lustre/obdclass/lu_object.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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  28  * Use is subject to license terms.
  29  *
  30  * Copyright (c) 2011, 2012, Whamcloud, Inc.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * lustre/obdclass/lu_object.c
  37  *
  38  * Lustre Object.
  39  * These are the only exported functions, they provide some generic
  40  * infrastructure for managing object devices
  41  *
  42  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  43  */
  44
  45 #define DEBUG_SUBSYSTEM S_CLASS
  46
  47 #include <libcfs/libcfs.h>
  48
  49 #ifdef __KERNEL__
  50 # include <linux/module.h>
  51 #endif
  52
  53 /* hash_long() */
  54 #include <libcfs/libcfs_hash.h>
  55 #include <obd_class.h>
  56 #include <obd_support.h>
  57 #include <lustre_disk.h>
  58 #include <lustre_fid.h>
  59 #include <lu_object.h>
  60 #include <libcfs/list.h>
  61 /* lu_time_global_{init,fini}() */
  62 #include <lu_time.h>
  63
  64 static void lu_object_free(const struct lu_env *env, struct lu_object *o);
  65
  66 /**
  67  * Decrease reference counter on object. If last reference is freed, return
  68  * object to the cache, unless lu_object_is_dying(o) holds. In the latter
  69  * case, free object immediately.
  70  */
  71 void lu_object_put(const struct lu_env *env, struct lu_object *o)
  72 {
  73         struct lu_site_bkt_data *bkt;
  74         struct lu_object_header *top;
  75         struct lu_site          *site;
  76         struct lu_object        *orig;
  77         cfs_hash_bd_t            bd;
  78
  79         top  = o->lo_header;
  80         site = o->lo_dev->ld_site;
  81         orig = o;
  82
  83         cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
  84         bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
  85
  86         if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) {
  87                 if (lu_object_is_dying(top)) {
  88
  89                         /*
  90                          * somebody may be waiting for this, currently only
  91                          * used for cl_object, see cl_object_put_last().
  92                          */
  93                         cfs_waitq_broadcast(&bkt->lsb_marche_funebre);
  94                 }
  95                 return;
  96         }
  97
  98         LASSERT(bkt->lsb_busy > 0);
  99         bkt->lsb_busy--;
 100         /*
 101          * When last reference is released, iterate over object
 102          * layers, and notify them that object is no longer busy.
 103          */
 104         cfs_list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
 105                 if (o->lo_ops->loo_object_release != NULL)
 106                         o->lo_ops->loo_object_release(env, o);
 107         }
 108
 109         if (!lu_object_is_dying(top)) {
 110                 LASSERT(cfs_list_empty(&top->loh_lru));
 111                 cfs_list_add_tail(&top->loh_lru, &bkt->lsb_lru);
 112                 cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 113                 return;
 114         }
 115
 116         /*
 117          * If object is dying (will not be cached), removed it
 118          * from hash table and LRU.
 119          *
 120          * This is done with hash table and LRU lists locked. As the only
 121          * way to acquire first reference to previously unreferenced
 122          * object is through hash-table lookup (lu_object_find()),
 123          * or LRU scanning (lu_site_purge()), that are done under hash-table
 124          * and LRU lock, no race with concurrent object lookup is possible
 125          * and we can safely destroy object below.
 126          */
 127         cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
 128         cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 129         /*
 130          * Object was already removed from hash and lru above, can
 131          * kill it.
 132          */
 133         lu_object_free(env, orig);
 134 }
 135 EXPORT_SYMBOL(lu_object_put);
 136
 137 /**
 138  * Put object and don't keep in cache. This is temporary solution for
 139  * multi-site objects when its layering is not constant.
 140  */
 141 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o)
 142 {
 143         cfs_set_bit(LU_OBJECT_HEARD_BANSHEE,
 144                     &o->lo_header->loh_flags);
 145         return lu_object_put(env, o);
 146 }
 147 EXPORT_SYMBOL(lu_object_put_nocache);
 148
 149 /**
 150  * Allocate new object.
 151  *
 152  * This follows object creation protocol, described in the comment within
 153  * struct lu_device_operations definition.
 154  */
 155 static struct lu_object *lu_object_alloc(const struct lu_env *env,
 156                                          struct lu_device *dev,
 157                                          const struct lu_fid *f,
 158                                          const struct lu_object_conf *conf)
 159 {
 160         struct lu_object *scan;
 161         struct lu_object *top;
 162         cfs_list_t *layers;
 163         int clean;
 164         int result;
 165         ENTRY;
 166
 167         /*
 168          * Create top-level object slice. This will also create
 169          * lu_object_header.
 170          */
 171         top = dev->ld_ops->ldo_object_alloc(env, NULL, dev);
 172         if (top == NULL)
 173                 RETURN(ERR_PTR(-ENOMEM));
 174         /*
 175          * This is the only place where object fid is assigned. It's constant
 176          * after this point.
 177          */
 178         LASSERT(fid_is_igif(f) || fid_ver(f) == 0);
 179         top->lo_header->loh_fid = *f;
 180         layers = &top->lo_header->loh_layers;
 181         do {
 182                 /*
 183                  * Call ->loo_object_init() repeatedly, until no more new
 184                  * object slices are created.
 185                  */
 186                 clean = 1;
 187                 cfs_list_for_each_entry(scan, layers, lo_linkage) {
 188                         if (scan->lo_flags & LU_OBJECT_ALLOCATED)
 189                                 continue;
 190                         clean = 0;
 191                         scan->lo_header = top->lo_header;
 192                         result = scan->lo_ops->loo_object_init(env, scan, conf);
 193                         if (result != 0) {
 194                                 lu_object_free(env, top);
 195                                 RETURN(ERR_PTR(result));
 196                         }
 197                         scan->lo_flags |= LU_OBJECT_ALLOCATED;
 198                 }
 199         } while (!clean);
 200
 201         cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
 202                 if (scan->lo_ops->loo_object_start != NULL) {
 203                         result = scan->lo_ops->loo_object_start(env, scan);
 204                         if (result != 0) {
 205                                 lu_object_free(env, top);
 206                                 RETURN(ERR_PTR(result));
 207                         }
 208                 }
 209         }
 210
 211         lprocfs_counter_incr(dev->ld_site->ls_stats, LU_SS_CREATED);
 212         RETURN(top);
 213 }
 214
 215 /**
 216  * Free an object.
 217  */
 218 static void lu_object_free(const struct lu_env *env, struct lu_object *o)
 219 {
 220         struct lu_site_bkt_data *bkt;
 221         struct lu_site          *site;
 222         struct lu_object        *scan;
 223         cfs_list_t              *layers;
 224         cfs_list_t               splice;
 225
 226         site   = o->lo_dev->ld_site;
 227         layers = &o->lo_header->loh_layers;
 228         bkt    = lu_site_bkt_from_fid(site, &o->lo_header->loh_fid);
 229         /*
 230          * First call ->loo_object_delete() method to release all resources.
 231          */
 232         cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
 233                 if (scan->lo_ops->loo_object_delete != NULL)
 234                         scan->lo_ops->loo_object_delete(env, scan);
 235         }
 236
 237         /*
 238          * Then, splice object layers into stand-alone list, and call
 239          * ->loo_object_free() on all layers to free memory. Splice is
 240          * necessary, because lu_object_header is freed together with the
 241          * top-level slice.
 242          */
 243         CFS_INIT_LIST_HEAD(&splice);
 244         cfs_list_splice_init(layers, &splice);
 245         while (!cfs_list_empty(&splice)) {
 246                 /*
 247                  * Free layers in bottom-to-top order, so that object header
 248                  * lives as long as possible and ->loo_object_free() methods
 249                  * can look at its contents.
 250                  */
 251                 o = container_of0(splice.prev, struct lu_object, lo_linkage);
 252                 cfs_list_del_init(&o->lo_linkage);
 253                 LASSERT(o->lo_ops->loo_object_free != NULL);
 254                 o->lo_ops->loo_object_free(env, o);
 255         }
 256
 257         if (cfs_waitq_active(&bkt->lsb_marche_funebre))
 258                 cfs_waitq_broadcast(&bkt->lsb_marche_funebre);
 259 }
 260
 261 /**
 262  * Free \a nr objects from the cold end of the site LRU list.
 263  */
 264 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
 265 {
 266         struct lu_object_header *h;
 267         struct lu_object_header *temp;
 268         struct lu_site_bkt_data *bkt;
 269         cfs_hash_bd_t            bd;
 270         cfs_hash_bd_t            bd2;
 271         cfs_list_t               dispose;
 272         int                      did_sth;
 273         int                      start;
 274         int                      count;
 275         int                      bnr;
 276         int                      i;
 277
 278         CFS_INIT_LIST_HEAD(&dispose);
 279         /*
 280          * Under LRU list lock, scan LRU list and move unreferenced objects to
 281          * the dispose list, removing them from LRU and hash table.
