drivers/staging/lustre/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, Intel Corporation.
  31  */
  32 /*
  33  * This file is part of Lustre, http://www.lustre.org/
  34  * Lustre is a trademark of Sun Microsystems, Inc.
  35  *
  36  * 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 <linux/libcfs/libcfs.h>
  48
  49 # include <linux/module.h>
  50
  51 /* hash_long() */
  52 #include <linux/libcfs/libcfs_hash.h>
  53 #include <obd_class.h>
  54 #include <obd_support.h>
  55 #include <lustre_disk.h>
  56 #include <lustre_fid.h>
  57 #include <lu_object.h>
  58 #include <lu_ref.h>
  59 #include <linux/list.h>
  60
  61 static void lu_object_free(const struct lu_env *env, struct lu_object *o);
  62
  63 /**
  64  * Decrease reference counter on object. If last reference is freed, return
  65  * object to the cache, unless lu_object_is_dying(o) holds. In the latter
  66  * case, free object immediately.
  67  */
  68 void lu_object_put(const struct lu_env *env, struct lu_object *o)
  69 {
  70         struct lu_site_bkt_data *bkt;
  71         struct lu_object_header *top;
  72         struct lu_site    *site;
  73         struct lu_object        *orig;
  74         cfs_hash_bd_t       bd;
  75         const struct lu_fid     *fid;
  76
  77         top  = o->lo_header;
  78         site = o->lo_dev->ld_site;
  79         orig = o;
  80
  81         /*
  82          * till we have full fids-on-OST implemented anonymous objects
  83          * are possible in OSP. such an object isn't listed in the site
  84          * so we should not remove it from the site.
  85          */
  86         fid = lu_object_fid(o);
  87         if (fid_is_zero(fid)) {
  88                 LASSERT(top->loh_hash.next == NULL
  89                         && top->loh_hash.pprev == NULL);
  90                 LASSERT(list_empty(&top->loh_lru));
  91                 if (!atomic_dec_and_test(&top->loh_ref))
  92                         return;
  93                 list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
  94                         if (o->lo_ops->loo_object_release != NULL)
  95                                 o->lo_ops->loo_object_release(env, o);
  96                 }
  97                 lu_object_free(env, orig);
  98                 return;
  99         }
 100
 101         cfs_hash_bd_get(site->ls_obj_hash, &top->loh_fid, &bd);
 102         bkt = cfs_hash_bd_extra_get(site->ls_obj_hash, &bd);
 103
 104         if (!cfs_hash_bd_dec_and_lock(site->ls_obj_hash, &bd, &top->loh_ref)) {
 105                 if (lu_object_is_dying(top)) {
 106
 107                         /*
 108                          * somebody may be waiting for this, currently only
 109                          * used for cl_object, see cl_object_put_last().
 110                          */
 111                         wake_up_all(&bkt->lsb_marche_funebre);
 112                 }
 113                 return;
 114         }
 115
 116         LASSERT(bkt->lsb_busy > 0);
 117         bkt->lsb_busy--;
 118         /*
 119          * When last reference is released, iterate over object
 120          * layers, and notify them that object is no longer busy.
 121          */
 122         list_for_each_entry_reverse(o, &top->loh_layers, lo_linkage) {
 123                 if (o->lo_ops->loo_object_release != NULL)
 124                         o->lo_ops->loo_object_release(env, o);
 125         }
 126
 127         if (!lu_object_is_dying(top)) {
 128                 LASSERT(list_empty(&top->loh_lru));
 129                 list_add_tail(&top->loh_lru, &bkt->lsb_lru);
 130                 cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 131                 return;
 132         }
 133
 134         /*
 135          * If object is dying (will not be cached), removed it
 136          * from hash table and LRU.
 137          *
 138          * This is done with hash table and LRU lists locked. As the only
 139          * way to acquire first reference to previously unreferenced
 140          * object is through hash-table lookup (lu_object_find()),
 141          * or LRU scanning (lu_site_purge()), that are done under hash-table
 142          * and LRU lock, no race with concurrent object lookup is possible
 143          * and we can safely destroy object below.
 144          */
 145         if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags))
 146                 cfs_hash_bd_del_locked(site->ls_obj_hash, &bd, &top->loh_hash);
 147         cfs_hash_bd_unlock(site->ls_obj_hash, &bd, 1);
 148         /*
 149          * Object was already removed from hash and lru above, can
 150          * kill it.
 151          */
 152         lu_object_free(env, orig);
 153 }
 154 EXPORT_SYMBOL(lu_object_put);
 155
 156 /**
 157  * Put object and don't keep in cache. This is temporary solution for
 158  * multi-site objects when its layering is not constant.
 159  */
 160 void lu_object_put_nocache(const struct lu_env *env, struct lu_object *o)
 161 {
 162         set_bit(LU_OBJECT_HEARD_BANSHEE, &o->lo_header->loh_flags);
 163         return lu_object_put(env, o);
 164 }
 165 EXPORT_SYMBOL(lu_object_put_nocache);
 166
 167 /**
 168  * Kill the object and take it out of LRU cache.
 169  * Currently used by client code for layout change.
 170  */
 171 void lu_object_unhash(const struct lu_env *env, struct lu_object *o)
 172 {
 173         struct lu_object_header *top;
 174
 175         top = o->lo_header;
 176         set_bit(LU_OBJECT_HEARD_BANSHEE, &top->loh_flags);
 177         if (!test_and_set_bit(LU_OBJECT_UNHASHED, &top->loh_flags)) {
 178                 cfs_hash_t *obj_hash = o->lo_dev->ld_site->ls_obj_hash;
 179                 cfs_hash_bd_t bd;
 180
 181                 cfs_hash_bd_get_and_lock(obj_hash, &top->loh_fid, &bd, 1);
 182                 list_del_init(&top->loh_lru);
 183                 cfs_hash_bd_del_locked(obj_hash, &bd, &top->loh_hash);
 184                 cfs_hash_bd_unlock(obj_hash, &bd, 1);
 185         }
 186 }
 187 EXPORT_SYMBOL(lu_object_unhash);
 188
 189 /**
 190  * Allocate new object.
 191  *
 192  * This follows object creation protocol, described in the comment within
 193  * struct lu_device_operations definition.
 194  */
 195 static struct lu_object *lu_object_alloc(const struct lu_env *env,
 196                                          struct lu_device *dev,
 197                                          const struct lu_fid *f,
 198                                          const struct lu_object_conf *conf)
 199 {
 200         struct lu_object *scan;
 201         struct lu_object *top;
 202         struct list_head *layers;
 203         int clean;
 204         int result;
 205
 206         /*
 207          * Create top-level object slice. This will also create
 208          * lu_object_header.
 209          */
 210         top = dev->ld_ops->ldo_object_alloc(env, NULL, dev);
 211         if (top == NULL)
 212                 return ERR_PTR(-ENOMEM);
 213         if (IS_ERR(top))
 214                 return top;
 215         /*
 216          * This is the only place where object fid is assigned. It's constant
 217          * after this point.
 218          */
 219         top->lo_header->loh_fid = *f;
 220         layers = &top->lo_header->loh_layers;
 221         do {
 222                 /*
 223                  * Call ->loo_object_init() repeatedly, until no more new
 224                  * object slices are created.
 225                  */
 226                 clean = 1;
 227                 list_for_each_entry(scan, layers, lo_linkage) {
 228                         if (scan->lo_flags & LU_OBJECT_ALLOCATED)
 229                                 continue;
 230                         clean = 0;
 231                         scan->lo_header = top->lo_header;
 232                         result = scan->lo_ops->loo_object_init(env, scan, conf);
 233                         if (result != 0) {
 234                                 lu_object_free(env, top);
 235                                 return ERR_PTR(result);
 236                         }
 237                         scan->lo_flags |= LU_OBJECT_ALLOCATED;
 238                 }
 239         } while (!clean);
 240
 241         list_for_each_entry_reverse(scan, layers, lo_linkage) {
 242                 if (scan->lo_ops->loo_object_start != NULL) {
 243                         result = scan->lo_ops->loo_object_start(env, scan);
 244                         if (result != 0) {
 245                                 lu_object_free(env, top);
 246                                 return ERR_PTR(result);
 247                         }
 248                 }
 249         }
 250
 251         lprocfs_counter_incr(dev->ld_site->ls_stats, LU_SS_CREATED);
 252         return top;
 253 }
 254
 255 /**
 256  * Free an object.
 257  */
 258 static void lu_object_free(const struct lu_env *env, struct lu_object *o)
 259 {
 260         struct lu_site_bkt_data *bkt;
 261         struct lu_site    *site;
 262         struct lu_object        *scan;
 263         struct list_head              *layers;
 264         struct list_head               splice;
 265
 266         site   = o->lo_dev->ld_site;
 267         layers = &o->lo_header->loh_layers;
 268         bkt    = lu_site_bkt_from_fid(site, &o->lo_header->loh_fid);
 269         /*
 270          * First call ->loo_object_delete() method to release all resources.
