  * @defunit:    default unit suffix to use if none is provided
1853  *
1854  * Parses a string into a number. The number stored at @buffer is
1855  * potentially suffixed with K, M, G, T, P, E. Besides these other
1856  * valid suffix units are shown in the string_to_size() function.
1857  * If the string lacks a suffix then the defunit is used. The defunit
1858  * should be given as a binary unit (e.g. MiB) as that is the standard
1859  * for tunables in Lustre. If no unit suffix is given (e.g. 'G'), then
1860  * it is assumed to be in binary units.
1861  *
1862  * Returns:     0 on success or -errno on failure.
1863  */
1864 int sysfs_memparse(const char *buffer, size_t count, u64 *val,
1865                    const char *defunit)
1866 {
1867         const char *param = buffer;
1868         char tmp_buf[23];
1869         int rc;
1870
1871         count = strlen(buffer);
1872         while (count > 0 && isspace(buffer[count - 1]))
1873                 count--;
1874
1875         if (!count)
1876                 RETURN(-EINVAL);
1877
1878         /* If there isn't already a unit on this value, append @defunit.
1879          * Units of 'B' don't affect the value, so don't bother adding.
1880          */
1881         if (!isalpha(buffer[count - 1]) && defunit[0] != 'B') {
1882                 if (count + 3 >= sizeof(tmp_buf)) {
1883                         CERROR("count %zd > size %zd\n", count, sizeof(param));
1884                         RETURN(-E2BIG);
1885                 }
1886
1887                 scnprintf(tmp_buf, sizeof(tmp_buf), "%.*s%s", (int)count,
1888                           buffer, defunit);
1889                 param = tmp_buf;
1890                 count = strlen(param);
1891         }
1892
1893         rc = string_to_size(val, param, count);
1894
1895         return rc < 0 ? rc : 0;
1896 }
1897 EXPORT_SYMBOL(sysfs_memparse);
1898
1899 char *lprocfs_strnstr(const char *s1, const char *s2, size_t len)
1900 {
1901         size_t l2;
1902
1903         l2 = strlen(s2);
1904         if (!l2)
1905                 return (char *)s1;
1906         while (len >= l2) {
1907                 len--;
1908                 if (!memcmp(s1, s2, l2))
1909                         return (char *)s1;
1910                 s1++;
1911         }
1912         return NULL;
1913 }
1914 EXPORT_SYMBOL(lprocfs_strnstr);
1915
1916 /**
1917  * Find the string \a name in the input \a buffer, and return a pointer to the
1918  * value immediately following \a name, reducing \a count appropriately.
1919  * If \a name is not found the original \a buffer is returned.
1920  */
1921 char *lprocfs_find_named_value(const char *buffer, const char *name,
1922                                 size_t *count)
1923 {
1924         char *val;
1925         size_t buflen = *count;
1926
1927         /* there is no strnstr() in rhel5 and ubuntu kernels */
1928         val = lprocfs_strnstr(buffer, name, buflen);
1929         if (!val)
1930                 return (char *)buffer;
1931
1932         val += strlen(name);                             /* skip prefix */
1933         while (val < buffer + buflen && isspace(*val)) /* skip separator */
1934                 val++;
1935
1936         *count = 0;
1937         while (val < buffer + buflen && isalnum(*val)) {
1938                 ++*count;
1939                 ++val;
1940         }
1941
1942         return val - *count;
1943 }
1944 EXPORT_SYMBOL(lprocfs_find_named_value);
1945
1946 int lprocfs_seq_create(struct proc_dir_entry *parent,
1947                        const char *name,
1948                        mode_t mode,
1949                        const struct proc_ops *seq_fops,