 282          */
 283         start = s->ls_purge_start;
 284         bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1;
 285  again:
 286         did_sth = 0;
 287         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
 288                 if (i < start)
 289                         continue;
 290                 count = bnr;
 291                 cfs_hash_bd_lock(s->ls_obj_hash, &bd, 1);
 292                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 293
 294                 cfs_list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
 295                         LASSERT(cfs_atomic_read(&h->loh_ref) == 0);
 296
 297                         cfs_hash_bd_get(s->ls_obj_hash, &h->loh_fid, &bd2);
 298                         LASSERT(bd.bd_bucket == bd2.bd_bucket);
 299
 300                         cfs_hash_bd_del_locked(s->ls_obj_hash,
 301                                                &bd2, &h->loh_hash);
 302                         cfs_list_move(&h->loh_lru, &dispose);
 303                         if (did_sth == 0)
 304                                 did_sth = 1;
 305
 306                         if (nr != ~0 && --nr == 0)
 307                                 break;
 308
 309                         if (count > 0 && --count == 0)
 310                                 break;
 311
 312                 }
 313                 cfs_hash_bd_unlock(s->ls_obj_hash, &bd, 1);
 314                 cfs_cond_resched();
 315                 /*
 316                  * Free everything on the dispose list. This is safe against
 317                  * races due to the reasons described in lu_object_put().
 318                  */
 319                 while (!cfs_list_empty(&dispose)) {
 320                         h = container_of0(dispose.next,
 321                                           struct lu_object_header, loh_lru);
 322                         cfs_list_del_init(&h->loh_lru);
 323                         lu_object_free(env, lu_object_top(h));
 324                         lprocfs_counter_incr(s->ls_stats, LU_SS_LRU_PURGED);
 325                 }
 326
 327                 if (nr == 0)
 328                         break;
 329         }
 330
 331         if (nr != 0 && did_sth && start != 0) {
 332                 start = 0; /* restart from the first bucket */
 333                 goto again;
 334         }
 335         /* race on s->ls_purge_start, but nobody cares */
 336         s->ls_purge_start = i % CFS_HASH_NBKT(s->ls_obj_hash);
 337
 338         return nr;
 339 }
 340 EXPORT_SYMBOL(lu_site_purge);
 341
 342 /*
 343  * Object printing.
 344  *
 345  * Code below has to jump through certain loops to output object description
 346  * into libcfs_debug_msg-based log. The problem is that lu_object_print()
 347  * composes object description from strings that are parts of _lines_ of
 348  * output (i.e., strings that are not terminated by newline). This doesn't fit
 349  * very well into libcfs_debug_msg() interface that assumes that each message
 350  * supplied to it is a self-contained output line.
 351  *
 352  * To work around this, strings are collected in a temporary buffer
 353  * (implemented as a value of lu_cdebug_key key), until terminating newline
 354  * character is detected.
 355  *
 356  */
 357
 358 enum {
 359         /**
 360          * Maximal line size.
 361          *
 362          * XXX overflow is not handled correctly.
 363          */
 364         LU_CDEBUG_LINE = 512
 365 };
 366
 367 struct lu_cdebug_data {
 368         /**
 369          * Temporary buffer.
 370          */
 371         char lck_area[LU_CDEBUG_LINE];
 372 };
 373
 374 /* context key constructor/destructor: lu_global_key_init, lu_global_key_fini */
 375 LU_KEY_INIT_FINI(lu_global, struct lu_cdebug_data);
 376
 377 /**
 378  * Key, holding temporary buffer. This key is registered very early by
 379  * lu_global_init().
 380  */
 381 struct lu_context_key lu_global_key = {
 382         .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD |
 383                     LCT_MG_THREAD | LCT_CL_THREAD,
 384         .lct_init = lu_global_key_init,
 385         .lct_fini = lu_global_key_fini
 386 };
 387
 388 /**
 389  * Printer function emitting messages through libcfs_debug_msg().
 390  */
 391 int lu_cdebug_printer(const struct lu_env *env,
 392                       void *cookie, const char *format, ...)
 393 {
 394         struct libcfs_debug_msg_data *msgdata = cookie;
 395         struct lu_cdebug_data        *key;
 396         int used;
 397         int complete;
 398         va_list args;
 399
 400         va_start(args, format);
 401
 402         key = lu_context_key_get(&env->le_ctx, &lu_global_key);
 403         LASSERT(key != NULL);
 404
 405         used = strlen(key->lck_area);
 406         complete = format[strlen(format) - 1] == '\n';
 407         /*
 408          * Append new chunk to the buffer.
 409          */
 410         vsnprintf(key->lck_area + used,
 411                   ARRAY_SIZE(key->lck_area) - used, format, args);
 412         if (complete) {
 413                 if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
 414                         libcfs_debug_msg(msgdata, "%s", key->lck_area);
 415                 key->lck_area[0] = 0;
 416         }
 417         va_end(args);
 418         return 0;
 419 }
 420 EXPORT_SYMBOL(lu_cdebug_printer);
 421
 422 /**
 423  * Print object header.
 424  */
 425 void lu_object_header_print(const struct lu_env *env, void *cookie,
 426                             lu_printer_t printer,
 427                             const struct lu_object_header *hdr)
 428 {
 429         (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]",
 430                    hdr, hdr->loh_flags, cfs_atomic_read(&hdr->loh_ref),
 431                    PFID(&hdr->loh_fid),
 432                    cfs_hlist_unhashed(&hdr->loh_hash) ? "" : " hash",
 433                    cfs_list_empty((cfs_list_t *)&hdr->loh_lru) ? \
 434                    "" : " lru",
 435                    hdr->loh_attr & LOHA_EXISTS ? " exist":"");
 436 }
 437 EXPORT_SYMBOL(lu_object_header_print);
 438
 439 /**
 440  * Print human readable representation of the \a o to the \a printer.
 441  */
 442 void lu_object_print(const struct lu_env *env, void *cookie,
 443                      lu_printer_t printer, const struct lu_object *o)
 444 {
 445         static const char ruler[] = "........................................";
 446         struct lu_object_header *top;
 447         int depth;
 448
 449         top = o->lo_header;
 450         lu_object_header_print(env, cookie, printer, top);
 451         (*printer)(env, cookie, "{ \n");
 452         cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 453                 depth = o->lo_depth + 4;
 454
 455                 /*
 456                  * print `.' \a depth times followed by type name and address
 457                  */
 458                 (*printer)(env, cookie, "%*.*s%s@%p", depth, depth, ruler,
 459                            o->lo_dev->ld_type->ldt_name, o);
 460                 if (o->lo_ops->loo_object_print != NULL)
 461                         o->lo_ops->loo_object_print(env, cookie, printer, o);
 462                 (*printer)(env, cookie, "\n");
 463         }
 464         (*printer)(env, cookie, "} header@%p\n", top);
 465 }
 466 EXPORT_SYMBOL(lu_object_print);
 467
 468 /**
 469  * Check object consistency.
 470  */
 471 int lu_object_invariant(const struct lu_object *o)
 472 {
 473         struct lu_object_header *top;
 474
 475         top = o->lo_header;
 476         cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 477                 if (o->lo_ops->loo_object_invariant != NULL &&
 478                     !o->lo_ops->loo_object_invariant(o))
 479                         return 0;
 480         }
 481         return 1;
 482 }
 483 EXPORT_SYMBOL(lu_object_invariant);
 484
 485 static struct lu_object *htable_lookup(struct lu_site *s,
 486                                        cfs_hash_bd_t *bd,
 487                                        const struct lu_fid *f,
 488                                        cfs_waitlink_t *waiter,
 489                                        __u64 *version)
 490 {
 491         struct lu_site_bkt_data *bkt;
 492         struct lu_object_header *h;
 493         cfs_hlist_node_t        *hnode;
 494         __u64  ver = cfs_hash_bd_version_get(bd);
 495
 496         if (*version == ver)
 497                 return NULL;
 498
 499         *version = ver;
 500         bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
 501         /* cfs_hash_bd_lookup_intent is a somehow "internal" function
 502          * of cfs_hash, but we don't want refcount on object right now */
 503         hnode = cfs_hash_bd_lookup_locked(s->ls_obj_hash, bd, (void *)f);
 504         if (hnode == NULL) {
 505                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
 506                 return NULL;
 507         }
 508
 509         h = container_of0(hnode, struct lu_object_header, loh_hash);
 510         if (likely(!lu_object_is_dying(h))) {
 511                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
 512                 cfs_list_del_init(&h->loh_lru);
 513                 return lu_object_top(h);
 514         }
 515
 516         /*
 517          * Lookup found an object being destroyed this object cannot be
 518          * returned (to assure that references to dying objects are eventually
 519          * drained), and moreover, lookup has to wait until object is freed.
 520          */
 521         cfs_atomic_dec(&h->loh_ref);
 522
 523         cfs_waitlink_init(waiter);
 524         cfs_waitq_add(&bkt->lsb_marche_funebre, waiter);
 525         cfs_set_current_state(CFS_TASK_UNINT);
 526         lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE);
 527         return ERR_PTR(-EAGAIN);
 528 }
 529
 530 /**
 531  * Search cache for an object with the fid \a f. If such object is found,
 532  * return it. Otherwise, create new object, insert it into cache and return
 533  * it. In any case, additional reference is acquired on the returned object.