 271          */
 272         list_for_each_entry_reverse(scan, layers, lo_linkage) {
 273                 if (scan->lo_ops->loo_object_delete != NULL)
 274                         scan->lo_ops->loo_object_delete(env, scan);
 275         }
 276
 277         /*
 278          * Then, splice object layers into stand-alone list, and call
 279          * ->loo_object_free() on all layers to free memory. Splice is
 280          * necessary, because lu_object_header is freed together with the
 281          * top-level slice.
 282          */
 283         INIT_LIST_HEAD(&splice);
 284         list_splice_init(layers, &splice);
 285         while (!list_empty(&splice)) {
 286                 /*
 287                  * Free layers in bottom-to-top order, so that object header
 288                  * lives as long as possible and ->loo_object_free() methods
 289                  * can look at its contents.
 290                  */
 291                 o = container_of0(splice.prev, struct lu_object, lo_linkage);
 292                 list_del_init(&o->lo_linkage);
 293                 LASSERT(o->lo_ops->loo_object_free != NULL);
 294                 o->lo_ops->loo_object_free(env, o);
 295         }
 296
 297         if (waitqueue_active(&bkt->lsb_marche_funebre))
 298                 wake_up_all(&bkt->lsb_marche_funebre);
 299 }
 300
 301 /**
 302  * Free \a nr objects from the cold end of the site LRU list.
 303  */
 304 int lu_site_purge(const struct lu_env *env, struct lu_site *s, int nr)
 305 {
 306         struct lu_object_header *h;
 307         struct lu_object_header *temp;
 308         struct lu_site_bkt_data *bkt;
 309         cfs_hash_bd_t       bd;
 310         cfs_hash_bd_t       bd2;
 311         struct list_head               dispose;
 312         int                   did_sth;
 313         int                   start;
 314         int                   count;
 315         int                   bnr;
 316         int                   i;
 317
 318         if (OBD_FAIL_CHECK(OBD_FAIL_OBD_NO_LRU))
 319                 return 0;
 320
 321         INIT_LIST_HEAD(&dispose);
 322         /*
 323          * Under LRU list lock, scan LRU list and move unreferenced objects to
 324          * the dispose list, removing them from LRU and hash table.
 325          */
 326         start = s->ls_purge_start;
 327         bnr = (nr == ~0) ? -1 : nr / CFS_HASH_NBKT(s->ls_obj_hash) + 1;
 328  again:
 329         did_sth = 0;
 330         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
 331                 if (i < start)
 332                         continue;
 333                 count = bnr;
 334                 cfs_hash_bd_lock(s->ls_obj_hash, &bd, 1);
 335                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
 336
 337                 list_for_each_entry_safe(h, temp, &bkt->lsb_lru, loh_lru) {
 338                         LASSERT(atomic_read(&h->loh_ref) == 0);
 339
 340                         cfs_hash_bd_get(s->ls_obj_hash, &h->loh_fid, &bd2);
 341                         LASSERT(bd.bd_bucket == bd2.bd_bucket);
 342
 343                         cfs_hash_bd_del_locked(s->ls_obj_hash,
 344                                                &bd2, &h->loh_hash);
 345                         list_move(&h->loh_lru, &dispose);
 346                         if (did_sth == 0)
 347                                 did_sth = 1;
 348
 349                         if (nr != ~0 && --nr == 0)
 350                                 break;
 351
 352                         if (count > 0 && --count == 0)
 353                                 break;
 354
 355                 }
 356                 cfs_hash_bd_unlock(s->ls_obj_hash, &bd, 1);
 357                 cond_resched();
 358                 /*
 359                  * Free everything on the dispose list. This is safe against
 360                  * races due to the reasons described in lu_object_put().
 361                  */
 362                 while (!list_empty(&dispose)) {
 363                         h = container_of0(dispose.next,
 364                                           struct lu_object_header, loh_lru);
 365                         list_del_init(&h->loh_lru);
 366                         lu_object_free(env, lu_object_top(h));
 367                         lprocfs_counter_incr(s->ls_stats, LU_SS_LRU_PURGED);
 368                 }
 369
 370                 if (nr == 0)
 371                         break;
 372         }
 373
 374         if (nr != 0 && did_sth && start != 0) {
 375                 start = 0; /* restart from the first bucket */
 376                 goto again;
 377         }
 378         /* race on s->ls_purge_start, but nobody cares */
 379         s->ls_purge_start = i % CFS_HASH_NBKT(s->ls_obj_hash);
 380
 381         return nr;
 382 }
 383 EXPORT_SYMBOL(lu_site_purge);
 384
 385 /*
 386  * Object printing.
 387  *
 388  * Code below has to jump through certain loops to output object description
 389  * into libcfs_debug_msg-based log. The problem is that lu_object_print()
 390  * composes object description from strings that are parts of _lines_ of
 391  * output (i.e., strings that are not terminated by newline). This doesn't fit
 392  * very well into libcfs_debug_msg() interface that assumes that each message
 393  * supplied to it is a self-contained output line.
 394  *
 395  * To work around this, strings are collected in a temporary buffer
 396  * (implemented as a value of lu_cdebug_key key), until terminating newline
 397  * character is detected.
 398  *
 399  */
 400
 401 enum {
 402         /**
 403          * Maximal line size.
 404          *
 405          * XXX overflow is not handled correctly.
 406          */
 407         LU_CDEBUG_LINE = 512
 408 };
 409
 410 struct lu_cdebug_data {
 411         /**
 412          * Temporary buffer.
 413          */
 414         char lck_area[LU_CDEBUG_LINE];
 415 };
 416
 417 /* context key constructor/destructor: lu_global_key_init, lu_global_key_fini */
 418 LU_KEY_INIT_FINI(lu_global, struct lu_cdebug_data);
 419
 420 /**
 421  * Key, holding temporary buffer. This key is registered very early by
 422  * lu_global_init().
 423  */
 424 struct lu_context_key lu_global_key = {
 425         .lct_tags = LCT_MD_THREAD | LCT_DT_THREAD |
 426                     LCT_MG_THREAD | LCT_CL_THREAD,
 427         .lct_init = lu_global_key_init,
 428         .lct_fini = lu_global_key_fini
 429 };
 430
 431 /**
 432  * Printer function emitting messages through libcfs_debug_msg().
 433  */
 434 int lu_cdebug_printer(const struct lu_env *env,
 435                       void *cookie, const char *format, ...)
 436 {
 437         struct libcfs_debug_msg_data *msgdata = cookie;
 438         struct lu_cdebug_data   *key;
 439         int used;
 440         int complete;
 441         va_list args;
 442
 443         va_start(args, format);
 444
 445         key = lu_context_key_get(&env->le_ctx, &lu_global_key);
 446         LASSERT(key != NULL);
 447
 448         used = strlen(key->lck_area);
 449         complete = format[strlen(format) - 1] == '\n';
 450         /*
 451          * Append new chunk to the buffer.
 452          */
 453         vsnprintf(key->lck_area + used,
 454                   ARRAY_SIZE(key->lck_area) - used, format, args);
 455         if (complete) {
 456                 if (cfs_cdebug_show(msgdata->msg_mask, msgdata->msg_subsys))
 457                         libcfs_debug_msg(msgdata, "%s", key->lck_area);
 458                 key->lck_area[0] = 0;
 459         }
 460         va_end(args);
 461         return 0;
 462 }
 463 EXPORT_SYMBOL(lu_cdebug_printer);
 464
 465 /**
 466  * Print object header.
 467  */
 468 void lu_object_header_print(const struct lu_env *env, void *cookie,
 469                             lu_printer_t printer,
 470                             const struct lu_object_header *hdr)
 471 {
 472         (*printer)(env, cookie, "header@%p[%#lx, %d, "DFID"%s%s%s]",
 473                    hdr, hdr->loh_flags, atomic_read(&hdr->loh_ref),
 474                    PFID(&hdr->loh_fid),
 475                    hlist_unhashed(&hdr->loh_hash) ? "" : " hash",
 476                    list_empty((struct list_head *)&hdr->loh_lru) ? \
 477                    "" : " lru",
 478                    hdr->loh_attr & LOHA_EXISTS ? " exist":"");
 479 }
 480 EXPORT_SYMBOL(lu_object_header_print);
 481
 482 /**
 483  * Print human readable representation of the \a o to the \a printer.
 484  */
 485 void lu_object_print(const struct lu_env *env, void *cookie,
 486                      lu_printer_t printer, const struct lu_object *o)
 487 {
 488         static const char ruler[] = "........................................";
 489         struct lu_object_header *top;
 490         int depth;
 491
 492         top = o->lo_header;
 493         lu_object_header_print(env, cookie, printer, top);
 494         (*printer)(env, cookie, "{ \n");
 495         list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 496                 depth = o->lo_depth + 4;
 497
 498                 /*
 499                  * print `.' \a depth times followed by type name and address
 500                  */
 501                 (*printer)(env, cookie, "%*.*s%s@%p", depth, depth, ruler,
 502                            o->lo_dev->ld_type->ldt_name, o);
 503                 if (o->lo_ops->loo_object_print != NULL)
 504                         o->lo_ops->loo_object_print(env, cookie, printer, o);
 505                 (*printer)(env, cookie, "\n");
 506         }
 507         (*printer)(env, cookie, "} header@%p\n", top);
 508 }
 509 EXPORT_SYMBOL(lu_object_print);
 510
 511 /**
 512  * Check object consistency.