1950                        void *data)
1951 {
1952         struct proc_dir_entry *entry;
1953         ENTRY;
1954
1955         /* Disallow secretly (un)writable entries. */
1956         LASSERT(!seq_fops->proc_write == !(mode & 0222));
1957
1958         entry = proc_create_data(name, mode, parent, seq_fops, data);
1959
1960         if (!entry)
1961                 RETURN(-ENOMEM);
1962
1963         RETURN(0);
1964 }
1965 EXPORT_SYMBOL(lprocfs_seq_create);
1966
1967 int lprocfs_obd_seq_create(struct obd_device *obd,
1968                            const char *name,
1969                            mode_t mode,
1970                            const struct proc_ops *seq_fops,
1971                            void *data)
1972 {
1973         return lprocfs_seq_create(obd->obd_proc_entry, name,
1974                                   mode, seq_fops, data);
1975 }
1976 EXPORT_SYMBOL(lprocfs_obd_seq_create);
1977
1978 void lprocfs_oh_tally(struct obd_histogram *oh, unsigned int value)
1979 {
1980         if (value >= OBD_HIST_MAX)
1981                 value = OBD_HIST_MAX - 1;
1982
1983         spin_lock(&oh->oh_lock);
1984         oh->oh_buckets[value]++;
1985         spin_unlock(&oh->oh_lock);
1986 }
1987 EXPORT_SYMBOL(lprocfs_oh_tally);
1988
1989 void lprocfs_oh_tally_log2(struct obd_histogram *oh, unsigned int value)
1990 {
1991         unsigned int val = 0;
1992
1993         if (likely(value != 0))
1994                 val = min(fls(value - 1), OBD_HIST_MAX);
1995
1996         lprocfs_oh_tally(oh, val);
1997 }
1998 EXPORT_SYMBOL(lprocfs_oh_tally_log2);
1999
2000 unsigned long lprocfs_oh_sum(struct obd_histogram *oh)
2001 {
2002         unsigned long ret = 0;
2003         int i;
2004
2005         for (i = 0; i < OBD_HIST_MAX; i++)
2006                 ret +=  oh->oh_buckets[i];
2007         return ret;
2008 }
2009 EXPORT_SYMBOL(lprocfs_oh_sum);
2010
2011 void lprocfs_oh_clear(struct obd_histogram *oh)
2012 {
2013         spin_lock(&oh->oh_lock);
2014         memset(oh->oh_buckets, 0, sizeof(oh->oh_buckets));
2015         spin_unlock(&oh->oh_lock);
2016 }
2017 EXPORT_SYMBOL(lprocfs_oh_clear);
2018
2019 void lprocfs_oh_tally_pcpu(struct obd_hist_pcpu *oh,
2020                            unsigned int value)
2021 {
2022         if (value >= OBD_HIST_MAX)
2023                 value = OBD_HIST_MAX - 1;
2024
2025         percpu_counter_inc(&oh->oh_pc_buckets[value]);
2026 }
2027 EXPORT_SYMBOL(lprocfs_oh_tally_pcpu);
2028
2029 void lprocfs_oh_tally_log2_pcpu(struct obd_hist_pcpu *oh,
2030                                 unsigned int value)
2031 {
2032         unsigned int val = 0;
2033
2034         if (likely(value != 0))
2035                 val = min(fls(value - 1), OBD_HIST_MAX);
2036
2037         lprocfs_oh_tally_pcpu(oh, val);
2038 }
2039 EXPORT_SYMBOL(lprocfs_oh_tally_log2_pcpu);
2040
2041 unsigned long lprocfs_oh_counter_pcpu(struct obd_hist_pcpu *oh,
2042                                       unsigned int value)
2043 {
2044         return percpu_counter_sum(&oh->oh_pc_buckets[value]);
2045 }
2046 EXPORT_SYMBOL(lprocfs_oh_counter_pcpu);
2047
2048 unsigned long lprocfs_oh_sum_pcpu(struct obd_hist_pcpu *oh)
2049 {
2050         unsigned long ret = 0;
2051         int i;
2052
2053         for (i = 0; i < OBD_HIST_MAX; i++)
2054                 ret += percpu_counter_sum(&oh->oh_pc_buckets[i]);
2055
2056         return ret;
2057 }
2058 EXPORT_SYMBOL(lprocfs_oh_sum_pcpu);
2059
2060 int lprocfs_oh_alloc_pcpu(struct obd_hist_pcpu *oh)
2061 {
2062         int i, rc;
2063
2064         if (oh->oh_initialized)
2065                 return 0;
2066
2067         for (i = 0; i < OBD_HIST_MAX; i++) {