 534  */
 535 struct lu_object *lu_object_find(const struct lu_env *env,
 536                                  struct lu_device *dev, const struct lu_fid *f,
 537                                  const struct lu_object_conf *conf)
 538 {
 539         return lu_object_find_at(env, dev->ld_site->ls_top_dev, f, conf);
 540 }
 541 EXPORT_SYMBOL(lu_object_find);
 542
 543 static struct lu_object *lu_object_new(const struct lu_env *env,
 544                                        struct lu_device *dev,
 545                                        const struct lu_fid *f,
 546                                        const struct lu_object_conf *conf)
 547 {
 548         struct lu_object        *o;
 549         cfs_hash_t              *hs;
 550         cfs_hash_bd_t            bd;
 551         struct lu_site_bkt_data *bkt;
 552
 553         o = lu_object_alloc(env, dev, f, conf);
 554         if (unlikely(IS_ERR(o)))
 555                 return o;
 556
 557         hs = dev->ld_site->ls_obj_hash;
 558         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
 559         bkt = cfs_hash_bd_extra_get(hs, &bd);
 560         cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
 561         bkt->lsb_busy++;
 562         cfs_hash_bd_unlock(hs, &bd, 1);
 563         return o;
 564 }
 565
 566 /**
 567  * Core logic of lu_object_find*() functions.
 568  */
 569 static struct lu_object *lu_object_find_try(const struct lu_env *env,
 570                                             struct lu_device *dev,
 571                                             const struct lu_fid *f,
 572                                             const struct lu_object_conf *conf,
 573                                             cfs_waitlink_t *waiter)
 574 {
 575         struct lu_object      *o;
 576         struct lu_object      *shadow;
 577         struct lu_site        *s;
 578         cfs_hash_t            *hs;
 579         cfs_hash_bd_t          bd;
 580         __u64                  version = 0;
 581
 582         /*
 583          * This uses standard index maintenance protocol:
 584          *
 585          *     - search index under lock, and return object if found;
 586          *     - otherwise, unlock index, allocate new object;
 587          *     - lock index and search again;
 588          *     - if nothing is found (usual case), insert newly created
 589          *       object into index;
 590          *     - otherwise (race: other thread inserted object), free
 591          *       object just allocated.
 592          *     - unlock index;
 593          *     - return object.
 594          *
 595          * For "LOC_F_NEW" case, we are sure the object is new established.
 596          * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
 597          * just alloc and insert directly.
 598          *
 599          * If dying object is found during index search, add @waiter to the
 600          * site wait-queue and return ERR_PTR(-EAGAIN).
 601          */
 602         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 603                 return lu_object_new(env, dev, f, conf);
 604
 605         s  = dev->ld_site;
 606         hs = s->ls_obj_hash;
 607         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
 608         o = htable_lookup(s, &bd, f, waiter, &version);
 609         cfs_hash_bd_unlock(hs, &bd, 1);
 610         if (o != NULL)
 611                 return o;
 612
 613         /*
 614          * Allocate new object. This may result in rather complicated
 615          * operations, including fld queries, inode loading, etc.
 616          */
 617         o = lu_object_alloc(env, dev, f, conf);
 618         if (unlikely(IS_ERR(o)))
 619                 return o;
 620
 621         LASSERT(lu_fid_eq(lu_object_fid(o), f));
 622
 623         cfs_hash_bd_lock(hs, &bd, 1);
 624
 625         shadow = htable_lookup(s, &bd, f, waiter, &version);
 626         if (likely(shadow == NULL)) {
 627                 struct lu_site_bkt_data *bkt;
 628
 629                 bkt = cfs_hash_bd_extra_get(hs, &bd);
 630                 cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
 631                 bkt->lsb_busy++;
 632                 cfs_hash_bd_unlock(hs, &bd, 1);
 633                 return o;
 634         }
 635
 636         lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE);
 637         cfs_hash_bd_unlock(hs, &bd, 1);
 638         lu_object_free(env, o);
 639         return shadow;
 640 }
 641
 642 /**
 643  * Much like lu_object_find(), but top level device of object is specifically
 644  * \a dev rather than top level device of the site. This interface allows
 645  * objects of different "stacking" to be created within the same site.
 646  */
 647 struct lu_object *lu_object_find_at(const struct lu_env *env,
 648                                     struct lu_device *dev,
 649                                     const struct lu_fid *f,
 650                                     const struct lu_object_conf *conf)
 651 {
 652         struct lu_site_bkt_data *bkt;
 653         struct lu_object        *obj;
 654         cfs_waitlink_t           wait;
 655
 656         while (1) {
 657                 obj = lu_object_find_try(env, dev, f, conf, &wait);
 658                 if (obj != ERR_PTR(-EAGAIN))
 659                         return obj;
 660                 /*
 661                  * lu_object_find_try() already added waiter into the
 662                  * wait queue.
 663                  */
 664                 cfs_waitq_wait(&wait, CFS_TASK_UNINT);
 665                 bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f);
 666                 cfs_waitq_del(&bkt->lsb_marche_funebre, &wait);
 667         }
 668 }
 669 EXPORT_SYMBOL(lu_object_find_at);
 670
 671 /**
 672  * Find object with given fid, and return its slice belonging to given device.
 673  */
 674 struct lu_object *lu_object_find_slice(const struct lu_env *env,
 675                                        struct lu_device *dev,
 676                                        const struct lu_fid *f,
 677                                        const struct lu_object_conf *conf)
 678 {
 679         struct lu_object *top;
 680         struct lu_object *obj;
 681
 682         top = lu_object_find(env, dev, f, conf);
 683         if (!IS_ERR(top)) {
 684                 obj = lu_object_locate(top->lo_header, dev->ld_type);
 685                 if (obj == NULL)
 686                         lu_object_put(env, top);
 687         } else
 688                 obj = top;
 689         return obj;
 690 }
 691 EXPORT_SYMBOL(lu_object_find_slice);
 692
 693 /**
 694  * Global list of all device types.
 695  */
 696 static CFS_LIST_HEAD(lu_device_types);
 697
 698 int lu_device_type_init(struct lu_device_type *ldt)
 699 {
 700         int result;
 701
 702         CFS_INIT_LIST_HEAD(&ldt->ldt_linkage);
 703         result = ldt->ldt_ops->ldto_init(ldt);
 704         if (result == 0)
 705                 cfs_list_add(&ldt->ldt_linkage, &lu_device_types);
 706         return result;
 707 }
 708 EXPORT_SYMBOL(lu_device_type_init);
 709
 710 void lu_device_type_fini(struct lu_device_type *ldt)
 711 {
 712         cfs_list_del_init(&ldt->ldt_linkage);
 713         ldt->ldt_ops->ldto_fini(ldt);
 714 }
 715 EXPORT_SYMBOL(lu_device_type_fini);
 716
 717 void lu_types_stop(void)
 718 {
 719         struct lu_device_type *ldt;
 720
 721         cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
 722                 if (ldt->ldt_device_nr == 0)
 723                         ldt->ldt_ops->ldto_stop(ldt);
 724         }
 725 }
 726 EXPORT_SYMBOL(lu_types_stop);
 727
 728 /**
 729  * Global list of all sites on this node
 730  */
 731 static CFS_LIST_HEAD(lu_sites);
 732 static CFS_DEFINE_MUTEX(lu_sites_guard);
 733
 734 /**
 735  * Global environment used by site shrinker.
 736  */
 737 static struct lu_env lu_shrink_env;
 738
 739 struct lu_site_print_arg {
 740         struct lu_env   *lsp_env;
 741         void            *lsp_cookie;
 742         lu_printer_t     lsp_printer;
 743 };
 744
 745 static int
 746 lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 747                   cfs_hlist_node_t *hnode, void *data)
 748 {
 749         struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data;
 750         struct lu_object_header  *h;
 751
 752         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 753         if (!cfs_list_empty(&h->loh_layers)) {
 754                 const struct lu_object *o;
 755
 756                 o = lu_object_top(h);
 757                 lu_object_print(arg->lsp_env, arg->lsp_cookie,
 758                                 arg->lsp_printer, o);
 759         } else {
 760                 lu_object_header_print(arg->lsp_env, arg->lsp_cookie,
 761                                        arg->lsp_printer, h);
 762         }
 763         return 0;
 764 }
 765
 766 /**
 767  * Print all objects in \a s.
 768  */
 769 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
 770                    lu_printer_t printer)
 771 {
 772         struct lu_site_print_arg arg = {
 773                 .lsp_env     = (struct lu_env *)env,
 774                 .lsp_cookie  = cookie,
 775                 .lsp_printer = printer,
 776         };
 777
 778         cfs_hash_for_each(s->ls_obj_hash, lu_site_obj_print, &arg);
 779 }
 780 EXPORT_SYMBOL(lu_site_print);
 781
 782 enum {
 783         LU_CACHE_PERCENT_MAX     = 50,
 784         LU_CACHE_PERCENT_DEFAULT = 20
 785 };
 786
 787 static unsigned int lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
 788 CFS_MODULE_PARM(lu_cache_percent, "i", int, 0644,
 789                 "Percentage of memory to be used as lu_object cache");
 790
 791 /**
 792  * Return desired hash table order.
