drivers/staging/lustre/lnet/libcfs/hash.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.sun.com/software/products/lustre/docs/GPLv2.pdf
  19  *
  20  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  21  * CA 95054 USA or visit www.sun.com if you need additional information or
  22  * have any questions.
  23  *
  24  * GPL HEADER END
  25  */
  26 /*
  27  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * libcfs/libcfs/hash.c
  37  *
  38  * Implement a hash class for hash process in lustre system.
  39  *
  40  * Author: YuZhangyong <yzy@clusterfs.com>
  41  *
  42  * 2008-08-15: Brian Behlendorf <behlendorf1@llnl.gov>
  43  * - Simplified API and improved documentation
  44  * - Added per-hash feature flags:
  45  *   * CFS_HASH_DEBUG additional validation
  46  *   * CFS_HASH_REHASH dynamic rehashing
  47  * - Added per-hash statistics
  48  * - General performance enhancements
  49  *
  50  * 2009-07-31: Liang Zhen <zhen.liang@sun.com>
  51  * - move all stuff to libcfs
  52  * - don't allow cur_bits != max_bits without setting of CFS_HASH_REHASH
  53  * - ignore hs_rwlock if without CFS_HASH_REHASH setting
  54  * - buckets are allocated one by one(instead of contiguous memory),
  55  *   to avoid unnecessary cacheline conflict
  56  *
  57  * 2010-03-01: Liang Zhen <zhen.liang@sun.com>
  58  * - "bucket" is a group of hlist_head now, user can specify bucket size
  59  *   by bkt_bits of cfs_hash_create(), all hlist_heads in a bucket share
  60  *   one lock for reducing memory overhead.
  61  *
  62  * - support lockless hash, caller will take care of locks:
  63  *   avoid lock overhead for hash tables that are already protected
  64  *   by locking in the caller for another reason
  65  *
  66  * - support both spin_lock/rwlock for bucket:
  67  *   overhead of spinlock contention is lower than read/write
  68  *   contention of rwlock, so using spinlock to serialize operations on
  69  *   bucket is more reasonable for those frequently changed hash tables
  70  *
  71  * - support one-single lock mode:
  72  *   one lock to protect all hash operations to avoid overhead of
  73  *   multiple locks if hash table is always small
  74  *
  75  * - removed a lot of unnecessary addref & decref on hash element:
  76  *   addref & decref are atomic operations in many use-cases which
  77  *   are expensive.
  78  *
  79  * - support non-blocking cfs_hash_add() and cfs_hash_findadd():
  80  *   some lustre use-cases require these functions to be strictly
  81  *   non-blocking, we need to schedule required rehash on a different
  82  *   thread on those cases.
  83  *
  84  * - safer rehash on large hash table
  85  *   In old implementation, rehash function will exclusively lock the
  86  *   hash table and finish rehash in one batch, it's dangerous on SMP
  87  *   system because rehash millions of elements could take long time.
  88  *   New implemented rehash can release lock and relax CPU in middle
  89  *   of rehash, it's safe for another thread to search/change on the
  90  *   hash table even it's in rehasing.
  91  *
  92  * - support two different refcount modes
  93  *   . hash table has refcount on element
  94  *   . hash table doesn't change refcount on adding/removing element
  95  *
  96  * - support long name hash table (for param-tree)
  97  *
  98  * - fix a bug for cfs_hash_rehash_key:
  99  *   in old implementation, cfs_hash_rehash_key could screw up the
 100  *   hash-table because @key is overwritten without any protection.
 101  *   Now we need user to define hs_keycpy for those rehash enabled
 102  *   hash tables, cfs_hash_rehash_key will overwrite hash-key
 103  *   inside lock by calling hs_keycpy.
 104  *
 105  * - better hash iteration:
 106  *   Now we support both locked iteration & lockless iteration of hash
 107  *   table. Also, user can break the iteration by return 1 in callback.
 108  */
 109 #include <linux/seq_file.h>
 110 #include <linux/log2.h>
 111
 112 #include "../../include/linux/libcfs/libcfs.h"
 113
 114 #if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
 115 static unsigned int warn_on_depth = 8;
 116 module_param(warn_on_depth, uint, 0644);
 117 MODULE_PARM_DESC(warn_on_depth, "warning when hash depth is high.");
 118 #endif
 119
 120 struct cfs_wi_sched *cfs_sched_rehash;
 121
 122 static inline void
 123 cfs_hash_nl_lock(union cfs_hash_lock *lock, int exclusive) {}
 124
 125 static inline void
 126 cfs_hash_nl_unlock(union cfs_hash_lock *lock, int exclusive) {}
 127
 128 static inline void
 129 cfs_hash_spin_lock(union cfs_hash_lock *lock, int exclusive)
 130         __acquires(&lock->spin)
 131 {
 132         spin_lock(&lock->spin);
 133 }
 134
 135 static inline void
 136 cfs_hash_spin_unlock(union cfs_hash_lock *lock, int exclusive)
 137         __releases(&lock->spin)
 138 {
 139         spin_unlock(&lock->spin);
 140 }
 141
 142 static inline void
 143 cfs_hash_rw_lock(union cfs_hash_lock *lock, int exclusive)
 144         __acquires(&lock->rw)
 145 {
 146         if (!exclusive)
 147                 read_lock(&lock->rw);
 148         else
 149                 write_lock(&lock->rw);
 150 }
 151
 152 static inline void
 153 cfs_hash_rw_unlock(union cfs_hash_lock *lock, int exclusive)
 154         __releases(&lock->rw)
 155 {
 156         if (!exclusive)
 157                 read_unlock(&lock->rw);
 158         else
 159                 write_unlock(&lock->rw);
 160 }
 161
 162 /** No lock hash */
 163 static struct cfs_hash_lock_ops cfs_hash_nl_lops = {
 164         .hs_lock        = cfs_hash_nl_lock,
 165         .hs_unlock      = cfs_hash_nl_unlock,
 166         .hs_bkt_lock    = cfs_hash_nl_lock,
 167         .hs_bkt_unlock  = cfs_hash_nl_unlock,
 168 };
 169
 170 /** no bucket lock, one spinlock to protect everything */
 171 static struct cfs_hash_lock_ops cfs_hash_nbl_lops = {
 172         .hs_lock        = cfs_hash_spin_lock,
 173         .hs_unlock      = cfs_hash_spin_unlock,
 174         .hs_bkt_lock    = cfs_hash_nl_lock,
 175         .hs_bkt_unlock  = cfs_hash_nl_unlock,
 176 };
 177
 178 /** spin bucket lock, rehash is enabled */
 179 static struct cfs_hash_lock_ops cfs_hash_bkt_spin_lops = {
 180         .hs_lock        = cfs_hash_rw_lock,
 181         .hs_unlock      = cfs_hash_rw_unlock,
 182         .hs_bkt_lock    = cfs_hash_spin_lock,
 183         .hs_bkt_unlock  = cfs_hash_spin_unlock,
 184 };
 185
 186 /** rw bucket lock, rehash is enabled */
 187 static struct cfs_hash_lock_ops cfs_hash_bkt_rw_lops = {
 188         .hs_lock        = cfs_hash_rw_lock,
 189         .hs_unlock      = cfs_hash_rw_unlock,
 190         .hs_bkt_lock    = cfs_hash_rw_lock,
 191         .hs_bkt_unlock  = cfs_hash_rw_unlock,
 192 };
 193
 194 /** spin bucket lock, rehash is disabled */
 195 static struct cfs_hash_lock_ops cfs_hash_nr_bkt_spin_lops = {
 196         .hs_lock        = cfs_hash_nl_lock,
 197         .hs_unlock      = cfs_hash_nl_unlock,
 198         .hs_bkt_lock    = cfs_hash_spin_lock,
 199         .hs_bkt_unlock  = cfs_hash_spin_unlock,
 200 };
 201
 202 /** rw bucket lock, rehash is disabled */
 203 static struct cfs_hash_lock_ops cfs_hash_nr_bkt_rw_lops = {
 204         .hs_lock        = cfs_hash_nl_lock,
 205         .hs_unlock      = cfs_hash_nl_unlock,
 206         .hs_bkt_lock    = cfs_hash_rw_lock,
 207         .hs_bkt_unlock  = cfs_hash_rw_unlock,
 208 };
 209
 210 static void
 211 cfs_hash_lock_setup(struct cfs_hash *hs)
 212 {
 213         if (cfs_hash_with_no_lock(hs)) {
 214                 hs->hs_lops = &cfs_hash_nl_lops;
 215
 216         } else if (cfs_hash_with_no_bktlock(hs)) {
 217                 hs->hs_lops = &cfs_hash_nbl_lops;
 218                 spin_lock_init(&hs->hs_lock.spin);
 219
 220         } else if (cfs_hash_with_rehash(hs)) {
 221                 rwlock_init(&hs->hs_lock.rw);
 222
 223                 if (cfs_hash_with_rw_bktlock(hs))
 224                         hs->hs_lops = &cfs_hash_bkt_rw_lops;
 225                 else if (cfs_hash_with_spin_bktlock(hs))
 226                         hs->hs_lops = &cfs_hash_bkt_spin_lops;
 227                 else
 228                         LBUG();
 229         } else {
 230                 if (cfs_hash_with_rw_bktlock(hs))
 231                         hs->hs_lops = &cfs_hash_nr_bkt_rw_lops;
 232                 else if (cfs_hash_with_spin_bktlock(hs))