 513  */
 514 int lu_object_invariant(const struct lu_object *o)
 515 {
 516         struct lu_object_header *top;
 517
 518         top = o->lo_header;
 519         list_for_each_entry(o, &top->loh_layers, lo_linkage) {
 520                 if (o->lo_ops->loo_object_invariant != NULL &&
 521                     !o->lo_ops->loo_object_invariant(o))
 522                         return 0;
 523         }
 524         return 1;
 525 }
 526 EXPORT_SYMBOL(lu_object_invariant);
 527
 528 static struct lu_object *htable_lookup(struct lu_site *s,
 529                                        cfs_hash_bd_t *bd,
 530                                        const struct lu_fid *f,
 531                                        wait_queue_t *waiter,
 532                                        __u64 *version)
 533 {
 534         struct lu_site_bkt_data *bkt;
 535         struct lu_object_header *h;
 536         struct hlist_node       *hnode;
 537         __u64  ver = cfs_hash_bd_version_get(bd);
 538
 539         if (*version == ver)
 540                 return ERR_PTR(-ENOENT);
 541
 542         *version = ver;
 543         bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, bd);
 544         /* cfs_hash_bd_peek_locked is a somehow "internal" function
 545          * of cfs_hash, it doesn't add refcount on object. */
 546         hnode = cfs_hash_bd_peek_locked(s->ls_obj_hash, bd, (void *)f);
 547         if (hnode == NULL) {
 548                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_MISS);
 549                 return ERR_PTR(-ENOENT);
 550         }
 551
 552         h = container_of0(hnode, struct lu_object_header, loh_hash);
 553         if (likely(!lu_object_is_dying(h))) {
 554                 cfs_hash_get(s->ls_obj_hash, hnode);
 555                 lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_HIT);
 556                 list_del_init(&h->loh_lru);
 557                 return lu_object_top(h);
 558         }
 559
 560         /*
 561          * Lookup found an object being destroyed this object cannot be
 562          * returned (to assure that references to dying objects are eventually
 563          * drained), and moreover, lookup has to wait until object is freed.
 564          */
 565
 566         init_waitqueue_entry_current(waiter);
 567         add_wait_queue(&bkt->lsb_marche_funebre, waiter);
 568         set_current_state(TASK_UNINTERRUPTIBLE);
 569         lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_DEATH_RACE);
 570         return ERR_PTR(-EAGAIN);
 571 }
 572
 573 /**
 574  * Search cache for an object with the fid \a f. If such object is found,
 575  * return it. Otherwise, create new object, insert it into cache and return
 576  * it. In any case, additional reference is acquired on the returned object.
 577  */
 578 struct lu_object *lu_object_find(const struct lu_env *env,
 579                                  struct lu_device *dev, const struct lu_fid *f,
 580                                  const struct lu_object_conf *conf)
 581 {
 582         return lu_object_find_at(env, dev->ld_site->ls_top_dev, f, conf);
 583 }
 584 EXPORT_SYMBOL(lu_object_find);
 585
 586 static struct lu_object *lu_object_new(const struct lu_env *env,
 587                                        struct lu_device *dev,
 588                                        const struct lu_fid *f,
 589                                        const struct lu_object_conf *conf)
 590 {
 591         struct lu_object        *o;
 592         cfs_hash_t            *hs;
 593         cfs_hash_bd_t       bd;
 594         struct lu_site_bkt_data *bkt;
 595
 596         o = lu_object_alloc(env, dev, f, conf);
 597         if (unlikely(IS_ERR(o)))
 598                 return o;
 599
 600         hs = dev->ld_site->ls_obj_hash;
 601         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
 602         bkt = cfs_hash_bd_extra_get(hs, &bd);
 603         cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
 604         bkt->lsb_busy++;
 605         cfs_hash_bd_unlock(hs, &bd, 1);
 606         return o;
 607 }
 608
 609 /**
 610  * Core logic of lu_object_find*() functions.
 611  */
 612 static struct lu_object *lu_object_find_try(const struct lu_env *env,
 613                                             struct lu_device *dev,
 614                                             const struct lu_fid *f,
 615                                             const struct lu_object_conf *conf,
 616                                             wait_queue_t *waiter)
 617 {
 618         struct lu_object      *o;
 619         struct lu_object      *shadow;
 620         struct lu_site  *s;
 621         cfs_hash_t          *hs;
 622         cfs_hash_bd_t     bd;
 623         __u64             version = 0;
 624
 625         /*
 626          * This uses standard index maintenance protocol:
 627          *
 628          *     - search index under lock, and return object if found;
 629          *     - otherwise, unlock index, allocate new object;
 630          *     - lock index and search again;
 631          *     - if nothing is found (usual case), insert newly created
 632          *       object into index;
 633          *     - otherwise (race: other thread inserted object), free
 634          *       object just allocated.
 635          *     - unlock index;
 636          *     - return object.
 637          *
 638          * For "LOC_F_NEW" case, we are sure the object is new established.
 639          * It is unnecessary to perform lookup-alloc-lookup-insert, instead,
 640          * just alloc and insert directly.
 641          *
 642          * If dying object is found during index search, add @waiter to the
 643          * site wait-queue and return ERR_PTR(-EAGAIN).
 644          */
 645         if (conf != NULL && conf->loc_flags & LOC_F_NEW)
 646                 return lu_object_new(env, dev, f, conf);
 647
 648         s  = dev->ld_site;
 649         hs = s->ls_obj_hash;
 650         cfs_hash_bd_get_and_lock(hs, (void *)f, &bd, 1);
 651         o = htable_lookup(s, &bd, f, waiter, &version);
 652         cfs_hash_bd_unlock(hs, &bd, 1);
 653         if (!IS_ERR(o) || PTR_ERR(o) != -ENOENT)
 654                 return o;
 655
 656         /*
 657          * Allocate new object. This may result in rather complicated
 658          * operations, including fld queries, inode loading, etc.
 659          */
 660         o = lu_object_alloc(env, dev, f, conf);
 661         if (unlikely(IS_ERR(o)))
 662                 return o;
 663
 664         LASSERT(lu_fid_eq(lu_object_fid(o), f));
 665
 666         cfs_hash_bd_lock(hs, &bd, 1);
 667
 668         shadow = htable_lookup(s, &bd, f, waiter, &version);
 669         if (likely(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT)) {
 670                 struct lu_site_bkt_data *bkt;
 671
 672                 bkt = cfs_hash_bd_extra_get(hs, &bd);
 673                 cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
 674                 bkt->lsb_busy++;
 675                 cfs_hash_bd_unlock(hs, &bd, 1);
 676                 return o;
 677         }
 678
 679         lprocfs_counter_incr(s->ls_stats, LU_SS_CACHE_RACE);
 680         cfs_hash_bd_unlock(hs, &bd, 1);
 681         lu_object_free(env, o);
 682         return shadow;
 683 }
 684
 685 /**
 686  * Much like lu_object_find(), but top level device of object is specifically
 687  * \a dev rather than top level device of the site. This interface allows
 688  * objects of different "stacking" to be created within the same site.
 689  */
 690 struct lu_object *lu_object_find_at(const struct lu_env *env,
 691                                     struct lu_device *dev,
 692                                     const struct lu_fid *f,
 693                                     const struct lu_object_conf *conf)
 694 {
 695         struct lu_site_bkt_data *bkt;
 696         struct lu_object        *obj;
 697         wait_queue_t       wait;
 698
 699         while (1) {
 700                 obj = lu_object_find_try(env, dev, f, conf, &wait);
 701                 if (obj != ERR_PTR(-EAGAIN))
 702                         return obj;
 703                 /*
 704                  * lu_object_find_try() already added waiter into the
 705                  * wait queue.
 706                  */
 707                 waitq_wait(&wait, TASK_UNINTERRUPTIBLE);
 708                 bkt = lu_site_bkt_from_fid(dev->ld_site, (void *)f);
 709                 remove_wait_queue(&bkt->lsb_marche_funebre, &wait);
 710         }
 711 }
 712 EXPORT_SYMBOL(lu_object_find_at);
 713
 714 /**
 715  * Find object with given fid, and return its slice belonging to given device.
 716  */
 717 struct lu_object *lu_object_find_slice(const struct lu_env *env,
 718                                        struct lu_device *dev,
 719                                        const struct lu_fid *f,
 720                                        const struct lu_object_conf *conf)
 721 {
 722         struct lu_object *top;
 723         struct lu_object *obj;
 724
 725         top = lu_object_find(env, dev, f, conf);
 726         if (!IS_ERR(top)) {
 727                 obj = lu_object_locate(top->lo_header, dev->ld_type);
 728                 if (obj == NULL)
 729                         lu_object_put(env, top);
 730         } else
 731                 obj = top;
 732         return obj;
 733 }
 734 EXPORT_SYMBOL(lu_object_find_slice);
 735
 736 /**
 737  * Global list of all device types.