 793  */
 794 static int lu_htable_order(void)
 795 {
 796         unsigned long cache_size;
 797         int bits;
 798
 799         /*
 800          * Calculate hash table size, assuming that we want reasonable
 801          * performance when 20% of total memory is occupied by cache of
 802          * lu_objects.
 803          *
 804          * Size of lu_object is (arbitrary) taken as 1K (together with inode).
 805          */
 806         cache_size = cfs_num_physpages;
 807
 808 #if BITS_PER_LONG == 32
 809         /* limit hashtable size for lowmem systems to low RAM */
 810         if (cache_size > 1 << (30 - CFS_PAGE_SHIFT))
 811                 cache_size = 1 << (30 - CFS_PAGE_SHIFT) * 3 / 4;
 812 #endif
 813
 814         /* clear off unreasonable cache setting. */
 815         if (lu_cache_percent == 0 || lu_cache_percent > LU_CACHE_PERCENT_MAX) {
 816                 CWARN("obdclass: invalid lu_cache_percent: %u, it must be in"
 817                       " the range of (0, %u]. Will use default value: %u.\n",
 818                       lu_cache_percent, LU_CACHE_PERCENT_MAX,
 819                       LU_CACHE_PERCENT_DEFAULT);
 820
 821                 lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
 822         }
 823         cache_size = cache_size / 100 * lu_cache_percent *
 824                 (CFS_PAGE_SIZE / 1024);
 825
 826         for (bits = 1; (1 << bits) < cache_size; ++bits) {
 827                 ;
 828         }
 829         return bits;
 830 }
 831
 832 static unsigned lu_obj_hop_hash(cfs_hash_t *hs,
 833                                 const void *key, unsigned mask)
 834 {
 835         struct lu_fid  *fid = (struct lu_fid *)key;
 836         __u32           hash;
 837
 838         hash = fid_flatten32(fid);
 839         hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
 840         hash = cfs_hash_long(hash, hs->hs_bkt_bits);
 841
 842         /* give me another random factor */
 843         hash -= cfs_hash_long((unsigned long)hs, fid_oid(fid) % 11 + 3);
 844
 845         hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
 846         hash |= (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
 847
 848         return hash & mask;
 849 }
 850
 851 static void *lu_obj_hop_object(cfs_hlist_node_t *hnode)
 852 {
 853         return cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 854 }
 855
 856 static void *lu_obj_hop_key(cfs_hlist_node_t *hnode)
 857 {
 858         struct lu_object_header *h;
 859
 860         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 861         return &h->loh_fid;
 862 }
 863
 864 static int lu_obj_hop_keycmp(const void *key, cfs_hlist_node_t *hnode)
 865 {
 866         struct lu_object_header *h;
 867
 868         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 869         return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
 870 }
 871
 872 static void lu_obj_hop_get(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 873 {
 874         struct lu_object_header *h;
 875
 876         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 877         if (cfs_atomic_add_return(1, &h->loh_ref) == 1) {
 878                 struct lu_site_bkt_data *bkt;
 879                 cfs_hash_bd_t            bd;
 880
 881                 cfs_hash_bd_get(hs, &h->loh_fid, &bd);
 882                 bkt = cfs_hash_bd_extra_get(hs, &bd);
 883                 bkt->lsb_busy++;
 884         }
 885 }
 886
 887 static void lu_obj_hop_put_locked(cfs_hash_t *hs, cfs_hlist_node_t *hnode)
 888 {
 889         LBUG(); /* we should never called it */
 890 }
 891
 892 cfs_hash_ops_t lu_site_hash_ops = {
 893         .hs_hash        = lu_obj_hop_hash,
 894         .hs_key         = lu_obj_hop_key,
 895         .hs_keycmp      = lu_obj_hop_keycmp,
 896         .hs_object      = lu_obj_hop_object,
 897         .hs_get         = lu_obj_hop_get,
 898         .hs_put_locked  = lu_obj_hop_put_locked,
 899 };
 900
 901 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d)
 902 {
 903         cfs_spin_lock(&s->ls_ld_lock);
 904         if (cfs_list_empty(&d->ld_linkage))
 905                 cfs_list_add(&d->ld_linkage, &s->ls_ld_linkage);
 906         cfs_spin_unlock(&s->ls_ld_lock);
 907 }
 908 EXPORT_SYMBOL(lu_dev_add_linkage);
 909
 910 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d)
 911 {
 912         cfs_spin_lock(&s->ls_ld_lock);
 913         cfs_list_del_init(&d->ld_linkage);
 914         cfs_spin_unlock(&s->ls_ld_lock);
 915 }
 916 EXPORT_SYMBOL(lu_dev_del_linkage);
 917
 918 /**
 919  * Initialize site \a s, with \a d as the top level device.
 920  */
 921 #define LU_SITE_BITS_MIN    12
 922 #define LU_SITE_BITS_MAX    24
 923 /**
 924  * total 256 buckets, we don't want too many buckets because:
 925  * - consume too much memory
 926  * - avoid unbalanced LRU list
 927  */
 928 #define LU_SITE_BKT_BITS    8
 929
 930 int lu_site_init(struct lu_site *s, struct lu_device *top)
 931 {
 932         struct lu_site_bkt_data *bkt;
 933         cfs_hash_bd_t bd;
 934         char name[16];
 935         int bits;
 936         int i;
 937         ENTRY;
 938
 939         memset(s, 0, sizeof *s);
 940         bits = lu_htable_order();
 941         snprintf(name, 16, "lu_site_%s", top->ld_type->ldt_name);
 942         for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX);
 943              bits >= LU_SITE_BITS_MIN; bits--) {
 944                 s->ls_obj_hash = cfs_hash_create(name, bits, bits,
 945                                                  bits - LU_SITE_BKT_BITS,
 946                                                  sizeof(*bkt), 0, 0,
 947                                                  &lu_site_hash_ops,
 948                                                  CFS_HASH_SPIN_BKTLOCK |
 949                                                  CFS_HASH_NO_ITEMREF |
 950                                                  CFS_HASH_DEPTH |
 951                                                  CFS_HASH_ASSERT_EMPTY);
 952                 if (s->ls_obj_hash != NULL)
 953                         break;
 954         }
 955
 956         if (s->ls_obj_hash == NULL) {
 957                 CERROR("failed to create lu_site hash with bits: %d\n", bits);
 958                 return -ENOMEM;
 959         }
 960
 961         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
 962                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 963                 CFS_INIT_LIST_HEAD(&bkt->lsb_lru);
 964                 cfs_waitq_init(&bkt->lsb_marche_funebre);
 965         }
 966
 967         s->ls_stats = lprocfs_alloc_stats(LU_SS_LAST_STAT, 0);
 968         if (s->ls_stats == NULL) {
 969                 cfs_hash_putref(s->ls_obj_hash);
 970                 s->ls_obj_hash = NULL;
 971                 return -ENOMEM;
 972         }
 973
 974         lprocfs_counter_init(s->ls_stats, LU_SS_CREATED,
 975                              0, "created", "created");
 976         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_HIT,
 977                              0, "cache_hit", "cache_hit");
 978         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_MISS,
 979                              0, "cache_miss", "cache_miss");
 980         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_RACE,
 981                              0, "cache_race", "cache_race");
 982         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_DEATH_RACE,
 983                              0, "cache_death_race", "cache_death_race");
 984         lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED,
 985                              0, "lru_purged", "lru_purged");
 986
 987         CFS_INIT_LIST_HEAD(&s->ls_linkage);
 988         s->ls_top_dev = top;
 989         top->ld_site = s;
 990         lu_device_get(top);
 991         lu_ref_add(&top->ld_reference, "site-top", s);
 992
 993         CFS_INIT_LIST_HEAD(&s->ls_ld_linkage);
 994         cfs_spin_lock_init(&s->ls_ld_lock);
 995
 996         lu_dev_add_linkage(s, top);
 997
 998         RETURN(0);
 999 }
1000 EXPORT_SYMBOL(lu_site_init);
1001
1002 /**
1003  * Finalize \a s and release its resources.
1004  */
1005 void lu_site_fini(struct lu_site *s)
1006 {
1007         cfs_mutex_lock(&lu_sites_guard);
1008         cfs_list_del_init(&s->ls_linkage);
1009         cfs_mutex_unlock(&lu_sites_guard);
1010
1011         if (s->ls_obj_hash != NULL) {
1012                 cfs_hash_putref(s->ls_obj_hash);
1013                 s->ls_obj_hash = NULL;
1014         }
1015
1016         if (s->ls_top_dev != NULL) {
1017                 s->ls_top_dev->ld_site = NULL;
1018                 lu_ref_del(&s->ls_top_dev->ld_reference, "site-top", s);
1019                 lu_device_put(s->ls_top_dev);
1020                 s->ls_top_dev = NULL;
1021         }
1022
1023         if (s->ls_stats != NULL)
1024                 lprocfs_free_stats(&s->ls_stats);
1025 }
1026 EXPORT_SYMBOL(lu_site_fini);
1027
1028 /**
1029  * Called when initialization of stack for this site is completed.