2068                 rc = percpu_counter_init(&oh->oh_pc_buckets[i], 0, GFP_KERNEL);
2069                 if (rc)
2070                         goto out;
2071         }
2072
2073         oh->oh_initialized = true;
2074
2075         return 0;
2076
2077 out:
2078         for (i--; i >= 0; i--)
2079                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2080
2081         return rc;
2082 }
2083 EXPORT_SYMBOL(lprocfs_oh_alloc_pcpu);
2084
2085 void lprocfs_oh_clear_pcpu(struct obd_hist_pcpu *oh)
2086 {
2087         int i;
2088
2089         for (i = 0; i < OBD_HIST_MAX; i++)
2090                 percpu_counter_set(&oh->oh_pc_buckets[i], 0);
2091 }
2092 EXPORT_SYMBOL(lprocfs_oh_clear_pcpu);
2093
2094 void lprocfs_oh_release_pcpu(struct obd_hist_pcpu *oh)
2095 {
2096         int i;
2097
2098         if (!oh->oh_initialized)
2099                 return;
2100
2101         for (i = 0; i < OBD_HIST_MAX; i++)
2102                 percpu_counter_destroy(&oh->oh_pc_buckets[i]);
2103
2104         oh->oh_initialized = false;
2105 }
2106 EXPORT_SYMBOL(lprocfs_oh_release_pcpu);
2107
2108 ssize_t lustre_attr_show(struct kobject *kobj,
2109                          struct attribute *attr, char *buf)
2110 {
2111         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2112
2113         return a->show ? a->show(kobj, attr, buf) : 0;
2114 }
2115 EXPORT_SYMBOL_GPL(lustre_attr_show);
2116
2117 ssize_t lustre_attr_store(struct kobject *kobj, struct attribute *attr,
2118                           const char *buf, size_t len)
2119 {
2120         struct lustre_attr *a = container_of(attr, struct lustre_attr, attr);
2121
2122         return a->store ? a->store(kobj, attr, buf, len) : len;
2123 }
2124 EXPORT_SYMBOL_GPL(lustre_attr_store);
2125
2126 const struct sysfs_ops lustre_sysfs_ops = {
2127         .show  = lustre_attr_show,
2128         .store = lustre_attr_store,
2129 };
2130 EXPORT_SYMBOL_GPL(lustre_sysfs_ops);
2131
2132 int lprocfs_obd_max_pages_per_rpc_seq_show(struct seq_file *m, void *data)
2133 {
2134         struct obd_device *obd = data;
2135         struct client_obd *cli = &obd->u.cli;
2136
2137         spin_lock(&cli->cl_loi_list_lock);
2138         seq_printf(m, "%d\n", cli->cl_max_pages_per_rpc);
2139         spin_unlock(&cli->cl_loi_list_lock);
2140         return 0;
2141 }
2142 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_show);
2143
2144 ssize_t lprocfs_obd_max_pages_per_rpc_seq_write(struct file *file,
2145                                                 const char __user *buffer,
2146                                                 size_t count, loff_t *off)
2147 {
2148         struct seq_file *m = file->private_data;
2149         struct obd_device *obd = m->private;
2150         struct client_obd *cli = &obd->u.cli;
2151         struct obd_import *imp;
2152         struct obd_connect_data *ocd;
2153         int chunk_mask, rc;
2154         char kernbuf[22];
2155         u64 val;
2156
2157         if (count > sizeof(kernbuf) - 1)
2158                 return -EINVAL;
2159
2160         if (copy_from_user(kernbuf, buffer, count))
2161                 return -EFAULT;
2162
2163         kernbuf[count] = '\0';
2164
2165         rc = sysfs_memparse(kernbuf, count, &val, "B");
2166         if (rc)
2167                 return rc;
2168
2169         /* if the max_pages is specified in bytes, convert to pages */
2170         if (val >= ONE_MB_BRW_SIZE)
2171                 val >>= PAGE_SHIFT;
2172
2173         with_imp_locked(obd, imp, rc) {
2174                 ocd = &imp->imp_connect_data;
2175                 chunk_mask = ~((1 << (cli->cl_chunkbits - PAGE_SHIFT)) - 1);
2176                 /* max_pages_per_rpc must be chunk aligned */