 738  */
 739 static LIST_HEAD(lu_device_types);
 740
 741 int lu_device_type_init(struct lu_device_type *ldt)
 742 {
 743         int result = 0;
 744
 745         INIT_LIST_HEAD(&ldt->ldt_linkage);
 746         if (ldt->ldt_ops->ldto_init)
 747                 result = ldt->ldt_ops->ldto_init(ldt);
 748         if (result == 0)
 749                 list_add(&ldt->ldt_linkage, &lu_device_types);
 750         return result;
 751 }
 752 EXPORT_SYMBOL(lu_device_type_init);
 753
 754 void lu_device_type_fini(struct lu_device_type *ldt)
 755 {
 756         list_del_init(&ldt->ldt_linkage);
 757         if (ldt->ldt_ops->ldto_fini)
 758                 ldt->ldt_ops->ldto_fini(ldt);
 759 }
 760 EXPORT_SYMBOL(lu_device_type_fini);
 761
 762 void lu_types_stop(void)
 763 {
 764         struct lu_device_type *ldt;
 765
 766         list_for_each_entry(ldt, &lu_device_types, ldt_linkage) {
 767                 if (ldt->ldt_device_nr == 0 && ldt->ldt_ops->ldto_stop)
 768                         ldt->ldt_ops->ldto_stop(ldt);
 769         }
 770 }
 771 EXPORT_SYMBOL(lu_types_stop);
 772
 773 /**
 774  * Global list of all sites on this node
 775  */
 776 static LIST_HEAD(lu_sites);
 777 static DEFINE_MUTEX(lu_sites_guard);
 778
 779 /**
 780  * Global environment used by site shrinker.
 781  */
 782 static struct lu_env lu_shrink_env;
 783
 784 struct lu_site_print_arg {
 785         struct lu_env   *lsp_env;
 786         void        *lsp_cookie;
 787         lu_printer_t     lsp_printer;
 788 };
 789
 790 static int
 791 lu_site_obj_print(cfs_hash_t *hs, cfs_hash_bd_t *bd,
 792                   struct hlist_node *hnode, void *data)
 793 {
 794         struct lu_site_print_arg *arg = (struct lu_site_print_arg *)data;
 795         struct lu_object_header  *h;
 796
 797         h = hlist_entry(hnode, struct lu_object_header, loh_hash);
 798         if (!list_empty(&h->loh_layers)) {
 799                 const struct lu_object *o;
 800
 801                 o = lu_object_top(h);
 802                 lu_object_print(arg->lsp_env, arg->lsp_cookie,
 803                                 arg->lsp_printer, o);
 804         } else {
 805                 lu_object_header_print(arg->lsp_env, arg->lsp_cookie,
 806                                        arg->lsp_printer, h);
 807         }
 808         return 0;
 809 }
 810
 811 /**
 812  * Print all objects in \a s.
 813  */
 814 void lu_site_print(const struct lu_env *env, struct lu_site *s, void *cookie,
 815                    lu_printer_t printer)
 816 {
 817         struct lu_site_print_arg arg = {
 818                 .lsp_env     = (struct lu_env *)env,
 819                 .lsp_cookie  = cookie,
 820                 .lsp_printer = printer,
 821         };
 822
 823         cfs_hash_for_each(s->ls_obj_hash, lu_site_obj_print, &arg);
 824 }
 825 EXPORT_SYMBOL(lu_site_print);
 826
 827 enum {
 828         LU_CACHE_PERCENT_MAX     = 50,
 829         LU_CACHE_PERCENT_DEFAULT = 20
 830 };
 831
 832 static unsigned int lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
 833 CFS_MODULE_PARM(lu_cache_percent, "i", int, 0644,
 834                 "Percentage of memory to be used as lu_object cache");
 835
 836 /**
 837  * Return desired hash table order.
 838  */
 839 static int lu_htable_order(void)
 840 {
 841         unsigned long cache_size;
 842         int bits;
 843
 844         /*
 845          * Calculate hash table size, assuming that we want reasonable
 846          * performance when 20% of total memory is occupied by cache of
 847          * lu_objects.
 848          *
 849          * Size of lu_object is (arbitrary) taken as 1K (together with inode).
 850          */
 851         cache_size = totalram_pages;
 852
 853 #if BITS_PER_LONG == 32
 854         /* limit hashtable size for lowmem systems to low RAM */
 855         if (cache_size > 1 << (30 - PAGE_CACHE_SHIFT))
 856                 cache_size = 1 << (30 - PAGE_CACHE_SHIFT) * 3 / 4;
 857 #endif
 858
 859         /* clear off unreasonable cache setting. */
 860         if (lu_cache_percent == 0 || lu_cache_percent > LU_CACHE_PERCENT_MAX) {
 861                 CWARN("obdclass: invalid lu_cache_percent: %u, it must be in"
 862                       " the range of (0, %u]. Will use default value: %u.\n",
 863                       lu_cache_percent, LU_CACHE_PERCENT_MAX,
 864                       LU_CACHE_PERCENT_DEFAULT);
 865
 866                 lu_cache_percent = LU_CACHE_PERCENT_DEFAULT;
 867         }
 868         cache_size = cache_size / 100 * lu_cache_percent *
 869                 (PAGE_CACHE_SIZE / 1024);
 870
 871         for (bits = 1; (1 << bits) < cache_size; ++bits) {
 872                 ;
 873         }
 874         return bits;
 875 }
 876
 877 static unsigned lu_obj_hop_hash(cfs_hash_t *hs,
 878                                 const void *key, unsigned mask)
 879 {
 880         struct lu_fid  *fid = (struct lu_fid *)key;
 881         __u32      hash;
 882
 883         hash = fid_flatten32(fid);
 884         hash += (hash >> 4) + (hash << 12); /* mixing oid and seq */
 885         hash = cfs_hash_long(hash, hs->hs_bkt_bits);
 886
 887         /* give me another random factor */
 888         hash -= cfs_hash_long((unsigned long)hs, fid_oid(fid) % 11 + 3);
 889
 890         hash <<= hs->hs_cur_bits - hs->hs_bkt_bits;
 891         hash |= (fid_seq(fid) + fid_oid(fid)) & (CFS_HASH_NBKT(hs) - 1);
 892
 893         return hash & mask;
 894 }
 895
 896 static void *lu_obj_hop_object(struct hlist_node *hnode)
 897 {
 898         return hlist_entry(hnode, struct lu_object_header, loh_hash);
 899 }
 900
 901 static void *lu_obj_hop_key(struct hlist_node *hnode)
 902 {
 903         struct lu_object_header *h;
 904
 905         h = hlist_entry(hnode, struct lu_object_header, loh_hash);
 906         return &h->loh_fid;
 907 }
 908
 909 static int lu_obj_hop_keycmp(const void *key, struct hlist_node *hnode)
 910 {
 911         struct lu_object_header *h;
 912
 913         h = hlist_entry(hnode, struct lu_object_header, loh_hash);
 914         return lu_fid_eq(&h->loh_fid, (struct lu_fid *)key);
 915 }
 916
 917 static void lu_obj_hop_get(cfs_hash_t *hs, struct hlist_node *hnode)
 918 {
 919         struct lu_object_header *h;
 920
 921         h = hlist_entry(hnode, struct lu_object_header, loh_hash);
 922         if (atomic_add_return(1, &h->loh_ref) == 1) {
 923                 struct lu_site_bkt_data *bkt;
 924                 cfs_hash_bd_t       bd;
 925
 926                 cfs_hash_bd_get(hs, &h->loh_fid, &bd);
 927                 bkt = cfs_hash_bd_extra_get(hs, &bd);
 928                 bkt->lsb_busy++;
 929         }
 930 }
 931
 932 static void lu_obj_hop_put_locked(cfs_hash_t *hs, struct hlist_node *hnode)
 933 {
 934         LBUG(); /* we should never called it */
 935 }
 936
 937 cfs_hash_ops_t lu_site_hash_ops = {
 938         .hs_hash        = lu_obj_hop_hash,
 939         .hs_key  = lu_obj_hop_key,
 940         .hs_keycmp      = lu_obj_hop_keycmp,
 941         .hs_object      = lu_obj_hop_object,
 942         .hs_get  = lu_obj_hop_get,
 943         .hs_put_locked  = lu_obj_hop_put_locked,
 944 };
 945
 946 void lu_dev_add_linkage(struct lu_site *s, struct lu_device *d)
 947 {
 948         spin_lock(&s->ls_ld_lock);
 949         if (list_empty(&d->ld_linkage))
 950                 list_add(&d->ld_linkage, &s->ls_ld_linkage);
 951         spin_unlock(&s->ls_ld_lock);
 952 }
 953 EXPORT_SYMBOL(lu_dev_add_linkage);
 954
 955 void lu_dev_del_linkage(struct lu_site *s, struct lu_device *d)
 956 {
 957         spin_lock(&s->ls_ld_lock);
 958         list_del_init(&d->ld_linkage);
 959         spin_unlock(&s->ls_ld_lock);
 960 }
 961 EXPORT_SYMBOL(lu_dev_del_linkage);
 962
 963 /**
 964  * Initialize site \a s, with \a d as the top level device.