 233                         hs->hs_lops = &cfs_hash_nr_bkt_spin_lops;
 234                 else
 235                         LBUG();
 236         }
 237 }
 238
 239 /**
 240  * Simple hash head without depth tracking
 241  * new element is always added to head of hlist
 242  */
 243 struct cfs_hash_head {
 244         struct hlist_head       hh_head;        /**< entries list */
 245 };
 246
 247 static int
 248 cfs_hash_hh_hhead_size(struct cfs_hash *hs)
 249 {
 250         return sizeof(struct cfs_hash_head);
 251 }
 252
 253 static struct hlist_head *
 254 cfs_hash_hh_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
 255 {
 256         struct cfs_hash_head *head;
 257
 258         head = (struct cfs_hash_head *)&bd->bd_bucket->hsb_head[0];
 259         return &head[bd->bd_offset].hh_head;
 260 }
 261
 262 static int
 263 cfs_hash_hh_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 264                       struct hlist_node *hnode)
 265 {
 266         hlist_add_head(hnode, cfs_hash_hh_hhead(hs, bd));
 267         return -1; /* unknown depth */
 268 }
 269
 270 static int
 271 cfs_hash_hh_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 272                       struct hlist_node *hnode)
 273 {
 274         hlist_del_init(hnode);
 275         return -1; /* unknown depth */
 276 }
 277
 278 /**
 279  * Simple hash head with depth tracking
 280  * new element is always added to head of hlist
 281  */
 282 struct cfs_hash_head_dep {
 283         struct hlist_head       hd_head;        /**< entries list */
 284         unsigned int            hd_depth;       /**< list length */
 285 };
 286
 287 static int
 288 cfs_hash_hd_hhead_size(struct cfs_hash *hs)
 289 {
 290         return sizeof(struct cfs_hash_head_dep);
 291 }
 292
 293 static struct hlist_head *
 294 cfs_hash_hd_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
 295 {
 296         struct cfs_hash_head_dep   *head;
 297
 298         head = (struct cfs_hash_head_dep *)&bd->bd_bucket->hsb_head[0];
 299         return &head[bd->bd_offset].hd_head;
 300 }
 301
 302 static int
 303 cfs_hash_hd_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 304                       struct hlist_node *hnode)
 305 {
 306         struct cfs_hash_head_dep *hh;
 307
 308         hh = container_of(cfs_hash_hd_hhead(hs, bd),
 309                           struct cfs_hash_head_dep, hd_head);
 310         hlist_add_head(hnode, &hh->hd_head);
 311         return ++hh->hd_depth;
 312 }
 313
 314 static int
 315 cfs_hash_hd_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 316                       struct hlist_node *hnode)
 317 {
 318         struct cfs_hash_head_dep *hh;
 319
 320         hh = container_of(cfs_hash_hd_hhead(hs, bd),
 321                           struct cfs_hash_head_dep, hd_head);
 322         hlist_del_init(hnode);
 323         return --hh->hd_depth;
 324 }
 325
 326 /**
 327  * double links hash head without depth tracking
 328  * new element is always added to tail of hlist
 329  */
 330 struct cfs_hash_dhead {
 331         struct hlist_head       dh_head;        /**< entries list */
 332         struct hlist_node       *dh_tail;       /**< the last entry */
 333 };
 334
 335 static int
 336 cfs_hash_dh_hhead_size(struct cfs_hash *hs)
 337 {
 338         return sizeof(struct cfs_hash_dhead);
 339 }
 340
 341 static struct hlist_head *
 342 cfs_hash_dh_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
 343 {
 344         struct cfs_hash_dhead *head;
 345
 346         head = (struct cfs_hash_dhead *)&bd->bd_bucket->hsb_head[0];
 347         return &head[bd->bd_offset].dh_head;
 348 }
 349
 350 static int
 351 cfs_hash_dh_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 352                       struct hlist_node *hnode)
 353 {
 354         struct cfs_hash_dhead *dh;
 355
 356         dh = container_of(cfs_hash_dh_hhead(hs, bd),
 357                           struct cfs_hash_dhead, dh_head);
 358         if (dh->dh_tail) /* not empty */
 359                 hlist_add_behind(hnode, dh->dh_tail);
 360         else /* empty list */
 361                 hlist_add_head(hnode, &dh->dh_head);
 362         dh->dh_tail = hnode;
 363         return -1; /* unknown depth */
 364 }
 365
 366 static int
 367 cfs_hash_dh_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 368                       struct hlist_node *hnd)
 369 {
 370         struct cfs_hash_dhead *dh;
 371
 372         dh = container_of(cfs_hash_dh_hhead(hs, bd),
 373                           struct cfs_hash_dhead, dh_head);
 374         if (!hnd->next) { /* it's the tail */
 375                 dh->dh_tail = (hnd->pprev == &dh->dh_head.first) ? NULL :
 376                               container_of(hnd->pprev, struct hlist_node, next);
 377         }
 378         hlist_del_init(hnd);
 379         return -1; /* unknown depth */
 380 }
 381
 382 /**
 383  * double links hash head with depth tracking
 384  * new element is always added to tail of hlist
 385  */
 386 struct cfs_hash_dhead_dep {
 387         struct hlist_head       dd_head;        /**< entries list */
 388         struct hlist_node       *dd_tail;       /**< the last entry */
 389         unsigned int            dd_depth;       /**< list length */
 390 };
 391
 392 static int
 393 cfs_hash_dd_hhead_size(struct cfs_hash *hs)
 394 {
 395         return sizeof(struct cfs_hash_dhead_dep);
 396 }
 397
 398 static struct hlist_head *
 399 cfs_hash_dd_hhead(struct cfs_hash *hs, struct cfs_hash_bd *bd)
 400 {
 401         struct cfs_hash_dhead_dep *head;
 402
 403         head = (struct cfs_hash_dhead_dep *)&bd->bd_bucket->hsb_head[0];
 404         return &head[bd->bd_offset].dd_head;
 405 }
 406
 407 static int
 408 cfs_hash_dd_hnode_add(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 409                       struct hlist_node *hnode)
 410 {
 411         struct cfs_hash_dhead_dep *dh;
 412
 413         dh = container_of(cfs_hash_dd_hhead(hs, bd),
 414                           struct cfs_hash_dhead_dep, dd_head);
 415         if (dh->dd_tail) /* not empty */
 416                 hlist_add_behind(hnode, dh->dd_tail);
 417         else /* empty list */
 418                 hlist_add_head(hnode, &dh->dd_head);
 419         dh->dd_tail = hnode;
 420         return ++dh->dd_depth;
 421 }
 422
 423 static int
 424 cfs_hash_dd_hnode_del(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 425                       struct hlist_node *hnd)
 426 {
 427         struct cfs_hash_dhead_dep *dh;
 428
 429         dh = container_of(cfs_hash_dd_hhead(hs, bd),
 430                           struct cfs_hash_dhead_dep, dd_head);
 431         if (!hnd->next) { /* it's the tail */
 432                 dh->dd_tail = (hnd->pprev == &dh->dd_head.first) ? NULL :
 433                               container_of(hnd->pprev, struct hlist_node, next);
 434         }
 435         hlist_del_init(hnd);
 436         return --dh->dd_depth;
 437 }
 438
 439 static struct cfs_hash_hlist_ops cfs_hash_hh_hops = {
 440         .hop_hhead      = cfs_hash_hh_hhead,
 441         .hop_hhead_size = cfs_hash_hh_hhead_size,
 442         .hop_hnode_add  = cfs_hash_hh_hnode_add,
 443         .hop_hnode_del  = cfs_hash_hh_hnode_del,
 444 };
 445
 446 static struct cfs_hash_hlist_ops cfs_hash_hd_hops = {
 447         .hop_hhead      = cfs_hash_hd_hhead,
 448         .hop_hhead_size = cfs_hash_hd_hhead_size,
 449         .hop_hnode_add  = cfs_hash_hd_hnode_add,
 450         .hop_hnode_del  = cfs_hash_hd_hnode_del,
 451 };
 452
 453 static struct cfs_hash_hlist_ops cfs_hash_dh_hops = {
 454         .hop_hhead      = cfs_hash_dh_hhead,
 455         .hop_hhead_size = cfs_hash_dh_hhead_size,
 456         .hop_hnode_add  = cfs_hash_dh_hnode_add,
 457         .hop_hnode_del  = cfs_hash_dh_hnode_del,
 458 };
 459
 460 static struct cfs_hash_hlist_ops cfs_hash_dd_hops = {
 461         .hop_hhead      = cfs_hash_dd_hhead,
 462         .hop_hhead_size = cfs_hash_dd_hhead_size,
 463         .hop_hnode_add  = cfs_hash_dd_hnode_add,
 464         .hop_hnode_del  = cfs_hash_dd_hnode_del,
 465 };
 466
 467 static void
 468 cfs_hash_hlist_setup(struct cfs_hash *hs)
 469 {
 470         if (cfs_hash_with_add_tail(hs)) {
 471                 hs->hs_hops = cfs_hash_with_depth(hs) ?
 472                               &cfs_hash_dd_hops : &cfs_hash_dh_hops;
 473         } else {
 474                 hs->hs_hops = cfs_hash_with_depth(hs) ?