2177                 val = (val + ~chunk_mask) & chunk_mask;
2178                 if (val == 0 || (ocd->ocd_brw_size != 0 &&
2179                                  val > ocd->ocd_brw_size >> PAGE_SHIFT)) {
2180                         rc = -ERANGE;
2181                 } else {
2182                         spin_lock(&cli->cl_loi_list_lock);
2183                         cli->cl_max_pages_per_rpc = val;
2184                         client_adjust_max_dirty(cli);
2185                         spin_unlock(&cli->cl_loi_list_lock);
2186                 }
2187         }
2188
2189         return rc ?: count;
2190 }
2191 EXPORT_SYMBOL(lprocfs_obd_max_pages_per_rpc_seq_write);
2192
2193 ssize_t short_io_bytes_show(struct kobject *kobj, struct attribute *attr,
2194                             char *buf)
2195 {
2196         struct obd_device *obd = container_of(kobj, struct obd_device,
2197                                               obd_kset.kobj);
2198         struct client_obd *cli = &obd->u.cli;
2199         int rc;
2200
2201         spin_lock(&cli->cl_loi_list_lock);
2202         rc = sprintf(buf, "%d\n", cli->cl_max_short_io_bytes);
2203         spin_unlock(&cli->cl_loi_list_lock);
2204         return rc;
2205 }
2206 EXPORT_SYMBOL(short_io_bytes_show);
2207
2208 /* Used to catch people who think they're specifying pages. */
2209 #define MIN_SHORT_IO_BYTES 64U
2210
2211 ssize_t short_io_bytes_store(struct kobject *kobj, struct attribute *attr,
2212                              const char *buffer, size_t count)
2213 {
2214         struct obd_device *obd = container_of(kobj, struct obd_device,
2215                                               obd_kset.kobj);
2216         struct client_obd *cli = &obd->u.cli;
2217         u64 val;
2218         int rc;
2219
2220         if (strcmp(buffer, "-1") == 0) {
2221                 val = OBD_DEF_SHORT_IO_BYTES;
2222         } else {
2223                 rc = sysfs_memparse(buffer, count, &val, "B");
2224                 if (rc)
2225                         GOTO(out, rc);
2226         }
2227
2228         if (val && (val < MIN_SHORT_IO_BYTES || val > LNET_MTU))
2229                 GOTO(out, rc = -ERANGE);
2230
2231         rc = count;
2232
2233         spin_lock(&cli->cl_loi_list_lock);
2234         cli->cl_max_short_io_bytes = min_t(u64, val, OST_MAX_SHORT_IO_BYTES);
2235         spin_unlock(&cli->cl_loi_list_lock);
2236
2237 out:
2238         return rc;
2239 }
2240 EXPORT_SYMBOL(short_io_bytes_store);
2241
2242 int lprocfs_wr_root_squash(const char __user *buffer, unsigned long count,
2243                            struct root_squash_info *squash, char *name)
2244 {
2245         int rc;
2246         char kernbuf[64], *tmp, *errmsg;
2247         unsigned long uid, gid;
2248         ENTRY;
2249
2250         if (count >= sizeof(kernbuf)) {
2251                 errmsg = "string too long";
2252                 GOTO(failed_noprint, rc = -EINVAL);
2253         }
2254         if (copy_from_user(kernbuf, buffer, count)) {
2255                 errmsg = "bad address";
2256                 GOTO(failed_noprint, rc = -EFAULT);
2257         }
2258         kernbuf[count] = '\0';
2259
2260         /* look for uid gid separator */
2261         tmp = strchr(kernbuf, ':');
2262         if (!tmp) {
2263                 errmsg = "needs uid:gid format";
2264                 GOTO(failed, rc = -EINVAL);
2265         }
2266         *tmp = '\0';
2267         tmp++;
2268
2269         /* parse uid */
2270         if (kstrtoul(kernbuf, 0, &uid) != 0) {
2271                 errmsg = "bad uid";
2272                 GOTO(failed, rc = -EINVAL);
2273         }
2274
2275         /* parse gid */
