lustre/obdclass/lu_object.c

   1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
   2  * vim:expandtab:shiftwidth=8:tabstop=8:
   3  *
   4  * GPL HEADER START
   5  *
   6  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   7  *
   8  * This program is free software; you can redistribute it and/or modify
   9  * it under the terms of the GNU General Public License version 2 only,
  10  * as published by the Free Software Foundation.
  11  *
  12  * This program is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License version 2 for more details (a copy is included
  16  * in the LICENSE file that accompanied this code).
  17  *
  18  * You should have received a copy of the GNU General Public License
  19  * version 2 along with this program; If not, see
  20  * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
  21  *
  22  * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
  23  * CA 95054 USA or visit www.sun.com if you need additional information or
  24  * have any questions.
  25  *
  26  * GPL HEADER END
  27  */
  28 /*
  29  * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
  30  * Use is subject to license terms.
  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 #ifndef EXPORT_SYMTAB
  47 # define EXPORT_SYMTAB
  48 #endif
  49
  50 #include <libcfs/libcfs.h>
  51
  52 #ifdef __KERNEL__
  53 # include <linux/module.h>
  54 #endif
  55
  56 /* hash_long() */
  57 #include <libcfs/libcfs_hash.h>
  58 #include <obd_class.h>
  59 #include <obd_support.h>
  60 #include <lustre_disk.h>
  61 #include <lustre_fid.h>
  62 #include <lu_object.h>
  63 #include <libcfs/list.h>
  64 /* lu_time_global_{init,fini}() */
  65 #include <lu_time.h>
  66
  67 static void lu_object_free(const struct lu_env *env, struct lu_object *o);
  68
  69 /**
  70  * Decrease reference counter on object. If last reference is freed, return
  71  * object to the cache, unless lu_object_is_dying(o) holds. In the latter
  72  * case, free object immediately.
  73  */
  74 void lu_object_put(const struct lu_env *env, struct lu_object *o)
  75 {
  76         struct lu_object_header *top;
  77         struct lu_site          *site;
  78         struct lu_object        *orig;
  79         cfs_hash_bd_t            bd;
  80
  81         top  = o->lo_header;
  82         site = o->lo_dev->ld_site;
  83         orig = o;
  84
  85         cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
  86         if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) {
  87                 if (lu_object_is_dying(top)) {
  88                         struct lu_site_bkt_data *bkt;
  89
  90                         /*
  91                          * somebody may be waiting for this, currently only
  92                          * used for cl_object, see cl_object_put_last().
  93                          */
  94                         bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
  95                         cfs_waitq_broadcast(&bkt->lsb_marche_funebre);
  96                 }
  97                 return;
  98         }
  99
 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                 cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 111                 return;
 112         }
 113
 114         /*
 115          * If object is dying (will not be cached), removed it
 116          * from hash table and LRU.
 117          *
 118          * This is done with hash table and LRU lists locked. As the only
 119          * way to acquire first reference to previously unreferenced
 120          * object is through hash-table lookup (lu_object_find()),
 121          * or LRU scanning (lu_site_purge()), that are done under hash-table
 122          * and LRU lock, no race with concurrent object lookup is possible
 123          * and we can safely destroy object below.
 124          */
 125         cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
 126         cfs_list_del_init(&top->loh_lru);
 127         cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 128         /*
 129          * Object was already removed from hash and lru above, can
 130          * kill it.
 131          */
 132         lu_object_free(env, orig);
 133 }
 134 EXPORT_SYMBOL(lu_object_put);
 135
 136 /**
 137  * Allocate new object.
 138  *
 139  * This follows object creation protocol, described in the comment within
 140  * struct lu_device_operations definition.
 141  */
 142 static struct lu_object *lu_object_alloc(const struct lu_env *env,
 143                                          struct lu_device *dev,
 144                                          const struct lu_fid *f,
 145                                          const struct lu_object_conf *conf)
 146 {
 147         struct lu_object *scan;
 148         struct lu_object *top;
 149         cfs_list_t *layers;
 150         int clean;
 151         int result;
 152         ENTRY;
 153
 154         /*
 155          * Create top-level object slice. This will also create
 156          * lu_object_header.
 157          */
 158         top = dev->ld_ops->ldo_object_alloc(env, NULL, dev);
 159         if (top == NULL)
 160                 RETURN(ERR_PTR(-ENOMEM));
 161         /*
 162          * This is the only place where object fid is assigned. It's constant
 163          * after this point.
 164          */
 165         LASSERT(fid_is_igif(f) || fid_ver(f) == 0);
 166         top->lo_header->loh_fid = *f;
 167         layers = &top->lo_header->loh_layers;
 168         do {
 169                 /*
 170                  * Call ->loo_object_init() repeatedly, until no more new
 171                  * object slices are created.
 172                  */
 173                 clean = 1;
 174                 cfs_list_for_each_entry(scan, layers, lo_linkage) {
 175                         if (scan->lo_flags & LU_OBJECT_ALLOCATED)
 176                                 continue;
 177                         clean = 0;
 178                         scan->lo_header = top->lo_header;
 179                         result = scan->lo_ops->loo_object_init(env, scan, conf);
 180                         if (result != 0) {
 181                                 lu_object_free(env, top);
 182                                 RETURN(ERR_PTR(result));
 183                         }
 184                         scan->lo_flags |= LU_OBJECT_ALLOCATED;
 185                 }
 186         } while (!clean);
 187
 188         cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
 189                 if (scan->lo_ops->loo_object_start != NULL) {
 190                         result = scan->lo_ops->loo_object_start(env, scan);
 191                         if (result != 0) {
 192                                 lu_object_free(env, top);
 193                                 RETURN(ERR_PTR(result));
 194                         }
 195                 }
 196         }
 197
 198         lprocfs_counter_incr(dev->ld_site->ls_stats, LU_SS_CREATED);
 199         RETURN(top);
 200 }
 201
 202 /**
 203  * Free an object.
 204  */
 205 static void lu_object_free(const struct lu_env *env, struct lu_object *o)
 206 {
 207         struct lu_site_bkt_data *bkt;
 208         struct lu_site          *site;
 209         struct lu_object        *scan;
 210         cfs_list_t              *layers;
 211         cfs_list_t               splice;
 212
 213         site   = o->lo_dev->ld_site;
 214         layers = &o->lo_header->loh_layers;
 215         bkt    = lu_site_bkt_from_fid(site, &o->lo_header->loh_fid);
 216         /*
 217          * First call ->loo_object_delete() method to release all resources.
 218          */
 219         cfs_list_for_each_entry_reverse(scan, layers, lo_linkage) {
 220                 if (scan->lo_ops->loo_object_delete != NULL)
 221                         scan->lo_ops->loo_object_delete(env, scan);
 222         }
 223
 224         /*
 225          * Then, splice object layers into stand-alone list, and call
 226          * ->loo_object_free() on all layers to free memory. Splice is
 227          * necessary, because lu_object_header is freed together with the
 228          * top-level slice.
 229          */
 230         CFS_INIT_LIST_HEAD(&splice);
 231         cfs_list_splice_init(layers, &splice);
 232         while (!cfs_list_empty(&splice)) {
 233                 /*
 234                  * Free layers in bottom-to-top order, so that object header
 235                  * lives as long as possible and ->loo_object_free() methods
 236                  * can look at its contents.
 237                  */
 238                 o = container_of0(splice.prev, struct lu_object, lo_linkage);
 239                 cfs_list_del_init(&o->lo_linkage);
 240                 LASSERT(o->lo_ops->loo_object_free != NULL);
 241                 o->lo_ops->loo_object_free(env, o);
 242         }
 243
 244         if (cfs_waitq_active(&bkt->lsb_marche_funebre))
 245                 cfs_waitq_broadcast(&bkt->lsb_marche_funebre);
 246 }
 247
 248 /**
 249  * Free \a nr objects from the cold end of the site LRU list.