 475                               &cfs_hash_hd_hops : &cfs_hash_hh_hops;
 476         }
 477 }
 478
 479 static void
 480 cfs_hash_bd_from_key(struct cfs_hash *hs, struct cfs_hash_bucket **bkts,
 481                      unsigned int bits, const void *key, struct cfs_hash_bd *bd)
 482 {
 483         unsigned int index = cfs_hash_id(hs, key, (1U << bits) - 1);
 484
 485         LASSERT(bits == hs->hs_cur_bits || bits == hs->hs_rehash_bits);
 486
 487         bd->bd_bucket = bkts[index & ((1U << (bits - hs->hs_bkt_bits)) - 1)];
 488         bd->bd_offset = index >> (bits - hs->hs_bkt_bits);
 489 }
 490
 491 void
 492 cfs_hash_bd_get(struct cfs_hash *hs, const void *key, struct cfs_hash_bd *bd)
 493 {
 494         /* NB: caller should hold hs->hs_rwlock if REHASH is set */
 495         if (likely(!hs->hs_rehash_buckets)) {
 496                 cfs_hash_bd_from_key(hs, hs->hs_buckets,
 497                                      hs->hs_cur_bits, key, bd);
 498         } else {
 499                 LASSERT(hs->hs_rehash_bits != 0);
 500                 cfs_hash_bd_from_key(hs, hs->hs_rehash_buckets,
 501                                      hs->hs_rehash_bits, key, bd);
 502         }
 503 }
 504 EXPORT_SYMBOL(cfs_hash_bd_get);
 505
 506 static inline void
 507 cfs_hash_bd_dep_record(struct cfs_hash *hs, struct cfs_hash_bd *bd, int dep_cur)
 508 {
 509         if (likely(dep_cur <= bd->bd_bucket->hsb_depmax))
 510                 return;
 511
 512         bd->bd_bucket->hsb_depmax = dep_cur;
 513 # if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
 514         if (likely(warn_on_depth == 0 ||
 515                    max(warn_on_depth, hs->hs_dep_max) >= dep_cur))
 516                 return;
 517
 518         spin_lock(&hs->hs_dep_lock);
 519         hs->hs_dep_max  = dep_cur;
 520         hs->hs_dep_bkt  = bd->bd_bucket->hsb_index;
 521         hs->hs_dep_off  = bd->bd_offset;
 522         hs->hs_dep_bits = hs->hs_cur_bits;
 523         spin_unlock(&hs->hs_dep_lock);
 524
 525         cfs_wi_schedule(cfs_sched_rehash, &hs->hs_dep_wi);
 526 # endif
 527 }
 528
 529 void
 530 cfs_hash_bd_add_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 531                        struct hlist_node *hnode)
 532 {
 533         int rc;
 534
 535         rc = hs->hs_hops->hop_hnode_add(hs, bd, hnode);
 536         cfs_hash_bd_dep_record(hs, bd, rc);
 537         bd->bd_bucket->hsb_version++;
 538         if (unlikely(bd->bd_bucket->hsb_version == 0))
 539                 bd->bd_bucket->hsb_version++;
 540         bd->bd_bucket->hsb_count++;
 541
 542         if (cfs_hash_with_counter(hs))
 543                 atomic_inc(&hs->hs_count);
 544         if (!cfs_hash_with_no_itemref(hs))
 545                 cfs_hash_get(hs, hnode);
 546 }
 547 EXPORT_SYMBOL(cfs_hash_bd_add_locked);
 548
 549 void
 550 cfs_hash_bd_del_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 551                        struct hlist_node *hnode)
 552 {
 553         hs->hs_hops->hop_hnode_del(hs, bd, hnode);
 554
 555         LASSERT(bd->bd_bucket->hsb_count > 0);
 556         bd->bd_bucket->hsb_count--;
 557         bd->bd_bucket->hsb_version++;
 558         if (unlikely(bd->bd_bucket->hsb_version == 0))
 559                 bd->bd_bucket->hsb_version++;
 560
 561         if (cfs_hash_with_counter(hs)) {
 562                 LASSERT(atomic_read(&hs->hs_count) > 0);
 563                 atomic_dec(&hs->hs_count);
 564         }
 565         if (!cfs_hash_with_no_itemref(hs))
 566                 cfs_hash_put_locked(hs, hnode);
 567 }
 568 EXPORT_SYMBOL(cfs_hash_bd_del_locked);
 569
 570 void
 571 cfs_hash_bd_move_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd_old,
 572                         struct cfs_hash_bd *bd_new, struct hlist_node *hnode)
 573 {
 574         struct cfs_hash_bucket *obkt = bd_old->bd_bucket;
 575         struct cfs_hash_bucket *nbkt = bd_new->bd_bucket;
 576         int rc;
 577
 578         if (cfs_hash_bd_compare(bd_old, bd_new) == 0)
 579                 return;
 580
 581         /* use cfs_hash_bd_hnode_add/del, to avoid atomic & refcount ops
 582          * in cfs_hash_bd_del/add_locked
 583          */
 584         hs->hs_hops->hop_hnode_del(hs, bd_old, hnode);
 585         rc = hs->hs_hops->hop_hnode_add(hs, bd_new, hnode);
 586         cfs_hash_bd_dep_record(hs, bd_new, rc);
 587
 588         LASSERT(obkt->hsb_count > 0);
 589         obkt->hsb_count--;
 590         obkt->hsb_version++;
 591         if (unlikely(obkt->hsb_version == 0))
 592                 obkt->hsb_version++;
 593         nbkt->hsb_count++;
 594         nbkt->hsb_version++;
 595         if (unlikely(nbkt->hsb_version == 0))
 596                 nbkt->hsb_version++;
 597 }
 598
 599 enum {
 600         /** always set, for sanity (avoid ZERO intent) */
 601         CFS_HS_LOOKUP_MASK_FIND = BIT(0),
 602         /** return entry with a ref */
 603         CFS_HS_LOOKUP_MASK_REF  = BIT(1),
 604         /** add entry if not existing */
 605         CFS_HS_LOOKUP_MASK_ADD  = BIT(2),
 606         /** delete entry, ignore other masks */
 607         CFS_HS_LOOKUP_MASK_DEL  = BIT(3),
 608 };
 609
 610 enum cfs_hash_lookup_intent {
 611         /** return item w/o refcount */
 612         CFS_HS_LOOKUP_IT_PEEK    = CFS_HS_LOOKUP_MASK_FIND,
 613         /** return item with refcount */
 614         CFS_HS_LOOKUP_IT_FIND    = (CFS_HS_LOOKUP_MASK_FIND |
 615                                     CFS_HS_LOOKUP_MASK_REF),
 616         /** return item w/o refcount if existed, otherwise add */
 617         CFS_HS_LOOKUP_IT_ADD     = (CFS_HS_LOOKUP_MASK_FIND |
 618                                     CFS_HS_LOOKUP_MASK_ADD),
 619         /** return item with refcount if existed, otherwise add */
 620         CFS_HS_LOOKUP_IT_FINDADD = (CFS_HS_LOOKUP_IT_FIND |
 621                                     CFS_HS_LOOKUP_MASK_ADD),
 622         /** delete if existed */
 623         CFS_HS_LOOKUP_IT_FINDDEL = (CFS_HS_LOOKUP_MASK_FIND |
 624                                     CFS_HS_LOOKUP_MASK_DEL)
 625 };
 626
 627 static struct hlist_node *
 628 cfs_hash_bd_lookup_intent(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 629                           const void *key, struct hlist_node *hnode,
 630                           enum cfs_hash_lookup_intent intent)
 631
 632 {
 633         struct hlist_head *hhead = cfs_hash_bd_hhead(hs, bd);
 634         struct hlist_node *ehnode;
 635         struct hlist_node *match;
 636         int intent_add = (intent & CFS_HS_LOOKUP_MASK_ADD) != 0;
 637
 638         /* with this function, we can avoid a lot of useless refcount ops,
 639          * which are expensive atomic operations most time.
 640          */
 641         match = intent_add ? NULL : hnode;
 642         hlist_for_each(ehnode, hhead) {
 643                 if (!cfs_hash_keycmp(hs, key, ehnode))
 644                         continue;
 645
 646                 if (match && match != ehnode) /* can't match */
 647                         continue;
 648
 649                 /* match and ... */
 650                 if ((intent & CFS_HS_LOOKUP_MASK_DEL) != 0) {
 651                         cfs_hash_bd_del_locked(hs, bd, ehnode);
 652                         return ehnode;
 653                 }
 654
 655                 /* caller wants refcount? */
 656                 if ((intent & CFS_HS_LOOKUP_MASK_REF) != 0)
 657                         cfs_hash_get(hs, ehnode);
 658                 return ehnode;
 659         }
 660         /* no match item */
 661         if (!intent_add)
 662                 return NULL;
 663
 664         LASSERT(hnode);
 665         cfs_hash_bd_add_locked(hs, bd, hnode);
 666         return hnode;
 667 }
 668
 669 struct hlist_node *
 670 cfs_hash_bd_lookup_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 671                           const void *key)
 672 {
 673         return cfs_hash_bd_lookup_intent(hs, bd, key, NULL,
 674                                          CFS_HS_LOOKUP_IT_FIND);
 675 }
 676 EXPORT_SYMBOL(cfs_hash_bd_lookup_locked);
 677
 678 struct hlist_node *
 679 cfs_hash_bd_peek_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
 680                         const void *key)
 681 {
 682         return cfs_hash_bd_lookup_intent(hs, bd, key, NULL,
 683                                          CFS_HS_LOOKUP_IT_PEEK);
 684 }
 685 EXPORT_SYMBOL(cfs_hash_bd_peek_locked);
 686
 687 static void
 688 cfs_hash_multi_bd_lock(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 689                        unsigned n, int excl)
 690 {
 691         struct cfs_hash_bucket *prev = NULL;
 692         int i;
 693
 694         /**
 695          * bds must be ascendantly ordered by bd->bd_bucket->hsb_index.
 696          * NB: it's possible that several bds point to the same bucket but
 697          * have different bd::bd_offset, so need take care of deadlock.
 698          */
 699         cfs_hash_for_each_bd(bds, n, i) {
 700                 if (prev == bds[i].bd_bucket)
 701                         continue;
 702
 703                 LASSERT(!prev || prev->hsb_index < bds[i].bd_bucket->hsb_index);
 704                 cfs_hash_bd_lock(hs, &bds[i], excl);
 705                 prev = bds[i].bd_bucket;
 706         }
 707 }
 708
 709 static void
 710 cfs_hash_multi_bd_unlock(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 711                          unsigned n, int excl)
 712 {
 713         struct cfs_hash_bucket *prev = NULL;
 714         int i;
 715
 716         cfs_hash_for_each_bd(bds, n, i) {
 717                 if (prev != bds[i].bd_bucket) {
 718                         cfs_hash_bd_unlock(hs, &bds[i], excl);
 719                         prev = bds[i].bd_bucket;
 720                 }
 721         }
 722 }
 723
 724 static struct hlist_node *
 725 cfs_hash_multi_bd_lookup_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 726                                 unsigned n, const void *key)
 727 {
 728         struct hlist_node *ehnode;
 729         unsigned i;
 730
 731         cfs_hash_for_each_bd(bds, n, i) {
 732                 ehnode = cfs_hash_bd_lookup_intent(hs, &bds[i], key, NULL,
 733                                                    CFS_HS_LOOKUP_IT_FIND);
 734                 if (ehnode)
 735                         return ehnode;
 736         }
 737         return NULL;
 738 }
 739
 740 static struct hlist_node *
 741 cfs_hash_multi_bd_findadd_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 742                                  unsigned n, const void *key,