1030  */
1031 int lu_site_init_finish(struct lu_site *s)
1032 {
1033         int result;
1034         cfs_mutex_lock(&lu_sites_guard);
1035         result = lu_context_refill(&lu_shrink_env.le_ctx);
1036         if (result == 0)
1037                 cfs_list_add(&s->ls_linkage, &lu_sites);
1038         cfs_mutex_unlock(&lu_sites_guard);
1039         return result;
1040 }
1041 EXPORT_SYMBOL(lu_site_init_finish);
1042
1043 /**
1044  * Acquire additional reference on device \a d
1045  */
1046 void lu_device_get(struct lu_device *d)
1047 {
1048         cfs_atomic_inc(&d->ld_ref);
1049 }
1050 EXPORT_SYMBOL(lu_device_get);
1051
1052 /**
1053  * Release reference on device \a d.
1054  */
1055 void lu_device_put(struct lu_device *d)
1056 {
1057         LASSERT(cfs_atomic_read(&d->ld_ref) > 0);
1058         cfs_atomic_dec(&d->ld_ref);
1059 }
1060 EXPORT_SYMBOL(lu_device_put);
1061
1062 /**
1063  * Initialize device \a d of type \a t.
1064  */
1065 int lu_device_init(struct lu_device *d, struct lu_device_type *t)
1066 {
1067         if (t->ldt_device_nr++ == 0 && t->ldt_ops->ldto_start != NULL)
1068                 t->ldt_ops->ldto_start(t);
1069         memset(d, 0, sizeof *d);
1070         cfs_atomic_set(&d->ld_ref, 0);
1071         d->ld_type = t;
1072         lu_ref_init(&d->ld_reference);
1073         CFS_INIT_LIST_HEAD(&d->ld_linkage);
1074         return 0;
1075 }
1076 EXPORT_SYMBOL(lu_device_init);
1077
1078 /**
1079  * Finalize device \a d.
1080  */
1081 void lu_device_fini(struct lu_device *d)
1082 {
1083         struct lu_device_type *t;
1084
1085         t = d->ld_type;
1086         if (d->ld_obd != NULL) {
1087                 d->ld_obd->obd_lu_dev = NULL;
1088                 d->ld_obd = NULL;
1089         }
1090
1091         lu_ref_fini(&d->ld_reference);
1092         LASSERTF(cfs_atomic_read(&d->ld_ref) == 0,
1093                  "Refcount is %u\n", cfs_atomic_read(&d->ld_ref));
1094         LASSERT(t->ldt_device_nr > 0);
1095         if (--t->ldt_device_nr == 0 && t->ldt_ops->ldto_stop != NULL)
1096                 t->ldt_ops->ldto_stop(t);
1097 }
1098 EXPORT_SYMBOL(lu_device_fini);
1099
1100 /**
1101  * Initialize object \a o that is part of compound object \a h and was created
1102  * by device \a d.
1103  */
1104 int lu_object_init(struct lu_object *o,
1105                    struct lu_object_header *h, struct lu_device *d)
1106 {
1107         memset(o, 0, sizeof *o);
1108         o->lo_header = h;
1109         o->lo_dev    = d;
1110         lu_device_get(d);
1111         o->lo_dev_ref = lu_ref_add(&d->ld_reference, "lu_object", o);
1112         CFS_INIT_LIST_HEAD(&o->lo_linkage);
1113         return 0;
1114 }
1115 EXPORT_SYMBOL(lu_object_init);
1116
1117 /**
1118  * Finalize object and release its resources.
1119  */
1120 void lu_object_fini(struct lu_object *o)
1121 {
1122         struct lu_device *dev = o->lo_dev;
1123
1124         LASSERT(cfs_list_empty(&o->lo_linkage));
1125
1126         if (dev != NULL) {
1127                 lu_ref_del_at(&dev->ld_reference,
1128                               o->lo_dev_ref , "lu_object", o);
1129                 lu_device_put(dev);
1130                 o->lo_dev = NULL;
1131         }
1132 }
1133 EXPORT_SYMBOL(lu_object_fini);
1134
1135 /**
1136  * Add object \a o as first layer of compound object \a h
1137  *
1138  * This is typically called by the ->ldo_object_alloc() method of top-level
1139  * device.
1140  */
1141 void lu_object_add_top(struct lu_object_header *h, struct lu_object *o)
1142 {
1143         cfs_list_move(&o->lo_linkage, &h->loh_layers);
1144 }
1145 EXPORT_SYMBOL(lu_object_add_top);
1146
1147 /**
1148  * Add object \a o as a layer of compound object, going after \a before.
1149  *
1150  * This is typically called by the ->ldo_object_alloc() method of \a
1151  * before->lo_dev.
1152  */
1153 void lu_object_add(struct lu_object *before, struct lu_object *o)
1154 {
1155         cfs_list_move(&o->lo_linkage, &before->lo_linkage);
1156 }
1157 EXPORT_SYMBOL(lu_object_add);
1158
1159 /**
1160  * Initialize compound object.
1161  */
1162 int lu_object_header_init(struct lu_object_header *h)
1163 {
1164         memset(h, 0, sizeof *h);
1165         cfs_atomic_set(&h->loh_ref, 1);
1166         CFS_INIT_HLIST_NODE(&h->loh_hash);
1167         CFS_INIT_LIST_HEAD(&h->loh_lru);
1168         CFS_INIT_LIST_HEAD(&h->loh_layers);
1169         lu_ref_init(&h->loh_reference);
1170         return 0;
1171 }
1172 EXPORT_SYMBOL(lu_object_header_init);
1173
1174 /**
1175  * Finalize compound object.
1176  */
1177 void lu_object_header_fini(struct lu_object_header *h)
1178 {
1179         LASSERT(cfs_list_empty(&h->loh_layers));
1180         LASSERT(cfs_list_empty(&h->loh_lru));
1181         LASSERT(cfs_hlist_unhashed(&h->loh_hash));
1182         lu_ref_fini(&h->loh_reference);
1183 }
1184 EXPORT_SYMBOL(lu_object_header_fini);
1185
1186 /**
1187  * Given a compound object, find its slice, corresponding to the device type
1188  * \a dtype.
1189  */
1190 struct lu_object *lu_object_locate(struct lu_object_header *h,
1191                                    const struct lu_device_type *dtype)
1192 {
1193         struct lu_object *o;
1194
1195         cfs_list_for_each_entry(o, &h->loh_layers, lo_linkage) {
1196                 if (o->lo_dev->ld_type == dtype)
1197                         return o;
1198         }
1199         return NULL;
1200 }
1201 EXPORT_SYMBOL(lu_object_locate);
1202
1203
1204
1205 /**
1206  * Finalize and free devices in the device stack.
1207  *
1208  * Finalize device stack by purging object cache, and calling
1209  * lu_device_type_operations::ldto_device_fini() and
1210  * lu_device_type_operations::ldto_device_free() on all devices in the stack.
1211  */
1212 void lu_stack_fini(const struct lu_env *env, struct lu_device *top)
1213 {
1214         struct lu_site   *site = top->ld_site;
1215         struct lu_device *scan;
1216         struct lu_device *next;
1217
1218         lu_site_purge(env, site, ~0);
1219         for (scan = top; scan != NULL; scan = next) {
1220                 next = scan->ld_type->ldt_ops->ldto_device_fini(env, scan);
1221                 lu_ref_del(&scan->ld_reference, "lu-stack", &lu_site_init);
1222                 lu_device_put(scan);
1223         }
1224
1225         /* purge again. */
1226         lu_site_purge(env, site, ~0);
1227
1228         for (scan = top; scan != NULL; scan = next) {
1229                 const struct lu_device_type *ldt = scan->ld_type;
1230                 struct obd_type             *type;
1231
1232                 next = ldt->ldt_ops->ldto_device_free(env, scan);
1233                 type = ldt->ldt_obd_type;
1234                 if (type != NULL) {
1235                         type->typ_refcnt--;
1236                         class_put_type(type);
1237                 }
1238         }
1239 }
1240 EXPORT_SYMBOL(lu_stack_fini);
1241
1242 enum {
1243         /**
1244          * Maximal number of tld slots.
1245          */
1246         LU_CONTEXT_KEY_NR = 32
1247 };
1248
1249 static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
1250
1251 static DEFINE_SPINLOCK(lu_keys_guard);
1252
1253 /**
1254  * Global counter incremented whenever key is registered, unregistered,
1255  * revived or quiesced. This is used to void unnecessary calls to
1256  * lu_context_refill(). No locking is provided, as initialization and shutdown
1257  * are supposed to be externally serialized.
1258  */
1259 static unsigned key_set_version = 0;
1260
1261 /**
1262  * Register new key.