 250  */
 251 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
 252 {
 253         struct lu_object_header *h;
 254         struct lu_object_header *temp;
 255         struct lu_site_bkt_data *bkt;
 256         cfs_hash_bd_t            bd;
 257         cfs_hash_bd_t            bd2;
 258         cfs_list_t               dispose;
 259         int                      did_sth;
 260         int                      start;
 261         int                      count;
 262         int                      bnr;
 263         int                      i;
 264
 265         CFS_INIT_LIST_HEAD(&dispose);
 266         /*
 267          * Under LRU list lock, scan LRU list and move unreferenced objects to
 268          * the dispose list, removing them from LRU and hash table.
 269          */
 270         start = s->ls_purge_start;
 271         bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1;
 272  again:
 273         did_sth = 0;
 274         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
 275                 if (i < start)
 276                         continue;
 277                 count = bnr;
 278                 cfs_hash_bd_lock(s->ls_obj_hash, &bd, 1);
 279                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 280
 281                 cfs_list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
 282                         /*
 283                          * Objects are sorted in lru order, and "busy"
 284                          * objects (ones with h->loh_ref > 0) naturally tend to
 285                          * live near hot end that we scan last. Unfortunately,
 286                          * sites usually have small (less then ten) number of
 287                          * busy yet rarely accessed objects (some global
 288                          * objects, accessed directly through pointers,
 289                          * bypassing hash table).
 290                          * Currently algorithm scans them over and over again.
 291                          * Probably we should move busy objects out of LRU,
 292                          * or we can live with that.
 293                          */
 294                         if (cfs_atomic_read(&h->loh_ref) > 0)
 295                                 continue;
 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 = 256
 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|LCT_CL_THREAD,
 383         .lct_init = lu_global_key_init,
 384         .lct_fini = lu_global_key_fini
 385 };
 386
 387 /**
 388  * Printer function emitting messages through libcfs_debug_msg().
 389  */
 390 int lu_cdebug_printer(const struct lu_env *env,
 391                       void *cookie, const char *format, ...)
 392 {
 393         struct lu_cdebug_print_info *info = cookie;
 394         struct lu_cdebug_data       *key;
 395         int used;
 396         int complete;
 397         va_list args;
 398
 399         va_start(args, format);
 400
 401         key = lu_context_key_get(&env->le_ctx, &lu_global_key);
 402         LASSERT(key != NULL);
 403
 404         used = strlen(key->lck_area);
 405         complete = format[strlen(format) - 1] == '\n';
 406         /*
 407          * Append new chunk to the buffer.
 408          */
 409         vsnprintf(key->lck_area + used,
 410                   ARRAY_SIZE(key->lck_area) - used, format, args);
 411         if (complete) {
 412                 if (cfs_cdebug_show(info->lpi_mask, info->lpi_subsys))
 413                         libcfs_debug_msg(NULL, info->lpi_subsys, info->lpi_mask,
 414                                          (char *)info->lpi_file, info->lpi_fn,
 415                                          info->lpi_line, "%s", key->lck_area);
 416                 key->lck_area[0] = 0;
 417         }
 418         va_end(args);
 419         return 0;
 420 }
 421 EXPORT_SYMBOL(lu_cdebug_printer);
 422
 423 /**
 424  * Print object header.
 425  */
 426 void lu_object_header_print(const struct lu_env *env, void *cookie,
 427                             lu_printer_t printer,
 428                             const struct lu_object_header *hdr)
 429 {
 430         (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]",
 431                    hdr, hdr->loh_flags, cfs_atomic_read(&hdr->loh_ref),
 432                    PFID(&hdr->loh_fid),
 433                    cfs_hlist_unhashed(&hdr->loh_hash) ? "" : " hash",
 434                    cfs_list_empty((cfs_list_t *)&hdr->loh_lru) ? \
 435                    "" : " lru",
 436                    hdr->loh_attr & LOHA_EXISTS ? " exist":"");
 437 }
 438 EXPORT_SYMBOL(lu_object_header_print);
 439
 440 /**
 441  * Print human readable representation of the \a o to the \a printer.
 442  */
 443 void lu_object_print(const struct lu_env *env, void *cookie,
 444                      lu_printer_t printer, const struct lu_object *o)
 445 {
 446         static const char ruler[] = "........................................";
 447         struct lu_object_header *top;
 448         int depth;
 449
 450         top = o->lo_header;
 451         lu_object_header_print(env, cookie, printer, top);
 452         (*printer)(env, cookie, "{ \n");
 453         cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 454                 depth = o->lo_depth + 4;
 455
 456                 /*
 457                  * print `.' \a depth times followed by type name and address
 458                  */
 459                 (*printer)(env, cookie, "%*.*s%s@%p", depth, depth, ruler,
 460                            o->lo_dev->ld_type->ldt_name, o);
 461                 if (o->lo_ops->loo_object_print != NULL)
 462                         o->lo_ops->loo_object_print(env, cookie, printer, o);
 463                 (*printer)(env, cookie, "\n");
 464         }
 465         (*printer)(env, cookie, "} header@%p\n", top);
 466 }
 467 EXPORT_SYMBOL(lu_object_print);
 468
 469 /**
 470  * Check object consistency.
 471  */
 472 int lu_object_invariant(const struct lu_object *o)
 473 {
 474         struct lu_object_header *top;
 475
 476         top = o->lo_header;
 477         cfs_list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 478                 if (o->lo_ops->loo_object_invariant != NULL &&
 479                     !o->lo_ops->loo_object_invariant(o))
 480                         return 0;
 481         }
 482         return 1;
 483 }
 484 EXPORT_SYMBOL(lu_object_invariant);
 485
 486 static struct lu_object *htable_lookup(struct lu_site *s,
 487                                        cfs_hash_bd_t *bd,
 488                                        const struct lu_fid *f,
 489                                        cfs_waitlink_t *waiter,
 490                                        __u64 *version)
 491 {
 492         struct lu_site_bkt_data *bkt;
 493         struct lu_object_header *h;
 494         cfs_hlist_node_t        *hnode;
 495         __u64  ver = cfs_hash_bd_version_get(bd);
 496
 497         if (*version == ver)
 498                 return NULL;
 499
 500         *version = ver;
 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                 return lu_object_top(h);
 513         }
 514
 515         /*
 516          * Lookup found an object being destroyed this object cannot be
 517          * returned (to assure that references to dying objects are eventually
 518          * drained), and moreover, lookup has to wait until object is freed.
 519          */
 520         cfs_atomic_dec(&h->loh_ref);
 521
 522         cfs_waitlink_init(waiter);
 523         bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
 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 /**
 544  * Core logic of lu_object_find*() functions.
 545  */
 546 static struct lu_object *lu_object_find_try(const struct lu_env *env,
 547                                             struct lu_device *dev,
 548                                             const struct lu_fid *f,
 549                                             const struct lu_object_conf *conf,
 550                                             cfs_waitlink_t *waiter)
 551 {
 552         struct lu_object      *o;
 553         struct lu_object      *shadow;
 554         struct lu_site        *s;
 555         cfs_hash_t            *hs;
 556         cfs_hash_bd_t          bd;
 557         __u64                  version = 0;
 558
 559         /*
 560          * This uses standard index maintenance protocol:
 561          *
 562          *     - search index under lock, and return object if found;
 563          *     - otherwise, unlock index, allocate new object;
 564          *     - lock index and search again;
 565          *     - if nothing is found (usual case), insert newly created
 566          *       object into index;
 567          *     - otherwise (race: other thread inserted object), free
 568          *       object just allocated.
 569          *     - unlock index;
 570          *     - return object.
 571          *
 572          * If dying object is found during index search, add @waiter to the
 573          * site wait-queue and return ERR_PTR(-EAGAIN).