 743                                  struct hlist_node *hnode, int noref)
 744 {
 745         struct hlist_node *ehnode;
 746         int intent;
 747         unsigned i;
 748
 749         LASSERT(hnode);
 750         intent = (!noref * CFS_HS_LOOKUP_MASK_REF) | CFS_HS_LOOKUP_IT_PEEK;
 751
 752         cfs_hash_for_each_bd(bds, n, i) {
 753                 ehnode = cfs_hash_bd_lookup_intent(hs, &bds[i], key,
 754                                                    NULL, intent);
 755                 if (ehnode)
 756                         return ehnode;
 757         }
 758
 759         if (i == 1) { /* only one bucket */
 760                 cfs_hash_bd_add_locked(hs, &bds[0], hnode);
 761         } else {
 762                 struct cfs_hash_bd mybd;
 763
 764                 cfs_hash_bd_get(hs, key, &mybd);
 765                 cfs_hash_bd_add_locked(hs, &mybd, hnode);
 766         }
 767
 768         return hnode;
 769 }
 770
 771 static struct hlist_node *
 772 cfs_hash_multi_bd_finddel_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 773                                  unsigned n, const void *key,
 774                                  struct hlist_node *hnode)
 775 {
 776         struct hlist_node *ehnode;
 777         unsigned int i;
 778
 779         cfs_hash_for_each_bd(bds, n, i) {
 780                 ehnode = cfs_hash_bd_lookup_intent(hs, &bds[i], key, hnode,
 781                                                    CFS_HS_LOOKUP_IT_FINDDEL);
 782                 if (ehnode)
 783                         return ehnode;
 784         }
 785         return NULL;
 786 }
 787
 788 static void
 789 cfs_hash_bd_order(struct cfs_hash_bd *bd1, struct cfs_hash_bd *bd2)
 790 {
 791         int rc;
 792
 793         if (!bd2->bd_bucket)
 794                 return;
 795
 796         if (!bd1->bd_bucket) {
 797                 *bd1 = *bd2;
 798                 bd2->bd_bucket = NULL;
 799                 return;
 800         }
 801
 802         rc = cfs_hash_bd_compare(bd1, bd2);
 803         if (!rc)
 804                 bd2->bd_bucket = NULL;
 805         else if (rc > 0)
 806                 swap(*bd1, *bd2); /* swap bd1 and bd2 */
 807 }
 808
 809 void
 810 cfs_hash_dual_bd_get(struct cfs_hash *hs, const void *key,
 811                      struct cfs_hash_bd *bds)
 812 {
 813         /* NB: caller should hold hs_lock.rw if REHASH is set */
 814         cfs_hash_bd_from_key(hs, hs->hs_buckets,
 815                              hs->hs_cur_bits, key, &bds[0]);
 816         if (likely(!hs->hs_rehash_buckets)) {
 817                 /* no rehash or not rehashing */
 818                 bds[1].bd_bucket = NULL;
 819                 return;
 820         }
 821
 822         LASSERT(hs->hs_rehash_bits != 0);
 823         cfs_hash_bd_from_key(hs, hs->hs_rehash_buckets,
 824                              hs->hs_rehash_bits, key, &bds[1]);
 825
 826         cfs_hash_bd_order(&bds[0], &bds[1]);
 827 }
 828
 829 void
 830 cfs_hash_dual_bd_lock(struct cfs_hash *hs, struct cfs_hash_bd *bds, int excl)
 831 {
 832         cfs_hash_multi_bd_lock(hs, bds, 2, excl);
 833 }
 834
 835 void
 836 cfs_hash_dual_bd_unlock(struct cfs_hash *hs, struct cfs_hash_bd *bds, int excl)
 837 {
 838         cfs_hash_multi_bd_unlock(hs, bds, 2, excl);
 839 }
 840
 841 struct hlist_node *
 842 cfs_hash_dual_bd_lookup_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 843                                const void *key)
 844 {
 845         return cfs_hash_multi_bd_lookup_locked(hs, bds, 2, key);
 846 }
 847
 848 struct hlist_node *
 849 cfs_hash_dual_bd_findadd_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 850                                 const void *key, struct hlist_node *hnode,
 851                                 int noref)
 852 {
 853         return cfs_hash_multi_bd_findadd_locked(hs, bds, 2, key,
 854                                                 hnode, noref);
 855 }
 856
 857 struct hlist_node *
 858 cfs_hash_dual_bd_finddel_locked(struct cfs_hash *hs, struct cfs_hash_bd *bds,
 859                                 const void *key, struct hlist_node *hnode)
 860 {
 861         return cfs_hash_multi_bd_finddel_locked(hs, bds, 2, key, hnode);
 862 }
 863
 864 static void
 865 cfs_hash_buckets_free(struct cfs_hash_bucket **buckets,
 866                       int bkt_size, int prev_size, int size)
 867 {
 868         int i;
 869
 870         for (i = prev_size; i < size; i++) {
 871                 if (buckets[i])
 872                         LIBCFS_FREE(buckets[i], bkt_size);
 873         }
 874
 875         LIBCFS_FREE(buckets, sizeof(buckets[0]) * size);
 876 }
 877
 878 /*
 879  * Create or grow bucket memory. Return old_buckets if no allocation was
 880  * needed, the newly allocated buckets if allocation was needed and
 881  * successful, and NULL on error.
 882  */
 883 static struct cfs_hash_bucket **
 884 cfs_hash_buckets_realloc(struct cfs_hash *hs, struct cfs_hash_bucket **old_bkts,
 885                          unsigned int old_size, unsigned int new_size)
 886 {
 887         struct cfs_hash_bucket **new_bkts;
 888         int i;
 889
 890         LASSERT(old_size == 0 || old_bkts);
 891
 892         if (old_bkts && old_size == new_size)
 893                 return old_bkts;
 894
 895         LIBCFS_ALLOC(new_bkts, sizeof(new_bkts[0]) * new_size);
 896         if (!new_bkts)
 897                 return NULL;
 898
 899         if (old_bkts) {
 900                 memcpy(new_bkts, old_bkts,
 901                        min(old_size, new_size) * sizeof(*old_bkts));
 902         }
 903
 904         for (i = old_size; i < new_size; i++) {
 905                 struct hlist_head *hhead;
 906                 struct cfs_hash_bd bd;
 907
 908                 LIBCFS_ALLOC(new_bkts[i], cfs_hash_bkt_size(hs));
 909                 if (!new_bkts[i]) {
 910                         cfs_hash_buckets_free(new_bkts, cfs_hash_bkt_size(hs),
 911                                               old_size, new_size);
 912                         return NULL;
 913                 }
 914
 915                 new_bkts[i]->hsb_index   = i;
 916                 new_bkts[i]->hsb_version = 1;  /* shouldn't be zero */
 917                 new_bkts[i]->hsb_depmax  = -1; /* unknown */
 918                 bd.bd_bucket = new_bkts[i];
 919                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead)
 920                         INIT_HLIST_HEAD(hhead);
 921
 922                 if (cfs_hash_with_no_lock(hs) ||
 923                     cfs_hash_with_no_bktlock(hs))
 924                         continue;
 925
 926                 if (cfs_hash_with_rw_bktlock(hs))
 927                         rwlock_init(&new_bkts[i]->hsb_lock.rw);
 928                 else if (cfs_hash_with_spin_bktlock(hs))
 929                         spin_lock_init(&new_bkts[i]->hsb_lock.spin);
 930                 else
 931                         LBUG(); /* invalid use-case */
 932         }
 933         return new_bkts;
 934 }
 935
 936 /**
 937  * Initialize new libcfs hash, where:
 938  * @name     - Descriptive hash name
 939  * @cur_bits - Initial hash table size, in bits
 940  * @max_bits - Maximum allowed hash table resize, in bits
 941  * @ops      - Registered hash table operations
 942  * @flags    - CFS_HASH_REHASH enable synamic hash resizing
 943  *           - CFS_HASH_SORT enable chained hash sort
 944  */
 945 static int cfs_hash_rehash_worker(struct cfs_workitem *wi);
 946
 947 #if CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1
 948 static int cfs_hash_dep_print(struct cfs_workitem *wi)
 949 {
 950         struct cfs_hash *hs = container_of(wi, struct cfs_hash, hs_dep_wi);
 951         int dep;
 952         int bkt;
 953         int off;
 954         int bits;
 955
 956         spin_lock(&hs->hs_dep_lock);
 957         dep  = hs->hs_dep_max;
 958         bkt  = hs->hs_dep_bkt;
 959         off  = hs->hs_dep_off;
 960         bits = hs->hs_dep_bits;
 961         spin_unlock(&hs->hs_dep_lock);
 962
 963         LCONSOLE_WARN("#### HASH %s (bits: %d): max depth %d at bucket %d/%d\n",
 964                       hs->hs_name, bits, dep, bkt, off);
 965         spin_lock(&hs->hs_dep_lock);
 966         hs->hs_dep_bits = 0; /* mark as workitem done */
 967         spin_unlock(&hs->hs_dep_lock);
 968         return 0;
 969 }
 970
 971 static void cfs_hash_depth_wi_init(struct cfs_hash *hs)
 972 {
 973         spin_lock_init(&hs->hs_dep_lock);
 974         cfs_wi_init(&hs->hs_dep_wi, hs, cfs_hash_dep_print);
 975 }
 976
 977 static void cfs_hash_depth_wi_cancel(struct cfs_hash *hs)
 978 {
 979         if (cfs_wi_deschedule(cfs_sched_rehash, &hs->hs_dep_wi))
 980                 return;
 981
 982         spin_lock(&hs->hs_dep_lock);
 983         while (hs->hs_dep_bits != 0) {
 984                 spin_unlock(&hs->hs_dep_lock);
 985                 cond_resched();
 986                 spin_lock(&hs->hs_dep_lock);
 987         }
 988         spin_unlock(&hs->hs_dep_lock);
 989 }
 990
 991 #else /* CFS_HASH_DEBUG_LEVEL < CFS_HASH_DEBUG_1 */
 992
 993 static inline void cfs_hash_depth_wi_init(struct cfs_hash *hs) {}
 994 static inline void cfs_hash_depth_wi_cancel(struct cfs_hash *hs) {}
 995
 996 #endif /* CFS_HASH_DEBUG_LEVEL >= CFS_HASH_DEBUG_1 */
 997
 998 struct cfs_hash *
 999 cfs_hash_create(char *name, unsigned cur_bits, unsigned max_bits,
1000                 unsigned bkt_bits, unsigned extra_bytes,
1001                 unsigned min_theta, unsigned max_theta,
1002                 struct cfs_hash_ops *ops, unsigned flags)
1003 {
1004         struct cfs_hash *hs;
1005         int len;
1006
1007         CLASSERT(CFS_HASH_THETA_BITS < 15);
1008
1009         LASSERT(name);
1010         LASSERT(ops->hs_key);
1011         LASSERT(ops->hs_hash);
1012         LASSERT(ops->hs_object);
1013         LASSERT(ops->hs_keycmp);
1014         LASSERT(ops->hs_get);
1015         LASSERT(ops->hs_put_locked);
1016
1017         if ((flags & CFS_HASH_REHASH) != 0)
1018                 flags |= CFS_HASH_COUNTER; /* must have counter */
1019
1020         LASSERT(cur_bits > 0);
1021         LASSERT(cur_bits >= bkt_bits);
1022         LASSERT(max_bits >= cur_bits && max_bits < 31);
1023         LASSERT(ergo((flags & CFS_HASH_REHASH) == 0, cur_bits == max_bits));
1024         LASSERT(ergo((flags & CFS_HASH_REHASH) != 0,
1025                      (flags & CFS_HASH_NO_LOCK) == 0));
1026         LASSERT(ergo((flags & CFS_HASH_REHASH_KEY) != 0, ops->hs_keycpy));
1027
1028         len = (flags & CFS_HASH_BIGNAME) == 0 ?