1263  */
1264 int lu_context_key_register(struct lu_context_key *key)
1265 {
1266         int result;
1267         int i;
1268
1269         LASSERT(key->lct_init != NULL);
1270         LASSERT(key->lct_fini != NULL);
1271         LASSERT(key->lct_tags != 0);
1272         LASSERT(key->lct_owner != NULL);
1273
1274         result = -ENFILE;
1275         cfs_spin_lock(&lu_keys_guard);
1276         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1277                 if (lu_keys[i] == NULL) {
1278                         key->lct_index = i;
1279                         cfs_atomic_set(&key->lct_used, 1);
1280                         lu_keys[i] = key;
1281                         lu_ref_init(&key->lct_reference);
1282                         result = 0;
1283                         ++key_set_version;
1284                         break;
1285                 }
1286         }
1287         cfs_spin_unlock(&lu_keys_guard);
1288         return result;
1289 }
1290 EXPORT_SYMBOL(lu_context_key_register);
1291
1292 static void key_fini(struct lu_context *ctx, int index)
1293 {
1294         if (ctx->lc_value != NULL && ctx->lc_value[index] != NULL) {
1295                 struct lu_context_key *key;
1296
1297                 key = lu_keys[index];
1298                 LASSERT(key != NULL);
1299                 LASSERT(key->lct_fini != NULL);
1300                 LASSERT(cfs_atomic_read(&key->lct_used) > 1);
1301
1302                 key->lct_fini(ctx, key, ctx->lc_value[index]);
1303                 lu_ref_del(&key->lct_reference, "ctx", ctx);
1304                 cfs_atomic_dec(&key->lct_used);
1305
1306                 LASSERT(key->lct_owner != NULL);
1307                 if ((ctx->lc_tags & LCT_NOREF) == 0) {
1308                         LINVRNT(cfs_module_refcount(key->lct_owner) > 0);
1309                         cfs_module_put(key->lct_owner);
1310                 }
1311                 ctx->lc_value[index] = NULL;
1312         }
1313 }
1314
1315 /**
1316  * Deregister key.
1317  */
1318 void lu_context_key_degister(struct lu_context_key *key)
1319 {
1320         LASSERT(cfs_atomic_read(&key->lct_used) >= 1);
1321         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1322
1323         lu_context_key_quiesce(key);
1324
1325         ++key_set_version;
1326         cfs_spin_lock(&lu_keys_guard);
1327         key_fini(&lu_shrink_env.le_ctx, key->lct_index);
1328         if (lu_keys[key->lct_index]) {
1329                 lu_keys[key->lct_index] = NULL;
1330                 lu_ref_fini(&key->lct_reference);
1331         }
1332         cfs_spin_unlock(&lu_keys_guard);
1333
1334         LASSERTF(cfs_atomic_read(&key->lct_used) == 1,
1335                  "key has instances: %d\n",
1336                  cfs_atomic_read(&key->lct_used));
1337 }
1338 EXPORT_SYMBOL(lu_context_key_degister);
1339
1340 /**
1341  * Register a number of keys. This has to be called after all keys have been
1342  * initialized by a call to LU_CONTEXT_KEY_INIT().
1343  */
1344 int lu_context_key_register_many(struct lu_context_key *k, ...)
1345 {
1346         struct lu_context_key *key = k;
1347         va_list args;
1348         int result;
1349
1350         va_start(args, k);
1351         do {
1352                 result = lu_context_key_register(key);
1353                 if (result)
1354                         break;
1355                 key = va_arg(args, struct lu_context_key *);
1356         } while (key != NULL);
1357         va_end(args);
1358
1359         if (result != 0) {
1360                 va_start(args, k);
1361                 while (k != key) {
1362                         lu_context_key_degister(k);
1363                         k = va_arg(args, struct lu_context_key *);
1364                 }
1365                 va_end(args);
1366         }
1367
1368         return result;
1369 }
1370 EXPORT_SYMBOL(lu_context_key_register_many);
1371
1372 /**
1373  * De-register a number of keys. This is a dual to
1374  * lu_context_key_register_many().
1375  */
1376 void lu_context_key_degister_many(struct lu_context_key *k, ...)
1377 {
1378         va_list args;
1379
1380         va_start(args, k);
1381         do {
1382                 lu_context_key_degister(k);
1383                 k = va_arg(args, struct lu_context_key*);
1384         } while (k != NULL);
1385         va_end(args);
1386 }
1387 EXPORT_SYMBOL(lu_context_key_degister_many);
1388
1389 /**
1390  * Revive a number of keys.
1391  */
1392 void lu_context_key_revive_many(struct lu_context_key *k, ...)
1393 {
1394         va_list args;
1395
1396         va_start(args, k);
1397         do {
1398                 lu_context_key_revive(k);
1399                 k = va_arg(args, struct lu_context_key*);
1400         } while (k != NULL);
1401         va_end(args);
1402 }
1403 EXPORT_SYMBOL(lu_context_key_revive_many);
1404
1405 /**
1406  * Quiescent a number of keys.
1407  */
1408 void lu_context_key_quiesce_many(struct lu_context_key *k, ...)
1409 {
1410         va_list args;
1411
1412         va_start(args, k);
1413         do {
1414                 lu_context_key_quiesce(k);
1415                 k = va_arg(args, struct lu_context_key*);
1416         } while (k != NULL);
1417         va_end(args);
1418 }
1419 EXPORT_SYMBOL(lu_context_key_quiesce_many);
1420
1421 /**
1422  * Return value associated with key \a key in context \a ctx.
1423  */
1424 void *lu_context_key_get(const struct lu_context *ctx,
1425                          const struct lu_context_key *key)
1426 {
1427         LINVRNT(ctx->lc_state == LCS_ENTERED);
1428         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1429         LASSERT(lu_keys[key->lct_index] == key);
1430         return ctx->lc_value[key->lct_index];
1431 }
1432 EXPORT_SYMBOL(lu_context_key_get);
1433
1434 /**
1435  * List of remembered contexts. XXX document me.
1436  */
1437 static CFS_LIST_HEAD(lu_context_remembered);
1438
1439 /**
1440  * Destroy \a key in all remembered contexts. This is used to destroy key
1441  * values in "shared" contexts (like service threads), when a module owning
1442  * the key is about to be unloaded.
1443  */
1444 void lu_context_key_quiesce(struct lu_context_key *key)
1445 {
1446         struct lu_context *ctx;
1447         extern unsigned cl_env_cache_purge(unsigned nr);
1448
1449         if (!(key->lct_tags & LCT_QUIESCENT)) {
1450                 /*
1451                  * XXX layering violation.
1452                  */
1453                 cl_env_cache_purge(~0);
1454                 key->lct_tags |= LCT_QUIESCENT;
1455                 /*
1456                  * XXX memory barrier has to go here.
1457                  */
1458                 cfs_spin_lock(&lu_keys_guard);
1459                 cfs_list_for_each_entry(ctx, &lu_context_remembered,
1460                                         lc_remember)
1461                         key_fini(ctx, key->lct_index);
1462                 cfs_spin_unlock(&lu_keys_guard);
1463                 ++key_set_version;
1464         }
1465 }
1466 EXPORT_SYMBOL(lu_context_key_quiesce);
1467
1468 void lu_context_key_revive(struct lu_context_key *key)
1469 {
1470         key->lct_tags &= ~LCT_QUIESCENT;
1471         ++key_set_version;
1472 }
1473 EXPORT_SYMBOL(lu_context_key_revive);
1474
1475 static void keys_fini(struct lu_context *ctx)
1476 {
1477         int     i;
1478
1479         if (ctx->lc_value == NULL)
1480                 return;
1481
1482         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
1483                 key_fini(ctx, i);
1484
1485         OBD_FREE(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1486         ctx->lc_value = NULL;
1487 }
1488
1489 static int keys_fill(struct lu_context *ctx)
1490 {
1491         int i;
1492
1493         LINVRNT(ctx->lc_value != NULL);
1494         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1495                 struct lu_context_key *key;
1496
1497                 key = lu_keys[i];
1498                 if (ctx->lc_value[i] == NULL && key != NULL &&
1499                     (key->lct_tags & ctx->lc_tags) &&
1500                     /*
1501                      * Don't create values for a LCT_QUIESCENT key, as this
1502                      * will pin module owning a key.
1503                      */
1504                     !(key->lct_tags & LCT_QUIESCENT)) {
1505                         void *value;
1506
1507                         LINVRNT(key->lct_init != NULL);
1508                         LINVRNT(key->lct_index == i);
1509
1510                         value = key->lct_init(ctx, key);
1511                         if (unlikely(IS_ERR(value)))
1512                                 return PTR_ERR(value);
1513
1514                         LASSERT(key->lct_owner != NULL);
1515                         if (!(ctx->lc_tags & LCT_NOREF))
1516                                 cfs_try_module_get(key->lct_owner);
1517                         lu_ref_add_atomic(&key->lct_reference, "ctx", ctx);
1518                         cfs_atomic_inc(&key->lct_used);
1519                         /*
1520                          * This is the only place in the code, where an
1521                          * element of ctx->lc_value[] array is set to non-NULL
1522                          * value.
1523                          */
1524                         ctx->lc_value[i] = value;
1525                         if (key->lct_exit != NULL)
1526                                 ctx->lc_tags |= LCT_HAS_EXIT;
1527                 }
1528                 ctx->lc_version = key_set_version;
1529         }
1530         return 0;
1531 }
1532
1533 static int keys_init(struct lu_context *ctx)
1534 {
1535         OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1536         if (likely(ctx->lc_value != NULL))
1537                 return keys_fill(ctx);
1538
1539         return -ENOMEM;
1540 }
1541
1542 /**
1543  * Initialize context data-structure. Create values for all keys.