 574          */
 575         s  = dev->ld_site;
 576         hs = s->ls_obj_hash;
 577         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
 578         o = htable_lookup(s, &bd, f, waiter, &version);
 579         cfs_hash_bd_unlock(hs, &bd, 1);
 580         if (o != NULL)
 581                 return o;
 582
 583         /*
 584          * Allocate new object. This may result in rather complicated
 585          * operations, including fld queries, inode loading, etc.
 586          */
 587         o = lu_object_alloc(env, dev, f, conf);
 588         if (unlikely(IS_ERR(o)))
 589                 return o;
 590
 591         LASSERT(lu_fid_eq(lu_object_fid(o), f));
 592
 593         cfs_hash_bd_lock(hs, &bd, 1);
 594
 595         shadow = htable_lookup(s, &bd, f, waiter, &version);
 596         if (likely(shadow == NULL)) {
 597                 struct lu_site_bkt_data *bkt;
 598
 599                 bkt = cfs_hash_bd_extra_get(hs, &bd);
 600                 cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
 601                 cfs_list_add_tail(&o->lo_header->loh_lru, &bkt->lsb_lru);
 602                 cfs_hash_bd_unlock(hs, &bd, 1);
 603                 return o;
 604         }
 605
 606         lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE);
 607         cfs_hash_bd_unlock(hs, &bd, 1);
 608         lu_object_free(env, o);
 609         return shadow;
 610 }
 611
 612 /**
 613  * Much like lu_object_find(), but top level device of object is specifically
 614  * \a dev rather than top level device of the site. This interface allows
 615  * objects of different "stacking" to be created within the same site.
 616  */
 617 struct lu_object *lu_object_find_at(const struct lu_env *env,
 618                                     struct lu_device *dev,
 619                                     const struct lu_fid *f,
 620                                     const struct lu_object_conf *conf)
 621 {
 622         struct lu_site_bkt_data *bkt;
 623         struct lu_object        *obj;
 624         cfs_waitlink_t           wait;
 625
 626         while (1) {
 627                 obj = lu_object_find_try(env, dev, f, conf, &wait);
 628                 if (obj != ERR_PTR(-EAGAIN))
 629                         return obj;
 630                 /*
 631                  * lu_object_find_try() already added waiter into the
 632                  * wait queue.
 633                  */
 634                 cfs_waitq_wait(&wait, CFS_TASK_UNINT);
 635                 bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f);
 636                 cfs_waitq_del(&bkt->lsb_marche_funebre, &wait);
 637         }
 638 }
 639 EXPORT_SYMBOL(lu_object_find_at);
 640
 641 /**
 642  * Find object with given fid, and return its slice belonging to given device.
 643  */
 644 struct lu_object *lu_object_find_slice(const struct lu_env *env,
 645                                        struct lu_device *dev,
 646                                        const struct lu_fid *f,
 647                                        const struct lu_object_conf *conf)
 648 {
 649         struct lu_object *top;
 650         struct lu_object *obj;
 651
 652         top = lu_object_find(env, dev, f, conf);
 653         if (!IS_ERR(top)) {
 654                 obj = lu_object_locate(top->lo_header, dev->ld_type);
 655                 if (obj == NULL)
 656                         lu_object_put(env, top);
 657         } else
 658                 obj = top;
 659         return obj;
 660 }
 661 EXPORT_SYMBOL(lu_object_find_slice);
 662
 663 /**
 664  * Global list of all device types.
 665  */
 666 static CFS_LIST_HEAD(lu_device_types);
 667
 668 int lu_device_type_init(struct lu_device_type *ldt)
 669 {
 670         int result;
 671
 672         CFS_INIT_LIST_HEAD(&ldt->ldt_linkage);
 673         result = ldt->ldt_ops->ldto_init(ldt);
 674         if (result == 0)
 675                 cfs_list_add(&ldt->ldt_linkage, &lu_device_types);
 676         return result;
 677 }
 678 EXPORT_SYMBOL(lu_device_type_init);
 679
 680 void lu_device_type_fini(struct lu_device_type *ldt)
 681 {
 682         cfs_list_del_init(&ldt->ldt_linkage);
 683         ldt->ldt_ops->ldto_fini(ldt);
 684 }
 685 EXPORT_SYMBOL(lu_device_type_fini);
 686
 687 void lu_types_stop(void)
 688 {
 689         struct lu_device_type *ldt;
 690
 691         cfs_list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
 692                 if (ldt->ldt_device_nr == 0)
 693                         ldt->ldt_ops->ldto_stop(ldt);
 694         }
 695 }
 696 EXPORT_SYMBOL(lu_types_stop);
 697
 698 /**
 699  * Global list of all sites on this node
 700  */
 701 static CFS_LIST_HEAD(lu_sites);
 702 static CFS_DECLARE_MUTEX(lu_sites_guard);
 703
 704 /**
 705  * Global environment used by site shrinker.
 706  */
 707 static struct lu_env lu_shrink_env;
 708
 709 struct lu_site_print_arg {
 710         struct lu_env   *lsp_env;
 711         void            *lsp_cookie;
 712         lu_printer_t     lsp_printer;
 713 };
 714
 715 static int
 716 lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 717                   cfs_hlist_node_t *hnode, void *data)
 718 {
 719         struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data;
 720         struct lu_object_header  *h;
 721
 722         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 723         if (!cfs_list_empty(&h->loh_layers)) {
 724                 const struct lu_object *o;
 725
 726                 o = lu_object_top(h);
 727                 lu_object_print(arg->lsp_env, arg->lsp_cookie,
 728                                 arg->lsp_printer, o);
 729         } else {
 730                 lu_object_header_print(arg->lsp_env, arg->lsp_cookie,
 731                                        arg->lsp_printer, h);
 732         }
 733         return 0;
 734 }
 735
 736 /**
 737  * Print all objects in \a s.
 738  */
 739 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
 740                    lu_printer_t printer)
 741 {
 742         struct lu_site_print_arg arg = {
 743                 .lsp_env     = (struct lu_env *)env,
 744                 .lsp_cookie  = cookie,
 745                 .lsp_printer = printer,
 746         };
 747
 748         cfs_hash_for_each(s->ls_obj_hash, lu_site_obj_print, &arg);
 749 }
 750 EXPORT_SYMBOL(lu_site_print);
 751
 752 enum {
 753         LU_CACHE_PERCENT   = 20,
 754 };
 755
 756 /**
 757  * Return desired hash table order.
 758  */
 759 static int lu_htable_order(void)
 760 {
 761         unsigned long cache_size;
 762         int bits;
 763
 764         /*
 765          * Calculate hash table size, assuming that we want reasonable
 766          * performance when 20% of total memory is occupied by cache of
 767          * lu_objects.
 768          *
 769          * Size of lu_object is (arbitrary) taken as 1K (together with inode).