 965  */
 966 #define LU_SITE_BITS_MIN    12
 967 #define LU_SITE_BITS_MAX    24
 968 /**
 969  * total 256 buckets, we don't want too many buckets because:
 970  * - consume too much memory
 971  * - avoid unbalanced LRU list
 972  */
 973 #define LU_SITE_BKT_BITS    8
 974
 975 int lu_site_init(struct lu_site *s, struct lu_device *top)
 976 {
 977         struct lu_site_bkt_data *bkt;
 978         cfs_hash_bd_t bd;
 979         char name[16];
 980         int bits;
 981         int i;
 982
 983         memset(s, 0, sizeof *s);
 984         bits = lu_htable_order();
 985         snprintf(name, 16, "lu_site_%s", top->ld_type->ldt_name);
 986         for (bits = min(max(LU_SITE_BITS_MIN, bits), LU_SITE_BITS_MAX);
 987              bits >= LU_SITE_BITS_MIN; bits--) {
 988                 s->ls_obj_hash = cfs_hash_create(name, bits, bits,
 989                                                  bits - LU_SITE_BKT_BITS,
 990                                                  sizeof(*bkt), 0, 0,
 991                                                  &lu_site_hash_ops,
 992                                                  CFS_HASH_SPIN_BKTLOCK |
 993                                                  CFS_HASH_NO_ITEMREF |
 994                                                  CFS_HASH_DEPTH |
 995                                                  CFS_HASH_ASSERT_EMPTY);
 996                 if (s->ls_obj_hash != NULL)
 997                         break;
 998         }
 999
1000         if (s->ls_obj_hash == NULL) {
1001                 CERROR("failed to create lu_site hash with bits: %d\n", bits);
1002                 return -ENOMEM;
1003         }
1004
1005         cfs_hash_for_each_bucket(s->ls_obj_hash, &bd, i) {
1006                 bkt = cfs_hash_bd_extra_get(s->ls_obj_hash, &bd);
1007                 INIT_LIST_HEAD(&bkt->lsb_lru);
1008                 init_waitqueue_head(&bkt->lsb_marche_funebre);
1009         }
1010
1011         s->ls_stats = lprocfs_alloc_stats(LU_SS_LAST_STAT, 0);
1012         if (s->ls_stats == NULL) {
1013                 cfs_hash_putref(s->ls_obj_hash);
1014                 s->ls_obj_hash = NULL;
1015                 return -ENOMEM;
1016         }
1017
1018         lprocfs_counter_init(s->ls_stats, LU_SS_CREATED,
1019                              0, "created", "created");
1020         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_HIT,
1021                              0, "cache_hit", "cache_hit");
1022         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_MISS,
1023                              0, "cache_miss", "cache_miss");
1024         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_RACE,
1025                              0, "cache_race", "cache_race");
1026         lprocfs_counter_init(s->ls_stats, LU_SS_CACHE_DEATH_RACE,
1027                              0, "cache_death_race", "cache_death_race");
1028         lprocfs_counter_init(s->ls_stats, LU_SS_LRU_PURGED,
1029                              0, "lru_purged", "lru_purged");
1030
1031         INIT_LIST_HEAD(&s->ls_linkage);
1032         s->ls_top_dev = top;
1033         top->ld_site = s;
1034         lu_device_get(top);
1035         lu_ref_add(&top->ld_reference, "site-top", s);
1036
1037         INIT_LIST_HEAD(&s->ls_ld_linkage);
1038         spin_lock_init(&s->ls_ld_lock);
1039
1040         lu_dev_add_linkage(s, top);
1041
1042         return 0;
1043 }
1044 EXPORT_SYMBOL(lu_site_init);
1045
1046 /**
1047  * Finalize \a s and release its resources.
1048  */
1049 void lu_site_fini(struct lu_site *s)
1050 {
1051         mutex_lock(&lu_sites_guard);
1052         list_del_init(&s->ls_linkage);
1053         mutex_unlock(&lu_sites_guard);
1054
1055         if (s->ls_obj_hash != NULL) {
1056                 cfs_hash_putref(s->ls_obj_hash);
1057                 s->ls_obj_hash = NULL;
1058         }
1059
1060         if (s->ls_top_dev != NULL) {
1061                 s->ls_top_dev->ld_site = NULL;
1062                 lu_ref_del(&s->ls_top_dev->ld_reference, "site-top", s);
1063                 lu_device_put(s->ls_top_dev);
1064                 s->ls_top_dev = NULL;
1065         }
1066
1067         if (s->ls_stats != NULL)
1068                 lprocfs_free_stats(&s->ls_stats);
1069 }
1070 EXPORT_SYMBOL(lu_site_fini);
1071
1072 /**
1073  * Called when initialization of stack for this site is completed.
1074  */
1075 int lu_site_init_finish(struct lu_site *s)
1076 {
1077         int result;
1078         mutex_lock(&lu_sites_guard);
1079         result = lu_context_refill(&lu_shrink_env.le_ctx);
1080         if (result == 0)
1081                 list_add(&s->ls_linkage, &lu_sites);
1082         mutex_unlock(&lu_sites_guard);
1083         return result;
1084 }
1085 EXPORT_SYMBOL(lu_site_init_finish);
1086
1087 /**
1088  * Acquire additional reference on device \a d
1089  */
1090 void lu_device_get(struct lu_device *d)
1091 {
1092         atomic_inc(&d->ld_ref);
1093 }
1094 EXPORT_SYMBOL(lu_device_get);
1095
1096 /**
1097  * Release reference on device \a d.
1098  */
1099 void lu_device_put(struct lu_device *d)
1100 {
1101         LASSERT(atomic_read(&d->ld_ref) > 0);
1102         atomic_dec(&d->ld_ref);
1103 }
1104 EXPORT_SYMBOL(lu_device_put);
1105
1106 /**
1107  * Initialize device \a d of type \a t.
1108  */
1109 int lu_device_init(struct lu_device *d, struct lu_device_type *t)
1110 {
1111         if (t->ldt_device_nr++ == 0 && t->ldt_ops->ldto_start != NULL)
1112                 t->ldt_ops->ldto_start(t);
1113         memset(d, 0, sizeof *d);
1114         atomic_set(&d->ld_ref, 0);
1115         d->ld_type = t;
1116         lu_ref_init(&d->ld_reference);
1117         INIT_LIST_HEAD(&d->ld_linkage);
1118         return 0;
1119 }
1120 EXPORT_SYMBOL(lu_device_init);
1121
1122 /**
1123  * Finalize device \a d.
1124  */
1125 void lu_device_fini(struct lu_device *d)
1126 {
1127         struct lu_device_type *t;
1128
1129         t = d->ld_type;
1130         if (d->ld_obd != NULL) {
1131                 d->ld_obd->obd_lu_dev = NULL;
1132                 d->ld_obd = NULL;
1133         }
1134
1135         lu_ref_fini(&d->ld_reference);
1136         LASSERTF(atomic_read(&d->ld_ref) == 0,
1137                  "Refcount is %u\n", atomic_read(&d->ld_ref));
1138         LASSERT(t->ldt_device_nr > 0);
1139         if (--t->ldt_device_nr == 0 && t->ldt_ops->ldto_stop != NULL)
1140                 t->ldt_ops->ldto_stop(t);
1141 }
1142 EXPORT_SYMBOL(lu_device_fini);
1143
1144 /**
1145  * Initialize object \a o that is part of compound object \a h and was created
1146  * by device \a d.
1147  */
1148 int lu_object_init(struct lu_object *o, struct lu_object_header *h,
1149                    struct lu_device *d)
1150 {
1151         memset(o, 0, sizeof(*o));
1152         o->lo_header = h;
1153         o->lo_dev = d;
1154         lu_device_get(d);
1155         lu_ref_add_at(&d->ld_reference, &o->lo_dev_ref, "lu_object", o);
1156         INIT_LIST_HEAD(&o->lo_linkage);
1157
1158         return 0;
1159 }
1160 EXPORT_SYMBOL(lu_object_init);
1161
1162 /**
1163  * Finalize object and release its resources.
1164  */
1165 void lu_object_fini(struct lu_object *o)
1166 {
1167         struct lu_device *dev = o->lo_dev;
1168
1169         LASSERT(list_empty(&o->lo_linkage));
1170
1171         if (dev != NULL) {
1172                 lu_ref_del_at(&dev->ld_reference, &o->lo_dev_ref,
1173                               "lu_object", o);
1174                 lu_device_put(dev);
1175                 o->lo_dev = NULL;
1176         }
1177 }
1178 EXPORT_SYMBOL(lu_object_fini);
1179
1180 /**
1181  * Add object \a o as first layer of compound object \a h
1182  *
1183  * This is typically called by the ->ldo_object_alloc() method of top-level
1184  * device.