1029               CFS_HASH_NAME_LEN : CFS_HASH_BIGNAME_LEN;
1030         LIBCFS_ALLOC(hs, offsetof(struct cfs_hash, hs_name[len]));
1031         if (!hs)
1032                 return NULL;
1033
1034         strlcpy(hs->hs_name, name, len);
1035         hs->hs_flags = flags;
1036
1037         atomic_set(&hs->hs_refcount, 1);
1038         atomic_set(&hs->hs_count, 0);
1039
1040         cfs_hash_lock_setup(hs);
1041         cfs_hash_hlist_setup(hs);
1042
1043         hs->hs_cur_bits = (__u8)cur_bits;
1044         hs->hs_min_bits = (__u8)cur_bits;
1045         hs->hs_max_bits = (__u8)max_bits;
1046         hs->hs_bkt_bits = (__u8)bkt_bits;
1047
1048         hs->hs_ops         = ops;
1049         hs->hs_extra_bytes = extra_bytes;
1050         hs->hs_rehash_bits = 0;
1051         cfs_wi_init(&hs->hs_rehash_wi, hs, cfs_hash_rehash_worker);
1052         cfs_hash_depth_wi_init(hs);
1053
1054         if (cfs_hash_with_rehash(hs))
1055                 __cfs_hash_set_theta(hs, min_theta, max_theta);
1056
1057         hs->hs_buckets = cfs_hash_buckets_realloc(hs, NULL, 0,
1058                                                   CFS_HASH_NBKT(hs));
1059         if (hs->hs_buckets)
1060                 return hs;
1061
1062         LIBCFS_FREE(hs, offsetof(struct cfs_hash, hs_name[len]));
1063         return NULL;
1064 }
1065 EXPORT_SYMBOL(cfs_hash_create);
1066
1067 /**
1068  * Cleanup libcfs hash @hs.
1069  */
1070 static void
1071 cfs_hash_destroy(struct cfs_hash *hs)
1072 {
1073         struct hlist_node *hnode;
1074         struct hlist_node *pos;
1075         struct cfs_hash_bd bd;
1076         int i;
1077
1078         LASSERT(hs);
1079         LASSERT(!cfs_hash_is_exiting(hs) &&
1080                 !cfs_hash_is_iterating(hs));
1081
1082         /**
1083          * prohibit further rehashes, don't need any lock because
1084          * I'm the only (last) one can change it.
1085          */
1086         hs->hs_exiting = 1;
1087         if (cfs_hash_with_rehash(hs))
1088                 cfs_hash_rehash_cancel(hs);
1089
1090         cfs_hash_depth_wi_cancel(hs);
1091         /* rehash should be done/canceled */
1092         LASSERT(hs->hs_buckets && !hs->hs_rehash_buckets);
1093
1094         cfs_hash_for_each_bucket(hs, &bd, i) {
1095                 struct hlist_head *hhead;
1096
1097                 LASSERT(bd.bd_bucket);
1098                 /* no need to take this lock, just for consistent code */
1099                 cfs_hash_bd_lock(hs, &bd, 1);
1100
1101                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
1102                         hlist_for_each_safe(hnode, pos, hhead) {
1103                                 LASSERTF(!cfs_hash_with_assert_empty(hs),
1104                                          "hash %s bucket %u(%u) is not empty: %u items left\n",
1105                                          hs->hs_name, bd.bd_bucket->hsb_index,
1106                                          bd.bd_offset, bd.bd_bucket->hsb_count);
1107                                 /* can't assert key valicate, because we
1108                                  * can interrupt rehash
1109                                  */
1110                                 cfs_hash_bd_del_locked(hs, &bd, hnode);
1111                                 cfs_hash_exit(hs, hnode);
1112                         }
1113                 }
1114                 LASSERT(bd.bd_bucket->hsb_count == 0);
1115                 cfs_hash_bd_unlock(hs, &bd, 1);
1116                 cond_resched();
1117         }
1118
1119         LASSERT(atomic_read(&hs->hs_count) == 0);
1120
1121         cfs_hash_buckets_free(hs->hs_buckets, cfs_hash_bkt_size(hs),
1122                               0, CFS_HASH_NBKT(hs));
1123         i = cfs_hash_with_bigname(hs) ?
1124             CFS_HASH_BIGNAME_LEN : CFS_HASH_NAME_LEN;
1125         LIBCFS_FREE(hs, offsetof(struct cfs_hash, hs_name[i]));
1126 }
1127
1128 struct cfs_hash *cfs_hash_getref(struct cfs_hash *hs)
1129 {
1130         if (atomic_inc_not_zero(&hs->hs_refcount))
1131                 return hs;
1132         return NULL;
1133 }
1134 EXPORT_SYMBOL(cfs_hash_getref);
1135
1136 void cfs_hash_putref(struct cfs_hash *hs)
1137 {
1138         if (atomic_dec_and_test(&hs->hs_refcount))
1139                 cfs_hash_destroy(hs);
1140 }
1141 EXPORT_SYMBOL(cfs_hash_putref);
1142
1143 static inline int
1144 cfs_hash_rehash_bits(struct cfs_hash *hs)
1145 {
1146         if (cfs_hash_with_no_lock(hs) ||
1147             !cfs_hash_with_rehash(hs))
1148                 return -EOPNOTSUPP;
1149
1150         if (unlikely(cfs_hash_is_exiting(hs)))
1151                 return -ESRCH;
1152
1153         if (unlikely(cfs_hash_is_rehashing(hs)))
1154                 return -EALREADY;
1155
1156         if (unlikely(cfs_hash_is_iterating(hs)))
1157                 return -EAGAIN;
1158
1159         /* XXX: need to handle case with max_theta != 2.0
1160          *      and the case with min_theta != 0.5
1161          */
1162         if ((hs->hs_cur_bits < hs->hs_max_bits) &&
1163             (__cfs_hash_theta(hs) > hs->hs_max_theta))
1164                 return hs->hs_cur_bits + 1;
1165
1166         if (!cfs_hash_with_shrink(hs))
1167                 return 0;
1168
1169         if ((hs->hs_cur_bits > hs->hs_min_bits) &&
1170             (__cfs_hash_theta(hs) < hs->hs_min_theta))
1171                 return hs->hs_cur_bits - 1;
1172
1173         return 0;
1174 }
1175
1176 /**
1177  * don't allow inline rehash if:
1178  * - user wants non-blocking change (add/del) on hash table
1179  * - too many elements
1180  */
1181 static inline int
1182 cfs_hash_rehash_inline(struct cfs_hash *hs)
1183 {
1184         return !cfs_hash_with_nblk_change(hs) &&
1185                atomic_read(&hs->hs_count) < CFS_HASH_LOOP_HOG;
1186 }
1187
1188 /**
1189  * Add item @hnode to libcfs hash @hs using @key.  The registered
1190  * ops->hs_get function will be called when the item is added.
1191  */
1192 void
1193 cfs_hash_add(struct cfs_hash *hs, const void *key, struct hlist_node *hnode)
1194 {
1195         struct cfs_hash_bd bd;
1196         int bits;
1197
1198         LASSERT(hlist_unhashed(hnode));
1199
1200         cfs_hash_lock(hs, 0);
1201         cfs_hash_bd_get_and_lock(hs, key, &bd, 1);
1202
1203         cfs_hash_key_validate(hs, key, hnode);
1204         cfs_hash_bd_add_locked(hs, &bd, hnode);
1205
1206         cfs_hash_bd_unlock(hs, &bd, 1);
1207
1208         bits = cfs_hash_rehash_bits(hs);
1209         cfs_hash_unlock(hs, 0);
1210         if (bits > 0)
1211                 cfs_hash_rehash(hs, cfs_hash_rehash_inline(hs));
1212 }
1213 EXPORT_SYMBOL(cfs_hash_add);
1214
1215 static struct hlist_node *
1216 cfs_hash_find_or_add(struct cfs_hash *hs, const void *key,
1217                      struct hlist_node *hnode, int noref)
1218 {
1219         struct hlist_node *ehnode;
1220         struct cfs_hash_bd bds[2];
1221         int bits = 0;
1222
1223         LASSERT(hlist_unhashed(hnode));
1224
1225         cfs_hash_lock(hs, 0);
1226         cfs_hash_dual_bd_get_and_lock(hs, key, bds, 1);
1227
1228         cfs_hash_key_validate(hs, key, hnode);
1229         ehnode = cfs_hash_dual_bd_findadd_locked(hs, bds, key,
1230                                                  hnode, noref);
1231         cfs_hash_dual_bd_unlock(hs, bds, 1);
1232
1233         if (ehnode == hnode)    /* new item added */
1234                 bits = cfs_hash_rehash_bits(hs);
1235         cfs_hash_unlock(hs, 0);
1236         if (bits > 0)
1237                 cfs_hash_rehash(hs, cfs_hash_rehash_inline(hs));
1238
1239         return ehnode;
1240 }
1241
1242 /**
1243  * Add item @hnode to libcfs hash @hs using @key.  The registered
1244  * ops->hs_get function will be called if the item was added.
1245  * Returns 0 on success or -EALREADY on key collisions.
1246  */
1247 int
1248 cfs_hash_add_unique(struct cfs_hash *hs, const void *key,
1249                     struct hlist_node *hnode)
1250 {
1251         return cfs_hash_find_or_add(hs, key, hnode, 1) != hnode ?
1252                -EALREADY : 0;
1253 }
1254 EXPORT_SYMBOL(cfs_hash_add_unique);
1255
1256 /**
1257  * Add item @hnode to libcfs hash @hs using @key.  If this @key
1258  * already exists in the hash then ops->hs_get will be called on the
1259  * conflicting entry and that entry will be returned to the caller.
1260  * Otherwise ops->hs_get is called on the item which was added.
1261  */
1262 void *
1263 cfs_hash_findadd_unique(struct cfs_hash *hs, const void *key,
1264                         struct hlist_node *hnode)
1265 {
1266         hnode = cfs_hash_find_or_add(hs, key, hnode, 0);
1267
1268         return cfs_hash_object(hs, hnode);
1269 }
1270 EXPORT_SYMBOL(cfs_hash_findadd_unique);
1271
1272 /**
1273  * Delete item @hnode from the libcfs hash @hs using @key.  The @key
1274  * is required to ensure the correct hash bucket is locked since there
1275  * is no direct linkage from the item to the bucket.  The object
1276  * removed from the hash will be returned and obs->hs_put is called
1277  * on the removed object.
1278  */
1279 void *
1280 cfs_hash_del(struct cfs_hash *hs, const void *key, struct hlist_node *hnode)
1281 {
1282         void *obj = NULL;
1283         int bits = 0;
1284         struct cfs_hash_bd bds[2];
1285
1286         cfs_hash_lock(hs, 0);
1287         cfs_hash_dual_bd_get_and_lock(hs, key, bds, 1);
1288
1289         /* NB: do nothing if @hnode is not in hash table */
1290         if (!hnode || !hlist_unhashed(hnode)) {
1291                 if (!bds[1].bd_bucket && hnode) {
1292                         cfs_hash_bd_del_locked(hs, &bds[0], hnode);
1293                 } else {
1294                         hnode = cfs_hash_dual_bd_finddel_locked(hs, bds,
1295                                                                 key, hnode);
1296                 }
1297         }
1298
1299         if (hnode) {
1300                 obj  = cfs_hash_object(hs, hnode);
1301                 bits = cfs_hash_rehash_bits(hs);
1302         }
1303
1304         cfs_hash_dual_bd_unlock(hs, bds, 1);
1305         cfs_hash_unlock(hs, 0);
1306         if (bits > 0)
1307                 cfs_hash_rehash(hs, cfs_hash_rehash_inline(hs));
1308
1309         return obj;
1310 }
1311 EXPORT_SYMBOL(cfs_hash_del);
1312
1313 /**
1314  * Delete item given @key in libcfs hash @hs.  The first @key found in
1315  * the hash will be removed, if the key exists multiple times in the hash
1316  * @hs this function must be called once per key.  The removed object
1317  * will be returned and ops->hs_put is called on the removed object.