 770          */
 771         cache_size = cfs_num_physpages;
 772
 773 #if BITS_PER_LONG == 32
 774         /* limit hashtable size for lowmem systems to low RAM */
 775         if (cache_size > 1 << (30 - CFS_PAGE_SHIFT))
 776                 cache_size = 1 << (30 - CFS_PAGE_SHIFT) * 3 / 4;
 777 #endif
 778
 779         cache_size = cache_size / 100 * LU_CACHE_PERCENT *
 780                 (CFS_PAGE_SIZE / 1024);
 781
 782         for (bits = 1; (1 << bits) < cache_size; ++bits) {
 783                 ;
 784         }
 785         return bits;
 786 }
 787
 788 static unsigned lu_obj_hop_hash(cfs_hash_t *hs, void *key, unsigned mask)
 789 {
 790         struct lu_fid  *fid = (struct lu_fid *)key;
 791         unsigned        hash;
 792
 793         hash = (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
 794         hash += fid_hash(fid, hs->hs_bkt_bits) << hs->hs_bkt_bits;
 795         return hash & mask;
 796 }
 797
 798 static void *lu_obj_hop_object(cfs_hlist_node_t *hnode)
 799 {
 800         return cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 801 }
 802
 803 static void *lu_obj_hop_key(cfs_hlist_node_t *hnode)
 804 {
 805         struct lu_object_header *h;
 806
 807         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 808         return &h->loh_fid;
 809 }
 810
 811 static int lu_obj_hop_keycmp(void *key, cfs_hlist_node_t *hnode)
 812 {
 813         struct lu_object_header *h;
 814
 815         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 816         return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
 817 }
 818
 819 static void *lu_obj_hop_get(cfs_hlist_node_t *hnode)
 820 {
 821         struct lu_object_header *h;
 822
 823         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
 824         cfs_atomic_inc(&h->loh_ref);
 825         return h;
 826 }
 827
 828 static void *lu_obj_hop_put_locked(cfs_hlist_node_t *hnode)
 829 {
 830         LBUG(); /* we should never called it */
 831         return NULL;
 832 }
 833
 834 cfs_hash_ops_t lu_site_hash_ops = {
 835         .hs_hash        = lu_obj_hop_hash,
 836         .hs_key         = lu_obj_hop_key,
 837         .hs_keycmp      = lu_obj_hop_keycmp,
 838         .hs_object      = lu_obj_hop_object,
 839         .hs_get         = lu_obj_hop_get,
 840         .hs_put_locked  = lu_obj_hop_put_locked,
 841 };
 842
 843 /**
 844  * Initialize site \a s, with \a d as the top level device.
 845  */
 846 #define LU_SITE_BITS_MIN    10
 847 #define LU_SITE_BITS_MAX    23
 848
 849 int lu_site_init(struct lu_site *s, struct lu_device *top)
 850 {
 851         struct lu_site_bkt_data *bkt;
 852         cfs_hash_bd_t bd;
 853         int bits;
 854         int i;
 855         ENTRY;
 856
 857         memset(s, 0, sizeof *s);
 858         bits = lu_htable_order();
 859         for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX);
 860              bits >= LU_SITE_BITS_MIN; bits--) {
 861                 s->ls_obj_hash = cfs_hash_create("lu_site", bits,
 862                                                  bits, bits - LU_SITE_BITS_MIN,
 863                                                  sizeof(*bkt), 0, 0,
 864                                                  &lu_site_hash_ops,
 865                                                  CFS_HASH_SPIN_BKTLOCK |
 866                                                  CFS_HASH_NO_ITEMREF |
 867                                                  CFS_HASH_DEPTH |
 868                                                  CFS_HASH_ASSERT_EMPTY);
 869                 if (s->ls_obj_hash != NULL)
 870                         break;
 871         }
 872
 873         if (s->ls_obj_hash == NULL) {
 874                 CERROR("failed to create lu_site hash with bits: %d\n", bits);
 875                 return -ENOMEM;
 876         }
 877
 878         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
 879                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 880                 CFS_INIT_LIST_HEAD(&bkt->lsb_lru);
 881                 cfs_waitq_init(&bkt->lsb_marche_funebre);
 882         }
 883
 884         s->ls_stats = lprocfs_alloc_stats(LU_SS_LAST_STAT, 0);
 885         if (s->ls_stats == NULL) {
 886                 cfs_hash_putref(s->ls_obj_hash);
 887                 s->ls_obj_hash = NULL;
 888                 return -ENOMEM;
 889         }
 890
 891         lprocfs_counter_init(s->ls_stats, LU_SS_CREATED,
 892                              0, "created", "created");
 893         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_HIT,
 894                              0, "cache_hit", "cache_hit");
 895         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_MISS,
 896                              0, "cache_miss", "cache_miss");
 897         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_RACE,
 898                              0, "cache_race", "cache_race");
 899         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_DEATH_RACE,
 900                              0, "cache_death_race", "cache_death_race");
 901         lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED,
 902                              0, "lru_purged", "lru_purged");
 903
 904         CFS_INIT_LIST_HEAD(&s->ls_linkage);
 905         s->ls_top_dev = top;
 906         top->ld_site = s;
 907         lu_device_get(top);
 908         lu_ref_add(&top->ld_reference, "site-top", s);
 909
 910         RETURN(0);
 911 }
 912 EXPORT_SYMBOL(lu_site_init);
 913
 914 /**
 915  * Finalize \a s and release its resources.
 916  */
 917 void lu_site_fini(struct lu_site *s)
 918 {
 919         cfs_down(&lu_sites_guard);
 920         cfs_list_del_init(&s->ls_linkage);
 921         cfs_up(&lu_sites_guard);
 922
 923         if (s->ls_obj_hash != NULL) {
 924                 cfs_hash_putref(s->ls_obj_hash);
 925                 s->ls_obj_hash = NULL;
 926         }
 927
 928         if (s->ls_top_dev != NULL) {
 929                 s->ls_top_dev->ld_site = NULL;
 930                 lu_ref_del(&s->ls_top_dev->ld_reference, "site-top", s);
 931                 lu_device_put(s->ls_top_dev);
 932                 s->ls_top_dev = NULL;
 933         }
 934
 935         if (s->ls_stats != NULL)
 936                 lprocfs_free_stats(&s->ls_stats);
 937 }
 938 EXPORT_SYMBOL(lu_site_fini);
 939
 940 /**
 941  * Called when initialization of stack for this site is completed.
 942  */
 943 int lu_site_init_finish(struct lu_site *s)
 944 {
 945         int result;
 946         cfs_down(&lu_sites_guard);
 947         result = lu_context_refill(&lu_shrink_env.le_ctx);
 948         if (result == 0)
 949                 cfs_list_add(&s->ls_linkage, &lu_sites);
 950         cfs_up(&lu_sites_guard);
 951         return result;
 952 }
 953 EXPORT_SYMBOL(lu_site_init_finish);
 954
 955 /**
 956  * Acquire additional reference on device \a d
 957  */
 958 void lu_device_get(struct lu_device *d)
 959 {
 960         cfs_atomic_inc(&d->ld_ref);
 961 }
 962 EXPORT_SYMBOL(lu_device_get);
 963
 964 /**
 965  * Release reference on device \a d.
 966  */
 967 void lu_device_put(struct lu_device *d)
 968 {
 969         LASSERT(cfs_atomic_read(&d->ld_ref) > 0);
 970         cfs_atomic_dec(&d->ld_ref);
 971 }
 972 EXPORT_SYMBOL(lu_device_put);
 973
 974 /**
 975  * Initialize device \a d of type \a t.
 976  */
 977 int lu_device_init(struct lu_device *d, struct lu_device_type *t)
 978 {
 979         if (t->ldt_device_nr++ == 0 && t->ldt_ops->ldto_start != NULL)
 980                 t->ldt_ops->ldto_start(t);
 981         memset(d, 0, sizeof *d);
 982         cfs_atomic_set(&d->ld_ref, 0);
 983         d->ld_type = t;
 984         lu_ref_init(&d->ld_reference);
 985         return 0;
 986 }
 987 EXPORT_SYMBOL(lu_device_init);
 988
 989 /**
 990  * Finalize device \a d.
 991  */
 992 void lu_device_fini(struct lu_device *d)
 993 {
 994         struct lu_device_type *t;
 995
 996         t = d->ld_type;
 997         if (d->ld_obd != NULL) {
 998                 d->ld_obd->obd_lu_dev = NULL;
 999                 d->ld_obd = NULL;
1000         }
1001
1002         lu_ref_fini(&d->ld_reference);
1003         LASSERTF(cfs_atomic_read(&d->ld_ref) == 0,
1004                  "Refcount is %u\n", cfs_atomic_read(&d->ld_ref));
1005         LASSERT(t->ldt_device_nr > 0);
1006         if (--t->ldt_device_nr == 0 && t->ldt_ops->ldto_stop != NULL)
1007                 t->ldt_ops->ldto_stop(t);
1008 }
1009 EXPORT_SYMBOL(lu_device_fini);
1010
1011 /**
1012  * Initialize object \a o that is part of compound object \a h and was created
1013  * by device \a d.