2276         if (kstrtoul(tmp, 0, &gid) != 0) {
2277                 errmsg = "bad gid";
2278                 GOTO(failed, rc = -EINVAL);
2279         }
2280
2281         squash->rsi_uid = uid;
2282         squash->rsi_gid = gid;
2283
2284         LCONSOLE_INFO("%s: root_squash is set to %u:%u\n",
2285                       name, squash->rsi_uid, squash->rsi_gid);
2286         RETURN(count);
2287
2288 failed:
2289         if (tmp) {
2290                 tmp--;
2291                 *tmp = ':';
2292         }
2293         CWARN("%s: failed to set root_squash to \"%s\", %s, rc = %d\n",
2294               name, kernbuf, errmsg, rc);
2295         RETURN(rc);
2296 failed_noprint:
2297         CWARN("%s: failed to set root_squash due to %s, rc = %d\n",
2298               name, errmsg, rc);
2299         RETURN(rc);
2300 }
2301 EXPORT_SYMBOL(lprocfs_wr_root_squash);
2302
2303
2304 int lprocfs_wr_nosquash_nids(const char __user *buffer, unsigned long count,
2305                              struct root_squash_info *squash, char *name)
2306 {
2307         int rc;
2308         char *kernbuf = NULL;
2309         char *errmsg;
2310         LIST_HEAD(tmp);
2311         int len = count;
2312         ENTRY;
2313
2314         if (count > 4096) {
2315                 errmsg = "string too long";
2316                 GOTO(failed, rc = -EINVAL);
2317         }
2318
2319         OBD_ALLOC(kernbuf, count + 1);
2320         if (!kernbuf) {
2321                 errmsg = "no memory";
2322                 GOTO(failed, rc = -ENOMEM);
2323         }
2324         if (copy_from_user(kernbuf, buffer, count)) {
2325                 errmsg = "bad address";
2326                 GOTO(failed, rc = -EFAULT);
2327         }
2328         kernbuf[count] = '\0';
2329
2330         if (count > 0 && kernbuf[count - 1] == '\n')
2331                 len = count - 1;
2332
2333         if ((len == 4 && strncmp(kernbuf, "NONE", len) == 0) ||
2334             (len == 5 && strncmp(kernbuf, "clear", len) == 0)) {
2335                 /* empty string is special case */
2336                 spin_lock(&squash->rsi_lock);
2337                 if (!list_empty(&squash->rsi_nosquash_nids))
2338                         cfs_free_nidlist(&squash->rsi_nosquash_nids);
2339                 spin_unlock(&squash->rsi_lock);
2340                 LCONSOLE_INFO("%s: nosquash_nids is cleared\n", name);
2341                 OBD_FREE(kernbuf, count + 1);
2342                 RETURN(count);
2343         }
2344
2345         if (cfs_parse_nidlist(kernbuf, count, &tmp) <= 0) {
2346                 errmsg = "can't parse";
2347                 GOTO(failed, rc = -EINVAL);
2348         }
2349         LCONSOLE_INFO("%s: nosquash_nids set to %s\n",
2350                       name, kernbuf);
2351         OBD_FREE(kernbuf, count + 1);
2352         kernbuf = NULL;
2353
2354         spin_lock(&squash->rsi_lock);
2355         if (!list_empty(&squash->rsi_nosquash_nids))
2356                 cfs_free_nidlist(&squash->rsi_nosquash_nids);
2357         list_splice(&tmp, &squash->rsi_nosquash_nids);
2358         spin_unlock(&squash->rsi_lock);
2359
2360         RETURN(count);
2361
2362 failed:
2363         if (kernbuf) {
2364                 CWARN("%s: failed to set nosquash_nids to \"%s\", %s rc = %d\n",
2365                       name, kernbuf, errmsg, rc);
2366                 OBD_FREE(kernbuf, count + 1);
2367         } else {
2368                 CWARN("%s: failed to set nosquash_nids due to %s rc = %d\n",
2369                       name, errmsg, rc);
2370         }
2371         RETURN(rc);
2372 }
2373 EXPORT_SYMBOL(lprocfs_wr_nosquash_nids);
2374
2375 #endif /* CONFIG_PROC_FS*/