1318  */
1319 void *
1320 cfs_hash_del_key(struct cfs_hash *hs, const void *key)
1321 {
1322         return cfs_hash_del(hs, key, NULL);
1323 }
1324 EXPORT_SYMBOL(cfs_hash_del_key);
1325
1326 /**
1327  * Lookup an item using @key in the libcfs hash @hs and return it.
1328  * If the @key is found in the hash hs->hs_get() is called and the
1329  * matching objects is returned.  It is the callers responsibility
1330  * to call the counterpart ops->hs_put using the cfs_hash_put() macro
1331  * when when finished with the object.  If the @key was not found
1332  * in the hash @hs NULL is returned.
1333  */
1334 void *
1335 cfs_hash_lookup(struct cfs_hash *hs, const void *key)
1336 {
1337         void *obj = NULL;
1338         struct hlist_node *hnode;
1339         struct cfs_hash_bd bds[2];
1340
1341         cfs_hash_lock(hs, 0);
1342         cfs_hash_dual_bd_get_and_lock(hs, key, bds, 0);
1343
1344         hnode = cfs_hash_dual_bd_lookup_locked(hs, bds, key);
1345         if (hnode)
1346                 obj = cfs_hash_object(hs, hnode);
1347
1348         cfs_hash_dual_bd_unlock(hs, bds, 0);
1349         cfs_hash_unlock(hs, 0);
1350
1351         return obj;
1352 }
1353 EXPORT_SYMBOL(cfs_hash_lookup);
1354
1355 static void
1356 cfs_hash_for_each_enter(struct cfs_hash *hs)
1357 {
1358         LASSERT(!cfs_hash_is_exiting(hs));
1359
1360         if (!cfs_hash_with_rehash(hs))
1361                 return;
1362         /*
1363          * NB: it's race on cfs_has_t::hs_iterating, but doesn't matter
1364          * because it's just an unreliable signal to rehash-thread,
1365          * rehash-thread will try to finish rehash ASAP when seeing this.
1366          */
1367         hs->hs_iterating = 1;
1368
1369         cfs_hash_lock(hs, 1);
1370         hs->hs_iterators++;
1371
1372         /* NB: iteration is mostly called by service thread,
1373          * we tend to cancel pending rehash-request, instead of
1374          * blocking service thread, we will relaunch rehash request
1375          * after iteration
1376          */
1377         if (cfs_hash_is_rehashing(hs))
1378                 cfs_hash_rehash_cancel_locked(hs);
1379         cfs_hash_unlock(hs, 1);
1380 }
1381
1382 static void
1383 cfs_hash_for_each_exit(struct cfs_hash *hs)
1384 {
1385         int remained;
1386         int bits;
1387
1388         if (!cfs_hash_with_rehash(hs))
1389                 return;
1390         cfs_hash_lock(hs, 1);
1391         remained = --hs->hs_iterators;
1392         bits = cfs_hash_rehash_bits(hs);
1393         cfs_hash_unlock(hs, 1);
1394         /* NB: it's race on cfs_has_t::hs_iterating, see above */
1395         if (remained == 0)
1396                 hs->hs_iterating = 0;
1397         if (bits > 0) {
1398                 cfs_hash_rehash(hs, atomic_read(&hs->hs_count) <
1399                                     CFS_HASH_LOOP_HOG);
1400         }
1401 }
1402
1403 /**
1404  * For each item in the libcfs hash @hs call the passed callback @func
1405  * and pass to it as an argument each hash item and the private @data.
1406  *
1407  * a) the function may sleep!
1408  * b) during the callback:
1409  *    . the bucket lock is held so the callback must never sleep.
1410  *    . if @removal_safe is true, use can remove current item by
1411  *      cfs_hash_bd_del_locked
1412  */
1413 static __u64
1414 cfs_hash_for_each_tight(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1415                         void *data, int remove_safe)
1416 {
1417         struct hlist_node *hnode;
1418         struct hlist_node *pos;
1419         struct cfs_hash_bd bd;
1420         __u64 count = 0;
1421         int excl = !!remove_safe;
1422         int loop = 0;
1423         int i;
1424
1425         cfs_hash_for_each_enter(hs);
1426
1427         cfs_hash_lock(hs, 0);
1428         LASSERT(!cfs_hash_is_rehashing(hs));
1429
1430         cfs_hash_for_each_bucket(hs, &bd, i) {
1431                 struct hlist_head *hhead;
1432
1433                 cfs_hash_bd_lock(hs, &bd, excl);
1434                 if (!func) { /* only glimpse size */
1435                         count += bd.bd_bucket->hsb_count;
1436                         cfs_hash_bd_unlock(hs, &bd, excl);
1437                         continue;
1438                 }
1439
1440                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
1441                         hlist_for_each_safe(hnode, pos, hhead) {
1442                                 cfs_hash_bucket_validate(hs, &bd, hnode);
1443                                 count++;
1444                                 loop++;
1445                                 if (func(hs, &bd, hnode, data)) {
1446                                         cfs_hash_bd_unlock(hs, &bd, excl);
1447                                         goto out;
1448                                 }
1449                         }
1450                 }
1451                 cfs_hash_bd_unlock(hs, &bd, excl);
1452                 if (loop < CFS_HASH_LOOP_HOG)
1453                         continue;
1454                 loop = 0;
1455                 cfs_hash_unlock(hs, 0);
1456                 cond_resched();
1457                 cfs_hash_lock(hs, 0);
1458         }
1459  out:
1460         cfs_hash_unlock(hs, 0);
1461
1462         cfs_hash_for_each_exit(hs);
1463         return count;
1464 }
1465
1466 struct cfs_hash_cond_arg {
1467         cfs_hash_cond_opt_cb_t  func;
1468         void                    *arg;
1469 };
1470
1471 static int
1472 cfs_hash_cond_del_locked(struct cfs_hash *hs, struct cfs_hash_bd *bd,
1473                          struct hlist_node *hnode, void *data)
1474 {
1475         struct cfs_hash_cond_arg *cond = data;
1476
1477         if (cond->func(cfs_hash_object(hs, hnode), cond->arg))
1478                 cfs_hash_bd_del_locked(hs, bd, hnode);
1479         return 0;
1480 }
1481
1482 /**
1483  * Delete item from the libcfs hash @hs when @func return true.
1484  * The write lock being hold during loop for each bucket to avoid
1485  * any object be reference.
1486  */
1487 void
1488 cfs_hash_cond_del(struct cfs_hash *hs, cfs_hash_cond_opt_cb_t func, void *data)
1489 {
1490         struct cfs_hash_cond_arg arg = {
1491                 .func   = func,
1492                 .arg    = data,
1493         };
1494
1495         cfs_hash_for_each_tight(hs, cfs_hash_cond_del_locked, &arg, 1);
1496 }
1497 EXPORT_SYMBOL(cfs_hash_cond_del);
1498
1499 void
1500 cfs_hash_for_each(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1501                   void *data)
1502 {
1503         cfs_hash_for_each_tight(hs, func, data, 0);
1504 }
1505 EXPORT_SYMBOL(cfs_hash_for_each);
1506
1507 void
1508 cfs_hash_for_each_safe(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1509                        void *data)
1510 {
1511         cfs_hash_for_each_tight(hs, func, data, 1);
1512 }
1513 EXPORT_SYMBOL(cfs_hash_for_each_safe);
1514
1515 static int
1516 cfs_hash_peek(struct cfs_hash *hs, struct cfs_hash_bd *bd,
1517               struct hlist_node *hnode, void *data)
1518 {
1519         *(int *)data = 0;
1520         return 1; /* return 1 to break the loop */
1521 }
1522
1523 int
1524 cfs_hash_is_empty(struct cfs_hash *hs)
1525 {
1526         int empty = 1;
1527
1528         cfs_hash_for_each_tight(hs, cfs_hash_peek, &empty, 0);
1529         return empty;
1530 }
1531 EXPORT_SYMBOL(cfs_hash_is_empty);
1532
1533 __u64
1534 cfs_hash_size_get(struct cfs_hash *hs)
1535 {
1536         return cfs_hash_with_counter(hs) ?
1537                atomic_read(&hs->hs_count) :
1538                cfs_hash_for_each_tight(hs, NULL, NULL, 0);
1539 }
1540 EXPORT_SYMBOL(cfs_hash_size_get);
1541
1542 /*
1543  * cfs_hash_for_each_relax:
1544  * Iterate the hash table and call @func on each item without
1545  * any lock. This function can't guarantee to finish iteration
1546  * if these features are enabled:
1547  *
1548  *  a. if rehash_key is enabled, an item can be moved from
1549  *     one bucket to another bucket
1550  *  b. user can remove non-zero-ref item from hash-table,
1551  *     so the item can be removed from hash-table, even worse,
1552  *     it's possible that user changed key and insert to another
1553  *     hash bucket.
1554  * there's no way for us to finish iteration correctly on previous
1555  * two cases, so iteration has to be stopped on change.