1544  */
1545 int lu_context_init(struct lu_context *ctx, __u32 tags)
1546 {
1547         int     rc;
1548
1549         memset(ctx, 0, sizeof *ctx);
1550         ctx->lc_state = LCS_INITIALIZED;
1551         ctx->lc_tags = tags;
1552         if (tags & LCT_REMEMBER) {
1553                 cfs_spin_lock(&lu_keys_guard);
1554                 cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
1555                 cfs_spin_unlock(&lu_keys_guard);
1556         } else {
1557                 CFS_INIT_LIST_HEAD(&ctx->lc_remember);
1558         }
1559
1560         rc = keys_init(ctx);
1561         if (rc != 0)
1562                 lu_context_fini(ctx);
1563
1564         return rc;
1565 }
1566 EXPORT_SYMBOL(lu_context_init);
1567
1568 /**
1569  * Finalize context data-structure. Destroy key values.
1570  */
1571 void lu_context_fini(struct lu_context *ctx)
1572 {
1573         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1574         ctx->lc_state = LCS_FINALIZED;
1575
1576         if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
1577                 LASSERT(cfs_list_empty(&ctx->lc_remember));
1578                 keys_fini(ctx);
1579
1580         } else { /* could race with key degister */
1581                 cfs_spin_lock(&lu_keys_guard);
1582                 keys_fini(ctx);
1583                 cfs_list_del_init(&ctx->lc_remember);
1584                 cfs_spin_unlock(&lu_keys_guard);
1585         }
1586 }
1587 EXPORT_SYMBOL(lu_context_fini);
1588
1589 /**
1590  * Called before entering context.
1591  */
1592 void lu_context_enter(struct lu_context *ctx)
1593 {
1594         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1595         ctx->lc_state = LCS_ENTERED;
1596 }
1597 EXPORT_SYMBOL(lu_context_enter);
1598
1599 /**
1600  * Called after exiting from \a ctx
1601  */
1602 void lu_context_exit(struct lu_context *ctx)
1603 {
1604         int i;
1605
1606         LINVRNT(ctx->lc_state == LCS_ENTERED);
1607         ctx->lc_state = LCS_LEFT;
1608         if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) {
1609                 for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1610                         if (ctx->lc_value[i] != NULL) {
1611                                 struct lu_context_key *key;
1612
1613                                 key = lu_keys[i];
1614                                 LASSERT(key != NULL);
1615                                 if (key->lct_exit != NULL)
1616                                         key->lct_exit(ctx,
1617                                                       key, ctx->lc_value[i]);
1618                         }
1619                 }
1620         }
1621 }
1622 EXPORT_SYMBOL(lu_context_exit);
1623
1624 /**
1625  * Allocate for context all missing keys that were registered after context
1626  * creation. key_set_version is only changed in rare cases when modules
1627  * are loaded and removed.
1628  */
1629 int lu_context_refill(struct lu_context *ctx)
1630 {
1631         return likely(ctx->lc_version == key_set_version) ? 0 : keys_fill(ctx);
1632 }
1633 EXPORT_SYMBOL(lu_context_refill);
1634
1635 /**
1636  * lu_ctx_tags/lu_ses_tags will be updated if there are new types of
1637  * obd being added. Currently, this is only used on client side, specifically
1638  * for echo device client, for other stack (like ptlrpc threads), context are
1639  * predefined when the lu_device type are registered, during the module probe
1640  * phase.
1641  */
1642 __u32 lu_context_tags_default = 0;
1643 __u32 lu_session_tags_default = 0;
1644
1645 void lu_context_tags_update(__u32 tags)
1646 {
1647         cfs_spin_lock(&lu_keys_guard);
1648         lu_context_tags_default |= tags;
1649         key_set_version ++;
1650         cfs_spin_unlock(&lu_keys_guard);
1651 }
1652 EXPORT_SYMBOL(lu_context_tags_update);
1653
1654 void lu_context_tags_clear(__u32 tags)
1655 {
1656         cfs_spin_lock(&lu_keys_guard);
1657         lu_context_tags_default &= ~tags;
1658         key_set_version ++;
1659         cfs_spin_unlock(&lu_keys_guard);
1660 }
1661 EXPORT_SYMBOL(lu_context_tags_clear);
1662
1663 void lu_session_tags_update(__u32 tags)
1664 {
1665         cfs_spin_lock(&lu_keys_guard);
1666         lu_session_tags_default |= tags;
1667         key_set_version ++;
1668         cfs_spin_unlock(&lu_keys_guard);
1669 }
1670 EXPORT_SYMBOL(lu_session_tags_update);
1671
1672 void lu_session_tags_clear(__u32 tags)
1673 {
1674         cfs_spin_lock(&lu_keys_guard);
1675         lu_session_tags_default &= ~tags;
1676         key_set_version ++;
1677         cfs_spin_unlock(&lu_keys_guard);
1678 }
1679 EXPORT_SYMBOL(lu_session_tags_clear);
1680
1681 int lu_env_init(struct lu_env *env, __u32 tags)
1682 {
1683         int result;
1684
1685         env->le_ses = NULL;
1686         result = lu_context_init(&env->le_ctx, tags);
1687         if (likely(result == 0))
1688                 lu_context_enter(&env->le_ctx);
1689         return result;
1690 }
1691 EXPORT_SYMBOL(lu_env_init);
1692
1693 void lu_env_fini(struct lu_env *env)
1694 {
1695         lu_context_exit(&env->le_ctx);
1696         lu_context_fini(&env->le_ctx);
1697         env->le_ses = NULL;
1698 }
1699 EXPORT_SYMBOL(lu_env_fini);
1700
1701 int lu_env_refill(struct lu_env *env)
1702 {
1703         int result;
1704
1705         result = lu_context_refill(&env->le_ctx);
1706         if (result == 0 && env->le_ses != NULL)
1707                 result = lu_context_refill(env->le_ses);
1708         return result;
1709 }
1710 EXPORT_SYMBOL(lu_env_refill);
1711
1712 /**
1713  * Currently, this API will only be used by echo client.
1714  * Because echo client and normal lustre client will share
1715  * same cl_env cache. So echo client needs to refresh
1716  * the env context after it get one from the cache, especially
1717  * when normal client and echo client co-exist in the same client.
1718  */
1719 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags,
1720                           __u32 stags)
1721 {
1722         int    result;
1723
1724         if ((env->le_ctx.lc_tags & ctags) != ctags) {
1725                 env->le_ctx.lc_version = 0;
1726                 env->le_ctx.lc_tags |= ctags;
1727         }
1728
1729         if (env->le_ses && (env->le_ses->lc_tags & stags) != stags) {
1730                 env->le_ses->lc_version = 0;
1731                 env->le_ses->lc_tags |= stags;
1732         }
1733
1734         result = lu_env_refill(env);
1735
1736         return result;
1737 }
1738 EXPORT_SYMBOL(lu_env_refill_by_tags);
1739
1740 static struct cfs_shrinker *lu_site_shrinker = NULL;
1741
1742 typedef struct lu_site_stats{
1743         unsigned        lss_populated;
1744         unsigned        lss_max_search;
1745         unsigned        lss_total;
1746         unsigned        lss_busy;
1747 } lu_site_stats_t;
1748
1749 static void lu_site_stats_get(cfs_hash_t *hs,
1750                               lu_site_stats_t *stats, int populated)
1751 {
1752         cfs_hash_bd_t bd;
1753         int           i;
1754
1755         cfs_hash_for_each_bucket(hs, &bd, i) {
1756                 struct lu_site_bkt_data *bkt = cfs_hash_bd_extra_get(hs, &bd);
1757                 cfs_hlist_head_t        *hhead;
1758
1759                 cfs_hash_bd_lock(hs, &bd, 1);
1760                 stats->lss_busy  += bkt->lsb_busy;
1761                 stats->lss_total += cfs_hash_bd_count_get(&bd);
1762                 stats->lss_max_search = max((int)stats->lss_max_search,
1763                                             cfs_hash_bd_depmax_get(&bd));
1764                 if (!populated) {
1765                         cfs_hash_bd_unlock(hs, &bd, 1);
1766                         continue;
1767                 }
1768
1769                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
1770                         if (!cfs_hlist_empty(hhead))
1771                                 stats->lss_populated++;
1772                 }
1773                 cfs_hash_bd_unlock(hs, &bd, 1);
1774         }
1775 }
1776
1777 #ifdef __KERNEL__
1778
1779 static int lu_cache_shrink(SHRINKER_ARGS(sc, nr_to_scan, gfp_mask))
1780 {
1781         lu_site_stats_t stats;
1782         struct lu_site *s;
1783         struct lu_site *tmp;
1784         int cached = 0;
1785         int remain = shrink_param(sc, nr_to_scan);
1786         CFS_LIST_HEAD(splice);
1787
1788         if (remain != 0) {
1789                 if (!(shrink_param(sc, gfp_mask) & __GFP_FS))
1790                         return -1;
1791                 CDEBUG(D_INODE, "Shrink %d objects\n", remain);
1792         }
1793
1794         cfs_mutex_lock(&lu_sites_guard);
1795         cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
1796                 if (shrink_param(sc, nr_to_scan) != 0) {
1797                         remain = lu_site_purge(&lu_shrink_env, s, remain);
1798                         /*
1799                          * Move just shrunk site to the tail of site list to
1800                          * assure shrinking fairness.