1185  */
1186 void lu_object_add_top(struct lu_object_header *h, struct lu_object *o)
1187 {
1188         list_move(&o->lo_linkage, &h->loh_layers);
1189 }
1190 EXPORT_SYMBOL(lu_object_add_top);
1191
1192 /**
1193  * Add object \a o as a layer of compound object, going after \a before.
1194  *
1195  * This is typically called by the ->ldo_object_alloc() method of \a
1196  * before->lo_dev.
1197  */
1198 void lu_object_add(struct lu_object *before, struct lu_object *o)
1199 {
1200         list_move(&o->lo_linkage, &before->lo_linkage);
1201 }
1202 EXPORT_SYMBOL(lu_object_add);
1203
1204 /**
1205  * Initialize compound object.
1206  */
1207 int lu_object_header_init(struct lu_object_header *h)
1208 {
1209         memset(h, 0, sizeof *h);
1210         atomic_set(&h->loh_ref, 1);
1211         INIT_HLIST_NODE(&h->loh_hash);
1212         INIT_LIST_HEAD(&h->loh_lru);
1213         INIT_LIST_HEAD(&h->loh_layers);
1214         lu_ref_init(&h->loh_reference);
1215         return 0;
1216 }
1217 EXPORT_SYMBOL(lu_object_header_init);
1218
1219 /**
1220  * Finalize compound object.
1221  */
1222 void lu_object_header_fini(struct lu_object_header *h)
1223 {
1224         LASSERT(list_empty(&h->loh_layers));
1225         LASSERT(list_empty(&h->loh_lru));
1226         LASSERT(hlist_unhashed(&h->loh_hash));
1227         lu_ref_fini(&h->loh_reference);
1228 }
1229 EXPORT_SYMBOL(lu_object_header_fini);
1230
1231 /**
1232  * Given a compound object, find its slice, corresponding to the device type
1233  * \a dtype.
1234  */
1235 struct lu_object *lu_object_locate(struct lu_object_header *h,
1236                                    const struct lu_device_type *dtype)
1237 {
1238         struct lu_object *o;
1239
1240         list_for_each_entry(o, &h->loh_layers, lo_linkage) {
1241                 if (o->lo_dev->ld_type == dtype)
1242                         return o;
1243         }
1244         return NULL;
1245 }
1246 EXPORT_SYMBOL(lu_object_locate);
1247
1248
1249
1250 /**
1251  * Finalize and free devices in the device stack.
1252  *
1253  * Finalize device stack by purging object cache, and calling
1254  * lu_device_type_operations::ldto_device_fini() and
1255  * lu_device_type_operations::ldto_device_free() on all devices in the stack.
1256  */
1257 void lu_stack_fini(const struct lu_env *env, struct lu_device *top)
1258 {
1259         struct lu_site   *site = top->ld_site;
1260         struct lu_device *scan;
1261         struct lu_device *next;
1262
1263         lu_site_purge(env, site, ~0);
1264         for (scan = top; scan != NULL; scan = next) {
1265                 next = scan->ld_type->ldt_ops->ldto_device_fini(env, scan);
1266                 lu_ref_del(&scan->ld_reference, "lu-stack", &lu_site_init);
1267                 lu_device_put(scan);
1268         }
1269
1270         /* purge again. */
1271         lu_site_purge(env, site, ~0);
1272
1273         for (scan = top; scan != NULL; scan = next) {
1274                 const struct lu_device_type *ldt = scan->ld_type;
1275                 struct obd_type      *type;
1276
1277                 next = ldt->ldt_ops->ldto_device_free(env, scan);
1278                 type = ldt->ldt_obd_type;
1279                 if (type != NULL) {
1280                         type->typ_refcnt--;
1281                         class_put_type(type);
1282                 }
1283         }
1284 }
1285 EXPORT_SYMBOL(lu_stack_fini);
1286
1287 enum {
1288         /**
1289          * Maximal number of tld slots.
1290          */
1291         LU_CONTEXT_KEY_NR = 40
1292 };
1293
1294 static struct lu_context_key *lu_keys[LU_CONTEXT_KEY_NR] = { NULL, };
1295
1296 static DEFINE_SPINLOCK(lu_keys_guard);
1297
1298 /**
1299  * Global counter incremented whenever key is registered, unregistered,
1300  * revived or quiesced. This is used to void unnecessary calls to
1301  * lu_context_refill(). No locking is provided, as initialization and shutdown
1302  * are supposed to be externally serialized.
1303  */
1304 static unsigned key_set_version = 0;
1305
1306 /**
1307  * Register new key.
1308  */
1309 int lu_context_key_register(struct lu_context_key *key)
1310 {
1311         int result;
1312         int i;
1313
1314         LASSERT(key->lct_init != NULL);
1315         LASSERT(key->lct_fini != NULL);
1316         LASSERT(key->lct_tags != 0);
1317
1318         result = -ENFILE;
1319         spin_lock(&lu_keys_guard);
1320         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1321                 if (lu_keys[i] == NULL) {
1322                         key->lct_index = i;
1323                         atomic_set(&key->lct_used, 1);
1324                         lu_keys[i] = key;
1325                         lu_ref_init(&key->lct_reference);
1326                         result = 0;
1327                         ++key_set_version;
1328                         break;
1329                 }
1330         }
1331         spin_unlock(&lu_keys_guard);
1332         return result;
1333 }
1334 EXPORT_SYMBOL(lu_context_key_register);
1335
1336 static void key_fini(struct lu_context *ctx, int index)
1337 {
1338         if (ctx->lc_value != NULL && ctx->lc_value[index] != NULL) {
1339                 struct lu_context_key *key;
1340
1341                 key = lu_keys[index];
1342                 LASSERT(key != NULL);
1343                 LASSERT(key->lct_fini != NULL);
1344                 LASSERT(atomic_read(&key->lct_used) > 1);
1345
1346                 key->lct_fini(ctx, key, ctx->lc_value[index]);
1347                 lu_ref_del(&key->lct_reference, "ctx", ctx);
1348                 atomic_dec(&key->lct_used);
1349
1350                 if ((ctx->lc_tags & LCT_NOREF) == 0) {
1351 #ifdef CONFIG_MODULE_UNLOAD
1352                         LINVRNT(module_refcount(key->lct_owner) > 0);
1353 #endif
1354                         module_put(key->lct_owner);
1355                 }
1356                 ctx->lc_value[index] = NULL;
1357         }
1358 }
1359
1360 /**
1361  * Deregister key.
1362  */
1363 void lu_context_key_degister(struct lu_context_key *key)
1364 {
1365         LASSERT(atomic_read(&key->lct_used) >= 1);
1366         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1367
1368         lu_context_key_quiesce(key);
1369
1370         ++key_set_version;
1371         spin_lock(&lu_keys_guard);
1372         key_fini(&lu_shrink_env.le_ctx, key->lct_index);
1373         if (lu_keys[key->lct_index]) {
1374                 lu_keys[key->lct_index] = NULL;
1375                 lu_ref_fini(&key->lct_reference);
1376         }
1377         spin_unlock(&lu_keys_guard);
1378
1379         LASSERTF(atomic_read(&key->lct_used) == 1,
1380                  "key has instances: %d\n",
1381                  atomic_read(&key->lct_used));
1382 }
1383 EXPORT_SYMBOL(lu_context_key_degister);
1384
1385 /**
1386  * Register a number of keys. This has to be called after all keys have been
1387  * initialized by a call to LU_CONTEXT_KEY_INIT().
1388  */
1389 int lu_context_key_register_many(struct lu_context_key *k, ...)
1390 {
1391         struct lu_context_key *key = k;
1392         va_list args;
1393         int result;
1394
1395         va_start(args, k);
1396         do {
1397                 result = lu_context_key_register(key);
1398                 if (result)
1399                         break;
1400                 key = va_arg(args, struct lu_context_key *);
1401         } while (key != NULL);
1402         va_end(args);
1403
1404         if (result != 0) {
1405                 va_start(args, k);
1406                 while (k != key) {
1407                         lu_context_key_degister(k);
1408                         k = va_arg(args, struct lu_context_key *);
1409                 }
1410                 va_end(args);
1411         }
1412
1413         return result;
1414 }
1415 EXPORT_SYMBOL(lu_context_key_register_many);
1416
1417 /**
1418  * De-register a number of keys. This is a dual to
1419  * lu_context_key_register_many().
1420  */
1421 void lu_context_key_degister_many(struct lu_context_key *k, ...)
1422 {
1423         va_list args;
1424
1425         va_start(args, k);
1426         do {
1427                 lu_context_key_degister(k);
1428                 k = va_arg(args, struct lu_context_key*);
1429         } while (k != NULL);
1430         va_end(args);
1431 }
1432 EXPORT_SYMBOL(lu_context_key_degister_many);
1433
1434 /**
1435  * Revive a number of keys.