1556  */
1557 static int
1558 cfs_hash_for_each_relax(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1559                         void *data)
1560 {
1561         struct hlist_node *hnode;
1562         struct hlist_node *tmp;
1563         struct cfs_hash_bd bd;
1564         __u32 version;
1565         int count = 0;
1566         int stop_on_change;
1567         int rc;
1568         int i;
1569
1570         stop_on_change = cfs_hash_with_rehash_key(hs) ||
1571                          !cfs_hash_with_no_itemref(hs) ||
1572                          !hs->hs_ops->hs_put_locked;
1573         cfs_hash_lock(hs, 0);
1574         LASSERT(!cfs_hash_is_rehashing(hs));
1575
1576         cfs_hash_for_each_bucket(hs, &bd, i) {
1577                 struct hlist_head *hhead;
1578
1579                 cfs_hash_bd_lock(hs, &bd, 0);
1580                 version = cfs_hash_bd_version_get(&bd);
1581
1582                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
1583                         for (hnode = hhead->first; hnode;) {
1584                                 cfs_hash_bucket_validate(hs, &bd, hnode);
1585                                 cfs_hash_get(hs, hnode);
1586                                 cfs_hash_bd_unlock(hs, &bd, 0);
1587                                 cfs_hash_unlock(hs, 0);
1588
1589                                 rc = func(hs, &bd, hnode, data);
1590                                 if (stop_on_change)
1591                                         cfs_hash_put(hs, hnode);
1592                                 cond_resched();
1593                                 count++;
1594
1595                                 cfs_hash_lock(hs, 0);
1596                                 cfs_hash_bd_lock(hs, &bd, 0);
1597                                 if (!stop_on_change) {
1598                                         tmp = hnode->next;
1599                                         cfs_hash_put_locked(hs, hnode);
1600                                         hnode = tmp;
1601                                 } else { /* bucket changed? */
1602                                         if (version !=
1603                                             cfs_hash_bd_version_get(&bd))
1604                                                 break;
1605                                         /* safe to continue because no change */
1606                                         hnode = hnode->next;
1607                                 }
1608                                 if (rc) /* callback wants to break iteration */
1609                                         break;
1610                         }
1611                         if (rc) /* callback wants to break iteration */
1612                                 break;
1613                 }
1614                 cfs_hash_bd_unlock(hs, &bd, 0);
1615                 if (rc) /* callback wants to break iteration */
1616                         break;
1617         }
1618         cfs_hash_unlock(hs, 0);
1619
1620         return count;
1621 }
1622
1623 int
1624 cfs_hash_for_each_nolock(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1625                          void *data)
1626 {
1627         if (cfs_hash_with_no_lock(hs) ||
1628             cfs_hash_with_rehash_key(hs) ||
1629             !cfs_hash_with_no_itemref(hs))
1630                 return -EOPNOTSUPP;
1631
1632         if (!hs->hs_ops->hs_get ||
1633             (!hs->hs_ops->hs_put && !hs->hs_ops->hs_put_locked))
1634                 return -EOPNOTSUPP;
1635
1636         cfs_hash_for_each_enter(hs);
1637         cfs_hash_for_each_relax(hs, func, data);
1638         cfs_hash_for_each_exit(hs);
1639
1640         return 0;
1641 }
1642 EXPORT_SYMBOL(cfs_hash_for_each_nolock);
1643
1644 /**
1645  * For each hash bucket in the libcfs hash @hs call the passed callback
1646  * @func until all the hash buckets are empty.  The passed callback @func
1647  * or the previously registered callback hs->hs_put must remove the item
1648  * from the hash.  You may either use the cfs_hash_del() or hlist_del()
1649  * functions.  No rwlocks will be held during the callback @func it is
1650  * safe to sleep if needed.  This function will not terminate until the
1651  * hash is empty.  Note it is still possible to concurrently add new
1652  * items in to the hash.  It is the callers responsibility to ensure
1653  * the required locking is in place to prevent concurrent insertions.
1654  */
1655 int
1656 cfs_hash_for_each_empty(struct cfs_hash *hs, cfs_hash_for_each_cb_t func,
1657                         void *data)
1658 {
1659         unsigned i = 0;
1660
1661         if (cfs_hash_with_no_lock(hs))
1662                 return -EOPNOTSUPP;
1663
1664         if (!hs->hs_ops->hs_get ||
1665             (!hs->hs_ops->hs_put && !hs->hs_ops->hs_put_locked))
1666                 return -EOPNOTSUPP;
1667
1668         cfs_hash_for_each_enter(hs);
1669         while (cfs_hash_for_each_relax(hs, func, data)) {
1670                 CDEBUG(D_INFO, "Try to empty hash: %s, loop: %u\n",
1671                        hs->hs_name, i++);
1672         }
1673         cfs_hash_for_each_exit(hs);
1674         return 0;
1675 }
1676 EXPORT_SYMBOL(cfs_hash_for_each_empty);
1677
1678 void
1679 cfs_hash_hlist_for_each(struct cfs_hash *hs, unsigned hindex,
1680                         cfs_hash_for_each_cb_t func, void *data)
1681 {
1682         struct hlist_head *hhead;
1683         struct hlist_node *hnode;
1684         struct cfs_hash_bd bd;
1685
1686         cfs_hash_for_each_enter(hs);
1687         cfs_hash_lock(hs, 0);
1688         if (hindex >= CFS_HASH_NHLIST(hs))
1689                 goto out;
1690
1691         cfs_hash_bd_index_set(hs, hindex, &bd);
1692
1693         cfs_hash_bd_lock(hs, &bd, 0);
1694         hhead = cfs_hash_bd_hhead(hs, &bd);
1695         hlist_for_each(hnode, hhead) {
1696                 if (func(hs, &bd, hnode, data))
1697                         break;
1698         }
1699         cfs_hash_bd_unlock(hs, &bd, 0);
1700 out:
1701         cfs_hash_unlock(hs, 0);
1702         cfs_hash_for_each_exit(hs);
1703 }
1704 EXPORT_SYMBOL(cfs_hash_hlist_for_each);
1705
1706 /*
1707  * For each item in the libcfs hash @hs which matches the @key call
1708  * the passed callback @func and pass to it as an argument each hash
1709  * item and the private @data. During the callback the bucket lock
1710  * is held so the callback must never sleep.
1711    */
1712 void
1713 cfs_hash_for_each_key(struct cfs_hash *hs, const void *key,
1714                       cfs_hash_for_each_cb_t func, void *data)
1715 {
1716         struct hlist_node *hnode;
1717         struct cfs_hash_bd bds[2];
1718         unsigned int i;
1719
1720         cfs_hash_lock(hs, 0);
1721
1722         cfs_hash_dual_bd_get_and_lock(hs, key, bds, 0);
1723
1724         cfs_hash_for_each_bd(bds, 2, i) {
1725                 struct hlist_head *hlist = cfs_hash_bd_hhead(hs, &bds[i]);
1726
1727                 hlist_for_each(hnode, hlist) {
1728                         cfs_hash_bucket_validate(hs, &bds[i], hnode);
1729
1730                         if (cfs_hash_keycmp(hs, key, hnode)) {
1731                                 if (func(hs, &bds[i], hnode, data))
1732                                         break;
1733                         }
1734                 }
1735         }
1736
1737         cfs_hash_dual_bd_unlock(hs, bds, 0);
1738         cfs_hash_unlock(hs, 0);
1739 }
1740 EXPORT_SYMBOL(cfs_hash_for_each_key);
1741
1742 /**
1743  * Rehash the libcfs hash @hs to the given @bits.  This can be used
1744  * to grow the hash size when excessive chaining is detected, or to
1745  * shrink the hash when it is larger than needed.  When the CFS_HASH_REHASH
1746  * flag is set in @hs the libcfs hash may be dynamically rehashed
1747  * during addition or removal if the hash's theta value exceeds
1748  * either the hs->hs_min_theta or hs->max_theta values.  By default
1749  * these values are tuned to keep the chained hash depth small, and
1750  * this approach assumes a reasonably uniform hashing function.  The
1751  * theta thresholds for @hs are tunable via cfs_hash_set_theta().
1752  */
1753 void
1754 cfs_hash_rehash_cancel_locked(struct cfs_hash *hs)
1755 {
1756         int i;
1757
1758         /* need hold cfs_hash_lock(hs, 1) */
1759         LASSERT(cfs_hash_with_rehash(hs) &&
1760                 !cfs_hash_with_no_lock(hs));
1761
1762         if (!cfs_hash_is_rehashing(hs))
1763                 return;
1764
1765         if (cfs_wi_deschedule(cfs_sched_rehash, &hs->hs_rehash_wi)) {
1766                 hs->hs_rehash_bits = 0;
1767                 return;
1768         }
1769
1770         for (i = 2; cfs_hash_is_rehashing(hs); i++) {
1771                 cfs_hash_unlock(hs, 1);
1772                 /* raise console warning while waiting too long */
1773                 CDEBUG(is_power_of_2(i >> 3) ? D_WARNING : D_INFO,
1774                        "hash %s is still rehashing, rescheded %d\n",
1775                        hs->hs_name, i - 1);
1776                 cond_resched();
1777                 cfs_hash_lock(hs, 1);
1778         }
1779 }
1780
1781 void
1782 cfs_hash_rehash_cancel(struct cfs_hash *hs)
1783 {
1784         cfs_hash_lock(hs, 1);
1785         cfs_hash_rehash_cancel_locked(hs);
1786         cfs_hash_unlock(hs, 1);
1787 }
1788
1789 int
1790 cfs_hash_rehash(struct cfs_hash *hs, int do_rehash)
1791 {
1792         int rc;
1793
1794         LASSERT(cfs_hash_with_rehash(hs) && !cfs_hash_with_no_lock(hs));
1795
1796         cfs_hash_lock(hs, 1);
1797
1798         rc = cfs_hash_rehash_bits(hs);
1799         if (rc <= 0) {
1800                 cfs_hash_unlock(hs, 1);
1801                 return rc;
1802         }
1803
1804         hs->hs_rehash_bits = rc;
1805         if (!do_rehash) {
1806                 /* launch and return */
1807                 cfs_wi_schedule(cfs_sched_rehash, &hs->hs_rehash_wi);
1808                 cfs_hash_unlock(hs, 1);
1809                 return 0;
1810         }
1811
1812         /* rehash right now */
1813         cfs_hash_unlock(hs, 1);
1814
1815         return cfs_hash_rehash_worker(&hs->hs_rehash_wi);
1816 }
1817
1818 static int
1819 cfs_hash_rehash_bd(struct cfs_hash *hs, struct cfs_hash_bd *old)
1820 {
1821         struct cfs_hash_bd new;
1822         struct hlist_head *hhead;
1823         struct hlist_node *hnode;
1824         struct hlist_node *pos;
1825         void *key;
1826         int c = 0;
1827
1828         /* hold cfs_hash_lock(hs, 1), so don't need any bucket lock */
1829         cfs_hash_bd_for_each_hlist(hs, old, hhead) {
1830                 hlist_for_each_safe(hnode, pos, hhead) {
1831                         key = cfs_hash_key(hs, hnode);
1832                         LASSERT(key);
1833                         /* Validate hnode is in the correct bucket. */
1834                         cfs_hash_bucket_validate(hs, old, hnode);
1835                         /*
1836                          * Delete from old hash bucket; move to new bucket.
1837                          * ops->hs_key must be defined.