1014  */
1015 int lu_object_init(struct lu_object *o,
1016                    struct lu_object_header *h, struct lu_device *d)
1017 {
1018         memset(o, 0, sizeof *o);
1019         o->lo_header = h;
1020         o->lo_dev    = d;
1021         lu_device_get(d);
1022         o->lo_dev_ref = lu_ref_add(&d->ld_reference, "lu_object", o);
1023         CFS_INIT_LIST_HEAD(&o->lo_linkage);
1024         return 0;
1025 }
1026 EXPORT_SYMBOL(lu_object_init);
1027
1028 /**
1029  * Finalize object and release its resources.
1030  */
1031 void lu_object_fini(struct lu_object *o)
1032 {
1033         struct lu_device *dev = o->lo_dev;
1034
1035         LASSERT(cfs_list_empty(&o->lo_linkage));
1036
1037         if (dev != NULL) {
1038                 lu_ref_del_at(&dev->ld_reference,
1039                               o->lo_dev_ref , "lu_object", o);
1040                 lu_device_put(dev);
1041                 o->lo_dev = NULL;
1042         }
1043 }
1044 EXPORT_SYMBOL(lu_object_fini);
1045
1046 /**
1047  * Add object \a o as first layer of compound object \a h
1048  *
1049  * This is typically called by the ->ldo_object_alloc() method of top-level
1050  * device.
1051  */
1052 void lu_object_add_top(struct lu_object_header *h, struct lu_object *o)
1053 {
1054         cfs_list_move(&o->lo_linkage, &h->loh_layers);
1055 }
1056 EXPORT_SYMBOL(lu_object_add_top);
1057
1058 /**
1059  * Add object \a o as a layer of compound object, going after \a before.
1060  *
1061  * This is typically called by the ->ldo_object_alloc() method of \a
1062  * before->lo_dev.
1063  */
1064 void lu_object_add(struct lu_object *before, struct lu_object *o)
1065 {
1066         cfs_list_move(&o->lo_linkage, &before->lo_linkage);
1067 }
1068 EXPORT_SYMBOL(lu_object_add);
1069
1070 /**
1071  * Initialize compound object.
1072  */
1073 int lu_object_header_init(struct lu_object_header *h)
1074 {
1075         memset(h, 0, sizeof *h);
1076         cfs_atomic_set(&h->loh_ref, 1);
1077         CFS_INIT_HLIST_NODE(&h->loh_hash);
1078         CFS_INIT_LIST_HEAD(&h->loh_lru);
1079         CFS_INIT_LIST_HEAD(&h->loh_layers);
1080         lu_ref_init(&h->loh_reference);
1081         return 0;
1082 }
1083 EXPORT_SYMBOL(lu_object_header_init);
1084
1085 /**
1086  * Finalize compound object.
1087  */
1088 void lu_object_header_fini(struct lu_object_header *h)
1089 {
1090         LASSERT(cfs_list_empty(&h->loh_layers));
1091         LASSERT(cfs_list_empty(&h->loh_lru));
1092         LASSERT(cfs_hlist_unhashed(&h->loh_hash));
1093         lu_ref_fini(&h->loh_reference);
1094 }
1095 EXPORT_SYMBOL(lu_object_header_fini);
1096
1097 /**
1098  * Given a compound object, find its slice, corresponding to the device type
1099  * \a dtype.
1100  */
1101 struct lu_object *lu_object_locate(struct lu_object_header *h,
1102                                    const struct lu_device_type *dtype)
1103 {
1104         struct lu_object *o;
1105
1106         cfs_list_for_each_entry(o, &h->loh_layers, lo_linkage) {
1107                 if (o->lo_dev->ld_type == dtype)
1108                         return o;
1109         }
1110         return NULL;
1111 }
1112 EXPORT_SYMBOL(lu_object_locate);
1113
1114
1115
1116 /**
1117  * Finalize and free devices in the device stack.
1118  *
1119  * Finalize device stack by purging object cache, and calling
1120  * lu_device_type_operations::ldto_device_fini() and
1121  * lu_device_type_operations::ldto_device_free() on all devices in the stack.
1122  */
1123 void lu_stack_fini(const struct lu_env *env, struct lu_device *top)
1124 {
1125         struct lu_site   *site = top->ld_site;
1126         struct lu_device *scan;
1127         struct lu_device *next;
1128
1129         lu_site_purge(env, site, ~0);
1130         for (scan = top; scan != NULL; scan = next) {
1131                 next = scan->ld_type->ldt_ops->ldto_device_fini(env, scan);
1132                 lu_ref_del(&scan->ld_reference, "lu-stack", &lu_site_init);
1133                 lu_device_put(scan);
1134         }
1135
1136         /* purge again. */
1137         lu_site_purge(env, site, ~0);
1138
1139         if (!cfs_hash_is_empty(site->ls_obj_hash)) {
1140                 /*
1141                  * Uh-oh, objects still exist.
1142                  */
1143                 static DECLARE_LU_CDEBUG_PRINT_INFO(cookie, D_ERROR);
1144
1145                 lu_site_print(env, site, &cookie, lu_cdebug_printer);
1146         }
1147
1148         for (scan = top; scan != NULL; scan = next) {
1149                 const struct lu_device_type *ldt = scan->ld_type;
1150                 struct obd_type             *type;
1151
1152                 next = ldt->ldt_ops->ldto_device_free(env, scan);
1153                 type = ldt->ldt_obd_type;
1154                 if (type != NULL) {
1155                         type->typ_refcnt--;
1156                         class_put_type(type);
1157                 }
1158         }
1159 }
1160 EXPORT_SYMBOL(lu_stack_fini);
1161
1162 enum {
1163         /**
1164          * Maximal number of tld slots.
1165          */
1166         LU_CONTEXT_KEY_NR = 32
1167 };
1168
1169 static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
1170
1171 static cfs_spinlock_t lu_keys_guard = CFS_SPIN_LOCK_UNLOCKED;
1172
1173 /**
1174  * Global counter incremented whenever key is registered, unregistered,
1175  * revived or quiesced. This is used to void unnecessary calls to
1176  * lu_context_refill(). No locking is provided, as initialization and shutdown
1177  * are supposed to be externally serialized.
1178  */
1179 static unsigned key_set_version = 0;
1180
1181 /**
1182  * Register new key.
1183  */
1184 int lu_context_key_register(struct lu_context_key *key)
1185 {
1186         int result;
1187         int i;
1188
1189         LASSERT(key->lct_init != NULL);
1190         LASSERT(key->lct_fini != NULL);
1191         LASSERT(key->lct_tags != 0);
1192         LASSERT(key->lct_owner != NULL);
1193
1194         result = -ENFILE;
1195         cfs_spin_lock(&lu_keys_guard);
1196         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1197                 if (lu_keys[i] == NULL) {
1198                         key->lct_index = i;
1199                         cfs_atomic_set(&key->lct_used, 1);
1200                         lu_keys[i] = key;
1201                         lu_ref_init(&key->lct_reference);
1202                         result = 0;
1203                         ++key_set_version;
1204                         break;
1205                 }
1206         }
1207         cfs_spin_unlock(&lu_keys_guard);
1208         return result;
1209 }
1210 EXPORT_SYMBOL(lu_context_key_register);
1211
1212 static void key_fini(struct lu_context *ctx, int index)
1213 {
1214         if (ctx->lc_value != NULL && ctx->lc_value[index] != NULL) {
1215                 struct lu_context_key *key;
1216
1217                 key = lu_keys[index];
1218                 LASSERT(key != NULL);
1219                 LASSERT(key->lct_fini != NULL);
1220                 LASSERT(cfs_atomic_read(&key->lct_used) > 1);
1221
1222                 key->lct_fini(ctx, key, ctx->lc_value[index]);
1223                 lu_ref_del(&key->lct_reference, "ctx", ctx);
1224                 cfs_atomic_dec(&key->lct_used);
1225                 LASSERT(key->lct_owner != NULL);
1226                 if (!(ctx->lc_tags & LCT_NOREF)) {
1227                         LASSERT(cfs_module_refcount(key->lct_owner) > 0);
1228                         cfs_module_put(key->lct_owner);
1229                 }
1230                 ctx->lc_value[index] = NULL;
1231         }
1232 }
1233
1234 /**
1235  * Deregister key.