1436  */
1437 void lu_context_key_revive_many(struct lu_context_key *k, ...)
1438 {
1439         va_list args;
1440
1441         va_start(args, k);
1442         do {
1443                 lu_context_key_revive(k);
1444                 k = va_arg(args, struct lu_context_key*);
1445         } while (k != NULL);
1446         va_end(args);
1447 }
1448 EXPORT_SYMBOL(lu_context_key_revive_many);
1449
1450 /**
1451  * Quiescent a number of keys.
1452  */
1453 void lu_context_key_quiesce_many(struct lu_context_key *k, ...)
1454 {
1455         va_list args;
1456
1457         va_start(args, k);
1458         do {
1459                 lu_context_key_quiesce(k);
1460                 k = va_arg(args, struct lu_context_key*);
1461         } while (k != NULL);
1462         va_end(args);
1463 }
1464 EXPORT_SYMBOL(lu_context_key_quiesce_many);
1465
1466 /**
1467  * Return value associated with key \a key in context \a ctx.
1468  */
1469 void *lu_context_key_get(const struct lu_context *ctx,
1470                          const struct lu_context_key *key)
1471 {
1472         LINVRNT(ctx->lc_state == LCS_ENTERED);
1473         LINVRNT(0 <= key->lct_index && key->lct_index < ARRAY_SIZE(lu_keys));
1474         LASSERT(lu_keys[key->lct_index] == key);
1475         return ctx->lc_value[key->lct_index];
1476 }
1477 EXPORT_SYMBOL(lu_context_key_get);
1478
1479 /**
1480  * List of remembered contexts. XXX document me.
1481  */
1482 static LIST_HEAD(lu_context_remembered);
1483
1484 /**
1485  * Destroy \a key in all remembered contexts. This is used to destroy key
1486  * values in "shared" contexts (like service threads), when a module owning
1487  * the key is about to be unloaded.
1488  */
1489 void lu_context_key_quiesce(struct lu_context_key *key)
1490 {
1491         struct lu_context *ctx;
1492
1493         if (!(key->lct_tags & LCT_QUIESCENT)) {
1494                 /*
1495                  * XXX layering violation.
1496                  */
1497                 key->lct_tags |= LCT_QUIESCENT;
1498                 /*
1499                  * XXX memory barrier has to go here.
1500                  */
1501                 spin_lock(&lu_keys_guard);
1502                 list_for_each_entry(ctx, &lu_context_remembered,
1503                                         lc_remember)
1504                         key_fini(ctx, key->lct_index);
1505                 spin_unlock(&lu_keys_guard);
1506                 ++key_set_version;
1507         }
1508 }
1509 EXPORT_SYMBOL(lu_context_key_quiesce);
1510
1511 void lu_context_key_revive(struct lu_context_key *key)
1512 {
1513         key->lct_tags &= ~LCT_QUIESCENT;
1514         ++key_set_version;
1515 }
1516 EXPORT_SYMBOL(lu_context_key_revive);
1517
1518 static void keys_fini(struct lu_context *ctx)
1519 {
1520         int     i;
1521
1522         if (ctx->lc_value == NULL)
1523                 return;
1524
1525         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i)
1526                 key_fini(ctx, i);
1527
1528         OBD_FREE(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1529         ctx->lc_value = NULL;
1530 }
1531
1532 static int keys_fill(struct lu_context *ctx)
1533 {
1534         int i;
1535
1536         LINVRNT(ctx->lc_value != NULL);
1537         for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1538                 struct lu_context_key *key;
1539
1540                 key = lu_keys[i];
1541                 if (ctx->lc_value[i] == NULL && key != NULL &&
1542                     (key->lct_tags & ctx->lc_tags) &&
1543                     /*
1544                      * Don't create values for a LCT_QUIESCENT key, as this
1545                      * will pin module owning a key.
1546                      */
1547                     !(key->lct_tags & LCT_QUIESCENT)) {
1548                         void *value;
1549
1550                         LINVRNT(key->lct_init != NULL);
1551                         LINVRNT(key->lct_index == i);
1552
1553                         value = key->lct_init(ctx, key);
1554                         if (unlikely(IS_ERR(value)))
1555                                 return PTR_ERR(value);
1556
1557                         if (!(ctx->lc_tags & LCT_NOREF))
1558                                 try_module_get(key->lct_owner);
1559                         lu_ref_add_atomic(&key->lct_reference, "ctx", ctx);
1560                         atomic_inc(&key->lct_used);
1561                         /*
1562                          * This is the only place in the code, where an
1563                          * element of ctx->lc_value[] array is set to non-NULL
1564                          * value.
1565                          */
1566                         ctx->lc_value[i] = value;
1567                         if (key->lct_exit != NULL)
1568                                 ctx->lc_tags |= LCT_HAS_EXIT;
1569                 }
1570                 ctx->lc_version = key_set_version;
1571         }
1572         return 0;
1573 }
1574
1575 static int keys_init(struct lu_context *ctx)
1576 {
1577         OBD_ALLOC(ctx->lc_value, ARRAY_SIZE(lu_keys) * sizeof ctx->lc_value[0]);
1578         if (likely(ctx->lc_value != NULL))
1579                 return keys_fill(ctx);
1580
1581         return -ENOMEM;
1582 }
1583
1584 /**
1585  * Initialize context data-structure. Create values for all keys.
1586  */
1587 int lu_context_init(struct lu_context *ctx, __u32 tags)
1588 {
1589         int     rc;
1590
1591         memset(ctx, 0, sizeof *ctx);
1592         ctx->lc_state = LCS_INITIALIZED;
1593         ctx->lc_tags = tags;
1594         if (tags & LCT_REMEMBER) {
1595                 spin_lock(&lu_keys_guard);
1596                 list_add(&ctx->lc_remember, &lu_context_remembered);
1597                 spin_unlock(&lu_keys_guard);
1598         } else {
1599                 INIT_LIST_HEAD(&ctx->lc_remember);
1600         }
1601
1602         rc = keys_init(ctx);
1603         if (rc != 0)
1604                 lu_context_fini(ctx);
1605
1606         return rc;
1607 }
1608 EXPORT_SYMBOL(lu_context_init);
1609
1610 /**
1611  * Finalize context data-structure. Destroy key values.
1612  */
1613 void lu_context_fini(struct lu_context *ctx)
1614 {
1615         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1616         ctx->lc_state = LCS_FINALIZED;
1617
1618         if ((ctx->lc_tags & LCT_REMEMBER) == 0) {
1619                 LASSERT(list_empty(&ctx->lc_remember));
1620                 keys_fini(ctx);
1621
1622         } else { /* could race with key degister */
1623                 spin_lock(&lu_keys_guard);
1624                 keys_fini(ctx);
1625                 list_del_init(&ctx->lc_remember);
1626                 spin_unlock(&lu_keys_guard);
1627         }
1628 }
1629 EXPORT_SYMBOL(lu_context_fini);
1630
1631 /**
1632  * Called before entering context.
1633  */
1634 void lu_context_enter(struct lu_context *ctx)
1635 {
1636         LINVRNT(ctx->lc_state == LCS_INITIALIZED || ctx->lc_state == LCS_LEFT);
1637         ctx->lc_state = LCS_ENTERED;
1638 }
1639 EXPORT_SYMBOL(lu_context_enter);
1640
1641 /**
1642  * Called after exiting from \a ctx
1643  */
1644 void lu_context_exit(struct lu_context *ctx)
1645 {
1646         int i;
1647
1648         LINVRNT(ctx->lc_state == LCS_ENTERED);
1649         ctx->lc_state = LCS_LEFT;
1650         if (ctx->lc_tags & LCT_HAS_EXIT && ctx->lc_value != NULL) {
1651                 for (i = 0; i < ARRAY_SIZE(lu_keys); ++i) {
1652                         if (ctx->lc_value[i] != NULL) {
1653                                 struct lu_context_key *key;
1654
1655                                 key = lu_keys[i];
1656                                 LASSERT(key != NULL);
1657                                 if (key->lct_exit != NULL)
1658                                         key->lct_exit(ctx,
1659                                                       key, ctx->lc_value[i]);
1660                         }
1661                 }
1662         }
1663 }
1664 EXPORT_SYMBOL(lu_context_exit);
1665
1666 /**
1667  * Allocate for context all missing keys that were registered after context
1668  * creation. key_set_version is only changed in rare cases when modules
1669  * are loaded and removed.
1670  */
1671 int lu_context_refill(struct lu_context *ctx)
1672 {
1673         return likely(ctx->lc_version == key_set_version) ? 0 : keys_fill(ctx);
1674 }
1675 EXPORT_SYMBOL(lu_context_refill);
1676
1677 /**
1678  * lu_ctx_tags/lu_ses_tags will be updated if there are new types of
1679  * obd being added. Currently, this is only used on client side, specifically
1680  * for echo device client, for other stack (like ptlrpc threads), context are
1681  * predefined when the lu_device type are registered, during the module probe
1682  * phase.