1838                          */
1839                         cfs_hash_bd_from_key(hs, hs->hs_rehash_buckets,
1840                                              hs->hs_rehash_bits, key, &new);
1841                         cfs_hash_bd_move_locked(hs, old, &new, hnode);
1842                         c++;
1843                 }
1844         }
1845
1846         return c;
1847 }
1848
1849 static int
1850 cfs_hash_rehash_worker(struct cfs_workitem *wi)
1851 {
1852         struct cfs_hash *hs = container_of(wi, struct cfs_hash, hs_rehash_wi);
1853         struct cfs_hash_bucket **bkts;
1854         struct cfs_hash_bd bd;
1855         unsigned int old_size;
1856         unsigned int new_size;
1857         int bsize;
1858         int count = 0;
1859         int rc = 0;
1860         int i;
1861
1862         LASSERT(hs && cfs_hash_with_rehash(hs));
1863
1864         cfs_hash_lock(hs, 0);
1865         LASSERT(cfs_hash_is_rehashing(hs));
1866
1867         old_size = CFS_HASH_NBKT(hs);
1868         new_size = CFS_HASH_RH_NBKT(hs);
1869
1870         cfs_hash_unlock(hs, 0);
1871
1872         /*
1873          * don't need hs::hs_rwlock for hs::hs_buckets,
1874          * because nobody can change bkt-table except me.
1875          */
1876         bkts = cfs_hash_buckets_realloc(hs, hs->hs_buckets,
1877                                         old_size, new_size);
1878         cfs_hash_lock(hs, 1);
1879         if (!bkts) {
1880                 rc = -ENOMEM;
1881                 goto out;
1882         }
1883
1884         if (bkts == hs->hs_buckets) {
1885                 bkts = NULL; /* do nothing */
1886                 goto out;
1887         }
1888
1889         rc = __cfs_hash_theta(hs);
1890         if ((rc >= hs->hs_min_theta) && (rc <= hs->hs_max_theta)) {
1891                 /* free the new allocated bkt-table */
1892                 old_size = new_size;
1893                 new_size = CFS_HASH_NBKT(hs);
1894                 rc = -EALREADY;
1895                 goto out;
1896         }
1897
1898         LASSERT(!hs->hs_rehash_buckets);
1899         hs->hs_rehash_buckets = bkts;
1900
1901         rc = 0;
1902         cfs_hash_for_each_bucket(hs, &bd, i) {
1903                 if (cfs_hash_is_exiting(hs)) {
1904                         rc = -ESRCH;
1905                         /* someone wants to destroy the hash, abort now */
1906                         if (old_size < new_size) /* OK to free old bkt-table */
1907                                 break;
1908                         /* it's shrinking, need free new bkt-table */
1909                         hs->hs_rehash_buckets = NULL;
1910                         old_size = new_size;
1911                         new_size = CFS_HASH_NBKT(hs);
1912                         goto out;
1913                 }
1914
1915                 count += cfs_hash_rehash_bd(hs, &bd);
1916                 if (count < CFS_HASH_LOOP_HOG ||
1917                     cfs_hash_is_iterating(hs)) { /* need to finish ASAP */
1918                         continue;
1919                 }
1920
1921                 count = 0;
1922                 cfs_hash_unlock(hs, 1);
1923                 cond_resched();
1924                 cfs_hash_lock(hs, 1);
1925         }
1926
1927         hs->hs_rehash_count++;
1928
1929         bkts = hs->hs_buckets;
1930         hs->hs_buckets = hs->hs_rehash_buckets;
1931         hs->hs_rehash_buckets = NULL;
1932
1933         hs->hs_cur_bits = hs->hs_rehash_bits;
1934 out:
1935         hs->hs_rehash_bits = 0;
1936         if (rc == -ESRCH) /* never be scheduled again */
1937                 cfs_wi_exit(cfs_sched_rehash, wi);
1938         bsize = cfs_hash_bkt_size(hs);
1939         cfs_hash_unlock(hs, 1);
1940         /* can't refer to @hs anymore because it could be destroyed */
1941         if (bkts)
1942                 cfs_hash_buckets_free(bkts, bsize, new_size, old_size);
1943         if (rc != 0)
1944                 CDEBUG(D_INFO, "early quit of rehashing: %d\n", rc);
1945         /* return 1 only if cfs_wi_exit is called */
1946         return rc == -ESRCH;
1947 }
1948
1949 /**
1950  * Rehash the object referenced by @hnode in the libcfs hash @hs.  The
1951  * @old_key must be provided to locate the objects previous location
1952  * in the hash, and the @new_key will be used to reinsert the object.
1953  * Use this function instead of a cfs_hash_add() + cfs_hash_del()
1954  * combo when it is critical that there is no window in time where the
1955  * object is missing from the hash.  When an object is being rehashed
1956  * the registered cfs_hash_get() and cfs_hash_put() functions will
1957  * not be called.
1958  */
1959 void cfs_hash_rehash_key(struct cfs_hash *hs, const void *old_key,
1960                          void *new_key, struct hlist_node *hnode)
1961 {
1962         struct cfs_hash_bd bds[3];
1963         struct cfs_hash_bd old_bds[2];
1964         struct cfs_hash_bd new_bd;
1965
1966         LASSERT(!hlist_unhashed(hnode));
1967
1968         cfs_hash_lock(hs, 0);
1969
1970         cfs_hash_dual_bd_get(hs, old_key, old_bds);
1971         cfs_hash_bd_get(hs, new_key, &new_bd);
1972
1973         bds[0] = old_bds[0];
1974         bds[1] = old_bds[1];
1975         bds[2] = new_bd;
1976
1977         /* NB: bds[0] and bds[1] are ordered already */
1978         cfs_hash_bd_order(&bds[1], &bds[2]);
1979         cfs_hash_bd_order(&bds[0], &bds[1]);
1980
1981         cfs_hash_multi_bd_lock(hs, bds, 3, 1);
1982         if (likely(!old_bds[1].bd_bucket)) {
1983                 cfs_hash_bd_move_locked(hs, &old_bds[0], &new_bd, hnode);
1984         } else {
1985                 cfs_hash_dual_bd_finddel_locked(hs, old_bds, old_key, hnode);
1986                 cfs_hash_bd_add_locked(hs, &new_bd, hnode);
1987         }
1988         /* overwrite key inside locks, otherwise may screw up with
1989          * other operations, i.e: rehash
1990          */
1991         cfs_hash_keycpy(hs, hnode, new_key);
1992
1993         cfs_hash_multi_bd_unlock(hs, bds, 3, 1);
1994         cfs_hash_unlock(hs, 0);
1995 }
1996 EXPORT_SYMBOL(cfs_hash_rehash_key);
1997
1998 void cfs_hash_debug_header(struct seq_file *m)
1999 {
2000         seq_printf(m, "%-*s   cur   min   max theta t-min t-max flags rehash   count  maxdep maxdepb distribution\n",
2001                    CFS_HASH_BIGNAME_LEN, "name");
2002 }
2003 EXPORT_SYMBOL(cfs_hash_debug_header);
2004
2005 static struct cfs_hash_bucket **
2006 cfs_hash_full_bkts(struct cfs_hash *hs)
2007 {
2008         /* NB: caller should hold hs->hs_rwlock if REHASH is set */
2009         if (!hs->hs_rehash_buckets)
2010                 return hs->hs_buckets;
2011
2012         LASSERT(hs->hs_rehash_bits != 0);
2013         return hs->hs_rehash_bits > hs->hs_cur_bits ?
2014                hs->hs_rehash_buckets : hs->hs_buckets;
2015 }
2016
2017 static unsigned int
2018 cfs_hash_full_nbkt(struct cfs_hash *hs)
2019 {
2020         /* NB: caller should hold hs->hs_rwlock if REHASH is set */
2021         if (!hs->hs_rehash_buckets)
2022                 return CFS_HASH_NBKT(hs);
2023
2024         LASSERT(hs->hs_rehash_bits != 0);
2025         return hs->hs_rehash_bits > hs->hs_cur_bits ?
2026                CFS_HASH_RH_NBKT(hs) : CFS_HASH_NBKT(hs);
2027 }
2028
2029 void cfs_hash_debug_str(struct cfs_hash *hs, struct seq_file *m)
2030 {
2031         int dist[8] = { 0, };
2032         int maxdep = -1;
2033         int maxdepb = -1;
2034         int total = 0;
2035         int theta;
2036         int i;
2037
2038         cfs_hash_lock(hs, 0);
2039         theta = __cfs_hash_theta(hs);
2040
2041         seq_printf(m, "%-*s %5d %5d %5d %d.%03d %d.%03d %d.%03d  0x%02x %6d ",
2042                    CFS_HASH_BIGNAME_LEN, hs->hs_name,
2043                    1 << hs->hs_cur_bits, 1 << hs->hs_min_bits,
2044                    1 << hs->hs_max_bits,
2045                    __cfs_hash_theta_int(theta), __cfs_hash_theta_frac(theta),
2046                    __cfs_hash_theta_int(hs->hs_min_theta),
2047                    __cfs_hash_theta_frac(hs->hs_min_theta),
2048                    __cfs_hash_theta_int(hs->hs_max_theta),
2049                    __cfs_hash_theta_frac(hs->hs_max_theta),
2050                    hs->hs_flags, hs->hs_rehash_count);
2051
2052         /*
2053          * The distribution is a summary of the chained hash depth in
2054          * each of the libcfs hash buckets.  Each buckets hsb_count is
2055          * divided by the hash theta value and used to generate a
2056          * histogram of the hash distribution.  A uniform hash will
2057          * result in all hash buckets being close to the average thus
2058          * only the first few entries in the histogram will be non-zero.
2059          * If you hash function results in a non-uniform hash the will
2060          * be observable by outlier bucks in the distribution histogram.
2061          *
2062          * Uniform hash distribution:           128/128/0/0/0/0/0/0
2063          * Non-Uniform hash distribution:       128/125/0/0/0/0/2/1
2064          */
2065         for (i = 0; i < cfs_hash_full_nbkt(hs); i++) {
2066                 struct cfs_hash_bd bd;
2067
2068                 bd.bd_bucket = cfs_hash_full_bkts(hs)[i];
2069                 cfs_hash_bd_lock(hs, &bd, 0);
2070                 if (maxdep < bd.bd_bucket->hsb_depmax) {
2071                         maxdep  = bd.bd_bucket->hsb_depmax;
2072                         maxdepb = ffz(~maxdep);
2073                 }
2074                 total += bd.bd_bucket->hsb_count;
2075                 dist[min(fls(bd.bd_bucket->hsb_count / max(theta, 1)), 7)]++;
2076                 cfs_hash_bd_unlock(hs, &bd, 0);
2077         }
2078
2079         seq_printf(m, "%7d %7d %7d ", total, maxdep, maxdepb);
2080         for (i = 0; i < 8; i++)
2081                 seq_printf(m, "%d%c",  dist[i], (i == 7) ? '\n' : '/');
2082
2083         cfs_hash_unlock(hs, 0);
2084 }
2085 EXPORT_SYMBOL(cfs_hash_debug_str);