1236  */
1237 void lu_context_key_degister(struct lu_context_key *key)
1238 {
1239         LASSERT(cfs_atomic_read(&key->lct_used) >= 1);
1240         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1241
1242         lu_context_key_quiesce(key);
1243
1244         ++key_set_version;
1245         cfs_spin_lock(&lu_keys_guard);
1246         key_fini(&lu_shrink_env.le_ctx, key->lct_index);
1247         if (lu_keys[key->lct_index]) {
1248                 lu_keys[key->lct_index] = NULL;
1249                 lu_ref_fini(&key->lct_reference);
1250         }
1251         cfs_spin_unlock(&lu_keys_guard);
1252
1253         LASSERTF(cfs_atomic_read(&key->lct_used) == 1,
1254                  "key has instances: %d\n",
1255                  cfs_atomic_read(&key->lct_used));
1256 }
1257 EXPORT_SYMBOL(lu_context_key_degister);
1258
1259 /**
1260  * Register a number of keys. This has to be called after all keys have been
1261  * initialized by a call to LU_CONTEXT_KEY_INIT().
1262  */
1263 int lu_context_key_register_many(struct lu_context_key *k, ...)
1264 {
1265         struct lu_context_key *key = k;
1266         va_list args;
1267         int result;
1268
1269         va_start(args, k);
1270         do {
1271                 result = lu_context_key_register(key);
1272                 if (result)
1273                         break;
1274                 key = va_arg(args, struct lu_context_key *);
1275         } while (key != NULL);
1276         va_end(args);
1277
1278         if (result != 0) {
1279                 va_start(args, k);
1280                 while (k != key) {
1281                         lu_context_key_degister(k);
1282                         k = va_arg(args, struct lu_context_key *);
1283                 }
1284                 va_end(args);
1285         }
1286
1287         return result;
1288 }
1289 EXPORT_SYMBOL(lu_context_key_register_many);
1290
1291 /**
1292  * De-register a number of keys. This is a dual to
1293  * lu_context_key_register_many().
1294  */
1295 void lu_context_key_degister_many(struct lu_context_key *k, ...)
1296 {
1297         va_list args;
1298
1299         va_start(args, k);
1300         do {
1301                 lu_context_key_degister(k);
1302                 k = va_arg(args, struct lu_context_key*);
1303         } while (k != NULL);
1304         va_end(args);
1305 }
1306 EXPORT_SYMBOL(lu_context_key_degister_many);
1307
1308 /**
1309  * Revive a number of keys.
1310  */
1311 void lu_context_key_revive_many(struct lu_context_key *k, ...)
1312 {
1313         va_list args;
1314
1315         va_start(args, k);
1316         do {
1317                 lu_context_key_revive(k);
1318                 k = va_arg(args, struct lu_context_key*);
1319         } while (k != NULL);
1320         va_end(args);
1321 }
1322 EXPORT_SYMBOL(lu_context_key_revive_many);
1323
1324 /**
1325  * Quiescent a number of keys.
1326  */
1327 void lu_context_key_quiesce_many(struct lu_context_key *k, ...)
1328 {
1329         va_list args;
1330
1331         va_start(args, k);
1332         do {
1333                 lu_context_key_quiesce(k);
1334                 k = va_arg(args, struct lu_context_key*);
1335         } while (k != NULL);
1336         va_end(args);
1337 }
1338 EXPORT_SYMBOL(lu_context_key_quiesce_many);
1339
1340 /**
1341  * Return value associated with key \a key in context \a ctx.
1342  */
1343 void *lu_context_key_get(const struct lu_context *ctx,
1344                          const struct lu_context_key *key)
1345 {
1346         LINVRNT(ctx->lc_state == LCS_ENTERED);
1347         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1348         LASSERT(lu_keys[key->lct_index] == key);
1349         return ctx->lc_value[key->lct_index];
1350 }
1351 EXPORT_SYMBOL(lu_context_key_get);
1352
1353 /**
1354  * List of remembered contexts. XXX document me.
1355  */
1356 static CFS_LIST_HEAD(lu_context_remembered);
1357
1358 /**
1359  * Destroy \a key in all remembered contexts. This is used to destroy key
1360  * values in "shared" contexts (like service threads), when a module owning
1361  * the key is about to be unloaded.
1362  */
1363 void lu_context_key_quiesce(struct lu_context_key *key)
1364 {
1365         struct lu_context *ctx;
1366         extern unsigned cl_env_cache_purge(unsigned nr);
1367
1368         if (!(key->lct_tags & LCT_QUIESCENT)) {
1369                 /*
1370                  * XXX layering violation.
1371                  */
1372                 cl_env_cache_purge(~0);
1373                 key->lct_tags |= LCT_QUIESCENT;
1374                 /*
1375                  * XXX memory barrier has to go here.
1376                  */
1377                 cfs_spin_lock(&lu_keys_guard);
1378                 cfs_list_for_each_entry(ctx, &lu_context_remembered,
1379                                         lc_remember)
1380                         key_fini(ctx, key->lct_index);
1381                 cfs_spin_unlock(&lu_keys_guard);
1382                 ++key_set_version;
1383         }
1384 }
1385 EXPORT_SYMBOL(lu_context_key_quiesce);
1386
1387 void lu_context_key_revive(struct lu_context_key *key)
1388 {
1389         key->lct_tags &= ~LCT_QUIESCENT;
1390         ++key_set_version;
1391 }
1392 EXPORT_SYMBOL(lu_context_key_revive);
1393
1394 static void keys_fini(struct lu_context *ctx)
1395 {
1396         int i;
1397
1398         cfs_spin_lock(&lu_keys_guard);
1399         if (ctx->lc_value != NULL) {
1400                 for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
1401                         key_fini(ctx, i);
1402                 OBD_FREE(ctx->lc_value,
1403                          ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1404                 ctx->lc_value = NULL;
1405         }
1406         cfs_spin_unlock(&lu_keys_guard);
1407 }
1408
1409 static int keys_fill(struct lu_context *ctx)
1410 {
1411         int i;
1412
1413         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1414                 struct lu_context_key *key;
1415
1416                 key = lu_keys[i];
1417                 if (ctx->lc_value[i] == NULL && key != NULL &&
1418                     (key->lct_tags & ctx->lc_tags) &&
1419                     /*
1420                      * Don't create values for a LCT_QUIESCENT key, as this
1421                      * will pin module owning a key.
1422                      */
1423                     !(key->lct_tags & LCT_QUIESCENT)) {
1424                         void *value;
1425
1426                         LINVRNT(key->lct_init != NULL);
1427                         LINVRNT(key->lct_index == i);
1428
1429                         value = key->lct_init(ctx, key);
1430                         if (unlikely(IS_ERR(value)))
1431                                 return PTR_ERR(value);
1432
1433                         LASSERT(key->lct_owner != NULL);
1434                         if (!(ctx->lc_tags & LCT_NOREF))
1435                                 cfs_try_module_get(key->lct_owner);
1436                         lu_ref_add_atomic(&key->lct_reference, "ctx", ctx);
1437                         cfs_atomic_inc(&key->lct_used);
1438                         /*
1439                          * This is the only place in the code, where an
1440                          * element of ctx->lc_value[] array is set to non-NULL
1441                          * value.