1801                          */
1802                         cfs_list_move_tail(&s->ls_linkage, &splice);
1803                 }
1804
1805                 memset(&stats, 0, sizeof(stats));
1806                 lu_site_stats_get(s->ls_obj_hash, &stats, 0);
1807                 cached += stats.lss_total - stats.lss_busy;
1808                 if (shrink_param(sc, nr_to_scan) && remain <= 0)
1809                         break;
1810         }
1811         cfs_list_splice(&splice, lu_sites.prev);
1812         cfs_mutex_unlock(&lu_sites_guard);
1813
1814         cached = (cached / 100) * sysctl_vfs_cache_pressure;
1815         if (shrink_param(sc, nr_to_scan) == 0)
1816                 CDEBUG(D_INODE, "%d objects cached\n", cached);
1817         return cached;
1818 }
1819
1820 /*
1821  * Debugging stuff.
1822  */
1823
1824 /**
1825  * Environment to be used in debugger, contains all tags.
1826  */
1827 struct lu_env lu_debugging_env;
1828
1829 /**
1830  * Debugging printer function using printk().
1831  */
1832 int lu_printk_printer(const struct lu_env *env,
1833                       void *unused, const char *format, ...)
1834 {
1835         va_list args;
1836
1837         va_start(args, format);
1838         vprintk(format, args);
1839         va_end(args);
1840         return 0;
1841 }
1842
1843 void lu_debugging_setup(void)
1844 {
1845         lu_env_init(&lu_debugging_env, ~0);
1846 }
1847
1848 void lu_context_keys_dump(void)
1849 {
1850         int i;
1851
1852         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1853                 struct lu_context_key *key;
1854
1855                 key = lu_keys[i];
1856                 if (key != NULL) {
1857                         CERROR("[%d]: %p %x (%p,%p,%p) %d %d \"%s\"@%p\n",
1858                                i, key, key->lct_tags,
1859                                key->lct_init, key->lct_fini, key->lct_exit,
1860                                key->lct_index, cfs_atomic_read(&key->lct_used),
1861                                key->lct_owner ? key->lct_owner->name : "",
1862                                key->lct_owner);
1863                         lu_ref_print(&key->lct_reference);
1864                 }
1865         }
1866 }
1867 EXPORT_SYMBOL(lu_context_keys_dump);
1868 #else  /* !__KERNEL__ */
1869 static int lu_cache_shrink(int nr, unsigned int gfp_mask)
1870 {
1871         return 0;
1872 }
1873 #endif /* __KERNEL__ */
1874
1875 int  cl_global_init(void);
1876 void cl_global_fini(void);
1877 int  lu_ref_global_init(void);
1878 void lu_ref_global_fini(void);
1879
1880 int dt_global_init(void);
1881 void dt_global_fini(void);
1882
1883 int llo_global_init(void);
1884 void llo_global_fini(void);
1885
1886 /**
1887  * Initialization of global lu_* data.
1888  */
1889 int lu_global_init(void)
1890 {
1891         int result;
1892
1893         CDEBUG(D_INFO, "Lustre LU module (%p).\n", &lu_keys);
1894
1895         result = lu_ref_global_init();
1896         if (result != 0)
1897                 return result;
1898
1899         LU_CONTEXT_KEY_INIT(&lu_global_key);
1900         result = lu_context_key_register(&lu_global_key);
1901         if (result != 0)
1902                 return result;
1903         /*
1904          * At this level, we don't know what tags are needed, so allocate them
1905          * conservatively. This should not be too bad, because this
1906          * environment is global.
1907          */
1908         cfs_mutex_lock(&lu_sites_guard);
1909         result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
1910         cfs_mutex_unlock(&lu_sites_guard);
1911         if (result != 0)
1912                 return result;
1913
1914         /*
1915          * seeks estimation: 3 seeks to read a record from oi, one to read
1916          * inode, one for ea. Unfortunately setting this high value results in
1917          * lu_object/inode cache consuming all the memory.
1918          */
1919         lu_site_shrinker = cfs_set_shrinker(CFS_DEFAULT_SEEKS, lu_cache_shrink);
1920         if (lu_site_shrinker == NULL)
1921                 return -ENOMEM;
1922
1923         result = lu_time_global_init();
1924         if (result)
1925                 GOTO(out, result);
1926
1927 #ifdef __KERNEL__
1928         result = dt_global_init();
1929         if (result)
1930                 GOTO(out, result);
1931
1932         result = llo_global_init();
1933         if (result)
1934                 GOTO(out, result);
1935 #endif
1936         result = cl_global_init();
1937 out:
1938
1939         return result;
1940 }
1941
1942 /**
1943  * Dual to lu_global_init().
1944  */
1945 void lu_global_fini(void)
1946 {
1947         cl_global_fini();
1948 #ifdef __KERNEL__
1949         llo_global_fini();
1950         dt_global_fini();
1951 #endif
1952         lu_time_global_fini();
1953         if (lu_site_shrinker != NULL) {
1954                 cfs_remove_shrinker(lu_site_shrinker);
1955                 lu_site_shrinker = NULL;
1956         }
1957
1958         lu_context_key_degister(&lu_global_key);
1959
1960         /*
1961          * Tear shrinker environment down _after_ de-registering
1962          * lu_global_key, because the latter has a value in the former.
1963          */
1964         cfs_mutex_lock(&lu_sites_guard);
1965         lu_env_fini(&lu_shrink_env);
1966         cfs_mutex_unlock(&lu_sites_guard);
1967
1968         lu_ref_global_fini();
1969 }
1970
1971 struct lu_buf LU_BUF_NULL = {
1972         .lb_buf = NULL,
1973         .lb_len = 0
1974 };
1975 EXPORT_SYMBOL(LU_BUF_NULL);
1976
1977 static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx)
1978 {
1979 #ifdef LPROCFS
1980         struct lprocfs_counter ret;
1981
1982         lprocfs_stats_collect(stats, idx, &ret);
1983         return (__u32)ret.lc_count;
1984 #else
1985         return 0;
1986 #endif
1987 }
1988
1989 /**
1990  * Output site statistical counters into a buffer. Suitable for
1991  * lprocfs_rd_*()-style functions.
1992  */
1993 int lu_site_stats_print(const struct lu_site *s, char *page, int count)
1994 {
1995         lu_site_stats_t stats;
1996
1997         memset(&stats, 0, sizeof(stats));
1998         lu_site_stats_get(s->ls_obj_hash, &stats, 1);
1999
2000         return snprintf(page, count, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
2001                         stats.lss_busy,
2002                         stats.lss_total,
2003                         stats.lss_populated,
2004                         CFS_HASH_NHLIST(s->ls_obj_hash),
2005                         stats.lss_max_search,
2006                         ls_stats_read(s->ls_stats, LU_SS_CREATED),
2007                         ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
2008                         ls_stats_read(s->ls_stats, LU_SS_CACHE_MISS),
2009                         ls_stats_read(s->ls_stats, LU_SS_CACHE_RACE),
2010                         ls_stats_read(s->ls_stats, LU_SS_CACHE_DEATH_RACE),
2011                         ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED));
2012 }
2013 EXPORT_SYMBOL(lu_site_stats_print);
2014
2015 const char *lu_time_names[LU_TIME_NR] = {
2016         [LU_TIME_FIND_LOOKUP] = "find_lookup",
2017         [LU_TIME_FIND_ALLOC]  = "find_alloc",
2018         [LU_TIME_FIND_INSERT] = "find_insert"
2019 };
2020 EXPORT_SYMBOL(lu_time_names);
2021
2022 /**
2023  * Helper function to initialize a number of kmem slab caches at once.
2024  */
2025 int lu_kmem_init(struct lu_kmem_descr *caches)
2026 {
2027         int result;
2028         struct lu_kmem_descr *iter = caches;
2029
2030         for (result = 0; iter->ckd_cache != NULL; ++iter) {
2031                 *iter->ckd_cache = cfs_mem_cache_create(iter->ckd_name,
2032                                                         iter->ckd_size,
2033                                                         0, 0);
2034                 if (*iter->ckd_cache == NULL) {
2035                         result = -ENOMEM;
2036                         /* free all previously allocated caches */
2037                         lu_kmem_fini(caches);
2038                         break;
2039                 }
2040         }
2041         return result;
2042 }
2043 EXPORT_SYMBOL(lu_kmem_init);
2044
2045 /**
2046  * Helper function to finalize a number of kmem slab cached at once. Dual to
2047  * lu_kmem_init().
2048  */
2049 void lu_kmem_fini(struct lu_kmem_descr *caches)
2050 {
2051         int rc;
2052
2053         for (; caches->ckd_cache != NULL; ++caches) {
2054                 if (*caches->ckd_cache != NULL) {
2055                         rc = cfs_mem_cache_destroy(*caches->ckd_cache);
2056                         LASSERTF(rc == 0, "couldn't destroy %s slab\n",
2057                                  caches->ckd_name);
2058                         *caches->ckd_cache = NULL;
2059                 }
2060         }
2061 }
2062 EXPORT_SYMBOL(lu_kmem_fini);