1442                          */
1443                         ctx->lc_value[i] = value;
1444                         if (key->lct_exit != NULL)
1445                                 ctx->lc_tags |= LCT_HAS_EXIT;
1446                 }
1447                 ctx->lc_version = key_set_version;
1448         }
1449         return 0;
1450 }
1451
1452 static int keys_init(struct lu_context *ctx)
1453 {
1454         int result;
1455
1456         OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1457         if (likely(ctx->lc_value != NULL))
1458                 result = keys_fill(ctx);
1459         else
1460                 result = -ENOMEM;
1461
1462         if (result != 0)
1463                 keys_fini(ctx);
1464         return result;
1465 }
1466
1467 /**
1468  * Initialize context data-structure. Create values for all keys.
1469  */
1470 int lu_context_init(struct lu_context *ctx, __u32 tags)
1471 {
1472         memset(ctx, 0, sizeof *ctx);
1473         ctx->lc_state = LCS_INITIALIZED;
1474         ctx->lc_tags = tags;
1475         if (tags & LCT_REMEMBER) {
1476                 cfs_spin_lock(&lu_keys_guard);
1477                 cfs_list_add(&ctx->lc_remember, &lu_context_remembered);
1478                 cfs_spin_unlock(&lu_keys_guard);
1479         } else
1480                 CFS_INIT_LIST_HEAD(&ctx->lc_remember);
1481         return keys_init(ctx);
1482 }
1483 EXPORT_SYMBOL(lu_context_init);
1484
1485 /**
1486  * Finalize context data-structure. Destroy key values.
1487  */
1488 void lu_context_fini(struct lu_context *ctx)
1489 {
1490         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1491         ctx->lc_state = LCS_FINALIZED;
1492         keys_fini(ctx);
1493         cfs_spin_lock(&lu_keys_guard);
1494         cfs_list_del_init(&ctx->lc_remember);
1495         cfs_spin_unlock(&lu_keys_guard);
1496 }
1497 EXPORT_SYMBOL(lu_context_fini);
1498
1499 /**
1500  * Called before entering context.
1501  */
1502 void lu_context_enter(struct lu_context *ctx)
1503 {
1504         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1505         ctx->lc_state = LCS_ENTERED;
1506 }
1507 EXPORT_SYMBOL(lu_context_enter);
1508
1509 /**
1510  * Called after exiting from \a ctx
1511  */
1512 void lu_context_exit(struct lu_context *ctx)
1513 {
1514         int i;
1515
1516         LINVRNT(ctx->lc_state == LCS_ENTERED);
1517         ctx->lc_state = LCS_LEFT;
1518         if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) {
1519                 for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1520                         if (ctx->lc_value[i] != NULL) {
1521                                 struct lu_context_key *key;
1522
1523                                 key = lu_keys[i];
1524                                 LASSERT(key != NULL);
1525                                 if (key->lct_exit != NULL)
1526                                         key->lct_exit(ctx,
1527                                                       key, ctx->lc_value[i]);
1528                         }
1529                 }
1530         }
1531 }
1532 EXPORT_SYMBOL(lu_context_exit);
1533
1534 /**
1535  * Allocate for context all missing keys that were registered after context
1536  * creation.
1537  */
1538 int lu_context_refill(struct lu_context *ctx)
1539 {
1540         LINVRNT(ctx->lc_value != NULL);
1541         return ctx->lc_version == key_set_version ? 0 : keys_fill(ctx);
1542 }
1543 EXPORT_SYMBOL(lu_context_refill);
1544
1545 int lu_env_init(struct lu_env *env, __u32 tags)
1546 {
1547         int result;
1548
1549         env->le_ses = NULL;
1550         result = lu_context_init(&env->le_ctx, tags);
1551         if (likely(result == 0))
1552                 lu_context_enter(&env->le_ctx);
1553         return result;
1554 }
1555 EXPORT_SYMBOL(lu_env_init);
1556
1557 void lu_env_fini(struct lu_env *env)
1558 {
1559         lu_context_exit(&env->le_ctx);
1560         lu_context_fini(&env->le_ctx);
1561         env->le_ses = NULL;
1562 }
1563 EXPORT_SYMBOL(lu_env_fini);
1564
1565 int lu_env_refill(struct lu_env *env)
1566 {
1567         int result;
1568
1569         result = lu_context_refill(&env->le_ctx);
1570         if (result == 0 && env->le_ses != NULL)
1571                 result = lu_context_refill(env->le_ses);
1572         return result;
1573 }
1574 EXPORT_SYMBOL(lu_env_refill);
1575
1576 static struct cfs_shrinker *lu_site_shrinker = NULL;
1577
1578 struct lu_site_stats_result {
1579         unsigned        lss_populated;
1580         unsigned        lss_max_search;
1581         unsigned        lss_total;
1582         unsigned        lss_busy;
1583         cfs_hash_bd_t   lss_bd;
1584 };
1585
1586 static int lu_site_stats_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
1587                              cfs_hlist_node_t *hnode, void *data)
1588 {
1589         struct lu_site_stats_result    *sa = data;
1590         struct lu_object_header        *h;
1591
1592         sa->lss_total++;
1593         h = cfs_hlist_entry(hnode, struct lu_object_header, loh_hash);
1594         if (cfs_atomic_read(&h->loh_ref) > 0)
1595                 sa->lss_busy++;
1596
1597         if (sa->lss_bd.bd_bucket == NULL ||
1598             cfs_hash_bd_compare(&sa->lss_bd, bd) != 0) {
1599                 if (sa->lss_max_search < cfs_hash_bd_depmax_get(bd))
1600                         sa->lss_max_search = cfs_hash_bd_depmax_get(bd);
1601                 sa->lss_populated++;
1602                 sa->lss_bd = *bd;
1603         }
1604         return 0;
1605 }
1606
1607 #ifdef __KERNEL__
1608 static int lu_cache_shrink(int nr, unsigned int gfp_mask)
1609 {
1610         struct lu_site_stats_result stats;
1611         struct lu_site *s;
1612         struct lu_site *tmp;
1613         int cached = 0;
1614         int remain = nr;
1615         CFS_LIST_HEAD(splice);
1616
1617         if (nr != 0) {
1618                 if (!(gfp_mask & __GFP_FS))
1619                         return -1;
1620                 CDEBUG(D_INODE, "Shrink %d objects\n", nr);
1621         }
1622
1623         cfs_down(&lu_sites_guard);
1624         cfs_list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
1625                 if (nr != 0) {
1626                         remain = lu_site_purge(&lu_shrink_env, s, remain);
1627                         /*
1628                          * Move just shrunk site to the tail of site list to
1629                          * assure shrinking fairness.
1630                          */
1631                         cfs_list_move_tail(&s->ls_linkage, &splice);
1632                 }
1633
1634                 memset(&stats, 0, sizeof(stats));
1635                 cfs_hash_for_each(s->ls_obj_hash, lu_site_stats_get, &stats);
1636                 cached += stats.lss_total - stats.lss_busy;
1637                 if (nr && remain <= 0)
1638                         break;
1639         }
1640         cfs_list_splice(&splice, lu_sites.prev);
1641         cfs_up(&lu_sites_guard);
1642
1643         cached = (cached / 100) * sysctl_vfs_cache_pressure;
1644         if (nr == 0)
1645                 CDEBUG(D_INODE, "%d objects cached\n", cached);
1646         return cached;
1647 }
1648
1649 /*
1650  * Debugging stuff.
1651  */
1652
1653 /**
1654  * Environment to be used in debugger, contains all tags.
1655  */
1656 struct lu_env lu_debugging_env;
1657
1658 /**
1659  * Debugging printer function using printk().
1660  */
1661 int lu_printk_printer(const struct lu_env *env,
1662                       void *unused, const char *format, ...)
1663 {
1664         va_list args;
1665
1666         va_start(args, format);
1667         vprintk(format, args);
1668         va_end(args);
1669         return 0;
1670 }
1671
1672 void lu_debugging_setup(void)
1673 {
1674         lu_env_init(&lu_debugging_env, ~0);
1675 }
1676
1677 void lu_context_keys_dump(void)
1678 {
1679         int i;
1680
1681         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1682                 struct lu_context_key *key;
1683
1684                 key = lu_keys[i];
1685                 if (key != NULL) {
1686                         CERROR("[%d]: %p %x (%p,%p,%p) %d %d \"%s\"@%p\n",
1687                                i, key, key->lct_tags,
1688                                key->lct_init, key->lct_fini, key->lct_exit,
1689                                key->lct_index, cfs_atomic_read(&key->lct_used),
1690                                key->lct_owner ? key->lct_owner->name : "",
1691                                key->lct_owner);
1692                         lu_ref_print(&key->lct_reference);
1693                 }
1694         }
1695 }
1696 EXPORT_SYMBOL(lu_context_keys_dump);
1697 #else  /* !__KERNEL__ */
1698 static int lu_cache_shrink(int nr, unsigned int gfp_mask)
1699 {
1700         return 0;
1701 }
1702 #endif /* __KERNEL__ */
1703
1704 int  cl_global_init(void);
1705 void cl_global_fini(void);
1706 int  lu_ref_global_init(void);
1707 void lu_ref_global_fini(void);
1708
1709 int dt_global_init(void);
1710 void dt_global_fini(void);
1711
1712 int llo_global_init(void);
1713 void llo_global_fini(void);
1714
1715 /**
1716  * Initialization of global lu_* data.
1717  */
1718 int lu_global_init(void)
1719 {
1720         int result;
1721
1722         CDEBUG(D_CONSOLE, "Lustre LU module (%p).\n", &lu_keys);
1723
1724         result = lu_ref_global_init();
1725         if (result != 0)
1726                 return result;
1727
1728         LU_CONTEXT_KEY_INIT(&lu_global_key);
1729         result = lu_context_key_register(&lu_global_key);
1730         if (result != 0)
1731                 return result;
1732         /*
1733          * At this level, we don't know what tags are needed, so allocate them
1734          * conservatively. This should not be too bad, because this
1735          * environment is global.
1736          */
1737         cfs_down(&lu_sites_guard);
1738         result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
1739         cfs_up(&lu_sites_guard);
1740         if (result != 0)
1741                 return result;
1742
1743         /*
1744          * seeks estimation: 3 seeks to read a record from oi, one to read
1745          * inode, one for ea. Unfortunately setting this high value results in
1746          * lu_object/inode cache consuming all the memory.
1747          */
1748         lu_site_shrinker = cfs_set_shrinker(CFS_DEFAULT_SEEKS, lu_cache_shrink);
1749         if (lu_site_shrinker == NULL)
1750                 return -ENOMEM;
1751
1752         result = lu_time_global_init();
1753         if (result)
1754                 GOTO(out, result);
1755
1756 #ifdef __KERNEL__
1757         result = dt_global_init();
1758         if (result)
1759                 GOTO(out, result);
1760
1761         result = llo_global_init();
1762         if (result)
1763                 GOTO(out, result);
1764 #endif
1765         result = cl_global_init();
1766 out:
1767
1768         return result;
1769 }
1770
1771 /**
1772  * Dual to lu_global_init().
1773  */
1774 void lu_global_fini(void)
1775 {
1776         cl_global_fini();
1777 #ifdef __KERNEL__
1778         llo_global_fini();
1779         dt_global_fini();
1780 #endif
1781         lu_time_global_fini();
1782         if (lu_site_shrinker != NULL) {
1783                 cfs_remove_shrinker(lu_site_shrinker);
1784                 lu_site_shrinker = NULL;
1785         }
1786
1787         lu_context_key_degister(&lu_global_key);
1788
1789         /*
1790          * Tear shrinker environment down _after_ de-registering
1791          * lu_global_key, because the latter has a value in the former.
1792          */
1793         cfs_down(&lu_sites_guard);
1794         lu_env_fini(&lu_shrink_env);
1795         cfs_up(&lu_sites_guard);
1796
1797         lu_ref_global_fini();
1798 }
1799
1800 struct lu_buf LU_BUF_NULL = {
1801         .lb_buf = NULL,
1802         .lb_len = 0
1803 };
1804 EXPORT_SYMBOL(LU_BUF_NULL);
1805
1806 static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx)
1807 {
1808 #ifdef LPROCFS
1809         struct lprocfs_counter ret;
1810
1811         lprocfs_stats_collect(stats, idx, &ret);
1812         return (__u32)ret.lc_count;
1813 #else
1814         return 0;
1815 #endif
1816 }
1817
1818 /**
1819  * Output site statistical counters into a buffer. Suitable for
1820  * lprocfs_rd_*()-style functions.
1821  */
1822 int lu_site_stats_print(const struct lu_site *s, char *page, int count)
1823 {
1824         struct lu_site_stats_result stats;
1825
1826         memset(&stats, 0, sizeof(stats));
1827         cfs_hash_for_each(s->ls_obj_hash, lu_site_stats_get, &stats);
1828
1829         return snprintf(page, count, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
1830                         stats.lss_busy,
1831                         stats.lss_total,
1832                         stats.lss_populated,
1833                         CFS_HASH_NHLIST(s->ls_obj_hash),
1834                         stats.lss_max_search,
1835                         ls_stats_read(s->ls_stats, LU_SS_CREATED),
1836                         ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
1837                         ls_stats_read(s->ls_stats, LU_SS_CACHE_MISS),
1838                         ls_stats_read(s->ls_stats, LU_SS_CACHE_RACE),
1839                         ls_stats_read(s->ls_stats, LU_SS_CACHE_DEATH_RACE),
1840                         ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED));
1841 }
1842 EXPORT_SYMBOL(lu_site_stats_print);
1843
1844 const char *lu_time_names[LU_TIME_NR] = {
1845         [LU_TIME_FIND_LOOKUP] = "find_lookup",
1846         [LU_TIME_FIND_ALLOC]  = "find_alloc",
1847         [LU_TIME_FIND_INSERT] = "find_insert"
1848 };
1849 EXPORT_SYMBOL(lu_time_names);
1850
1851 /**
1852  * Helper function to initialize a number of kmem slab caches at once.
1853  */
1854 int lu_kmem_init(struct lu_kmem_descr *caches)
1855 {
1856         int result;
1857
1858         for (result = 0; caches->ckd_cache != NULL; ++caches) {
1859                 *caches->ckd_cache = cfs_mem_cache_create(caches->ckd_name,
1860                                                           caches->ckd_size,
1861                                                           0, 0);
1862                 if (*caches->ckd_cache == NULL) {
1863                         result = -ENOMEM;
1864                         break;
1865                 }
1866         }
1867         return result;
1868 }
1869 EXPORT_SYMBOL(lu_kmem_init);
1870
1871 /**
1872  * Helper function to finalize a number of kmem slab cached at once. Dual to
1873  * lu_kmem_init().
1874  */
1875 void lu_kmem_fini(struct lu_kmem_descr *caches)
1876 {
1877         int rc;
1878
1879         for (; caches->ckd_cache != NULL; ++caches) {
1880                 if (*caches->ckd_cache != NULL) {
1881                         rc = cfs_mem_cache_destroy(*caches->ckd_cache);
1882                         LASSERTF(rc == 0, "couldn't destroy %s slab\n",
1883                                  caches->ckd_name);
1884                         *caches->ckd_cache = NULL;
1885                 }
1886         }
1887 }
1888 EXPORT_SYMBOL(lu_kmem_fini);