1683  */
1684 __u32 lu_context_tags_default = 0;
1685 __u32 lu_session_tags_default = 0;
1686
1687 void lu_context_tags_update(__u32 tags)
1688 {
1689         spin_lock(&lu_keys_guard);
1690         lu_context_tags_default |= tags;
1691         key_set_version++;
1692         spin_unlock(&lu_keys_guard);
1693 }
1694 EXPORT_SYMBOL(lu_context_tags_update);
1695
1696 void lu_context_tags_clear(__u32 tags)
1697 {
1698         spin_lock(&lu_keys_guard);
1699         lu_context_tags_default &= ~tags;
1700         key_set_version++;
1701         spin_unlock(&lu_keys_guard);
1702 }
1703 EXPORT_SYMBOL(lu_context_tags_clear);
1704
1705 void lu_session_tags_update(__u32 tags)
1706 {
1707         spin_lock(&lu_keys_guard);
1708         lu_session_tags_default |= tags;
1709         key_set_version++;
1710         spin_unlock(&lu_keys_guard);
1711 }
1712 EXPORT_SYMBOL(lu_session_tags_update);
1713
1714 void lu_session_tags_clear(__u32 tags)
1715 {
1716         spin_lock(&lu_keys_guard);
1717         lu_session_tags_default &= ~tags;
1718         key_set_version++;
1719         spin_unlock(&lu_keys_guard);
1720 }
1721 EXPORT_SYMBOL(lu_session_tags_clear);
1722
1723 int lu_env_init(struct lu_env *env, __u32 tags)
1724 {
1725         int result;
1726
1727         env->le_ses = NULL;
1728         result = lu_context_init(&env->le_ctx, tags);
1729         if (likely(result == 0))
1730                 lu_context_enter(&env->le_ctx);
1731         return result;
1732 }
1733 EXPORT_SYMBOL(lu_env_init);
1734
1735 void lu_env_fini(struct lu_env *env)
1736 {
1737         lu_context_exit(&env->le_ctx);
1738         lu_context_fini(&env->le_ctx);
1739         env->le_ses = NULL;
1740 }
1741 EXPORT_SYMBOL(lu_env_fini);
1742
1743 int lu_env_refill(struct lu_env *env)
1744 {
1745         int result;
1746
1747         result = lu_context_refill(&env->le_ctx);
1748         if (result == 0 && env->le_ses != NULL)
1749                 result = lu_context_refill(env->le_ses);
1750         return result;
1751 }
1752 EXPORT_SYMBOL(lu_env_refill);
1753
1754 /**
1755  * Currently, this API will only be used by echo client.
1756  * Because echo client and normal lustre client will share
1757  * same cl_env cache. So echo client needs to refresh
1758  * the env context after it get one from the cache, especially
1759  * when normal client and echo client co-exist in the same client.
1760  */
1761 int lu_env_refill_by_tags(struct lu_env *env, __u32 ctags,
1762                           __u32 stags)
1763 {
1764         int    result;
1765
1766         if ((env->le_ctx.lc_tags & ctags) != ctags) {
1767                 env->le_ctx.lc_version = 0;
1768                 env->le_ctx.lc_tags |= ctags;
1769         }
1770
1771         if (env->le_ses && (env->le_ses->lc_tags & stags) != stags) {
1772                 env->le_ses->lc_version = 0;
1773                 env->le_ses->lc_tags |= stags;
1774         }
1775
1776         result = lu_env_refill(env);
1777
1778         return result;
1779 }
1780 EXPORT_SYMBOL(lu_env_refill_by_tags);
1781
1782
1783 typedef struct lu_site_stats{
1784         unsigned        lss_populated;
1785         unsigned        lss_max_search;
1786         unsigned        lss_total;
1787         unsigned        lss_busy;
1788 } lu_site_stats_t;
1789
1790 static void lu_site_stats_get(cfs_hash_t *hs,
1791                               lu_site_stats_t *stats, int populated)
1792 {
1793         cfs_hash_bd_t bd;
1794         int        i;
1795
1796         cfs_hash_for_each_bucket(hs, &bd, i) {
1797                 struct lu_site_bkt_data *bkt = cfs_hash_bd_extra_get(hs, &bd);
1798                 struct hlist_head       *hhead;
1799
1800                 cfs_hash_bd_lock(hs, &bd, 1);
1801                 stats->lss_busy  += bkt->lsb_busy;
1802                 stats->lss_total += cfs_hash_bd_count_get(&bd);
1803                 stats->lss_max_search = max((int)stats->lss_max_search,
1804                                             cfs_hash_bd_depmax_get(&bd));
1805                 if (!populated) {
1806                         cfs_hash_bd_unlock(hs, &bd, 1);
1807                         continue;
1808                 }
1809
1810                 cfs_hash_bd_for_each_hlist(hs, &bd, hhead) {
1811                         if (!hlist_empty(hhead))
1812                                 stats->lss_populated++;
1813                 }
1814                 cfs_hash_bd_unlock(hs, &bd, 1);
1815         }
1816 }
1817
1818
1819 /*
1820  * There exists a potential lock inversion deadlock scenario when using
1821  * Lustre on top of ZFS. This occurs between one of ZFS's
1822  * buf_hash_table.ht_lock's, and Lustre's lu_sites_guard lock. Essentially,
1823  * thread A will take the lu_sites_guard lock and sleep on the ht_lock,
1824  * while thread B will take the ht_lock and sleep on the lu_sites_guard
1825  * lock. Obviously neither thread will wake and drop their respective hold
1826  * on their lock.
1827  *
1828  * To prevent this from happening we must ensure the lu_sites_guard lock is
1829  * not taken while down this code path. ZFS reliably does not set the
1830  * __GFP_FS bit in its code paths, so this can be used to determine if it
1831  * is safe to take the lu_sites_guard lock.
1832  *
1833  * Ideally we should accurately return the remaining number of cached
1834  * objects without taking the  lu_sites_guard lock, but this is not
1835  * possible in the current implementation.
1836  */
1837 static unsigned long lu_cache_shrink_count(struct shrinker *sk,
1838                                            struct shrink_control *sc)
1839 {
1840         lu_site_stats_t stats;
1841         struct lu_site *s;
1842         struct lu_site *tmp;
1843         unsigned long cached = 0;
1844
1845         if (!(sc->gfp_mask & __GFP_FS))
1846                 return 0;
1847
1848         mutex_lock(&lu_sites_guard);
1849         list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
1850                 memset(&stats, 0, sizeof(stats));
1851                 lu_site_stats_get(s->ls_obj_hash, &stats, 0);
1852                 cached += stats.lss_total - stats.lss_busy;
1853         }
1854         mutex_unlock(&lu_sites_guard);
1855
1856         cached = (cached / 100) * sysctl_vfs_cache_pressure;
1857         CDEBUG(D_INODE, "%ld objects cached\n", cached);
1858         return cached;
1859 }
1860
1861 static unsigned long lu_cache_shrink_scan(struct shrinker *sk,
1862                                           struct shrink_control *sc)
1863 {
1864         struct lu_site *s;
1865         struct lu_site *tmp;
1866         unsigned long remain = sc->nr_to_scan, freed = 0;
1867         LIST_HEAD(splice);
1868
1869         if (!(sc->gfp_mask & __GFP_FS))
1870                 /* We must not take the lu_sites_guard lock when
1871                  * __GFP_FS is *not* set because of the deadlock
1872                  * possibility detailed above. Additionally,
1873                  * since we cannot determine the number of
1874                  * objects in the cache without taking this
1875                  * lock, we're in a particularly tough spot. As
1876                  * a result, we'll just lie and say our cache is
1877                  * empty. This _should_ be ok, as we can't
1878                  * reclaim objects when __GFP_FS is *not* set
1879                  * anyways.
1880                  */
1881                 return SHRINK_STOP;
1882
1883         mutex_lock(&lu_sites_guard);
1884         list_for_each_entry_safe(s, tmp, &lu_sites, ls_linkage) {
1885                 freed = lu_site_purge(&lu_shrink_env, s, remain);
1886                 remain -= freed;
1887                 /*
1888                  * Move just shrunk site to the tail of site list to
1889                  * assure shrinking fairness.
1890                  */
1891                 list_move_tail(&s->ls_linkage, &splice);
1892         }
1893         list_splice(&splice, lu_sites.prev);
1894         mutex_unlock(&lu_sites_guard);
1895
1896         return sc->nr_to_scan - remain;
1897 }
1898
1899 /*
1900  * Debugging stuff.
1901  */
1902
1903 /**
1904  * Environment to be used in debugger, contains all tags.
1905  */
1906 struct lu_env lu_debugging_env;
1907
1908 /**
1909  * Debugging printer function using printk().
1910  */
1911 int lu_printk_printer(const struct lu_env *env,
1912                       void *unused, const char *format, ...)
1913 {
1914         va_list args;
1915
1916         va_start(args, format);
1917         vprintk(format, args);
1918         va_end(args);
1919         return 0;
1920 }
1921
1922 static struct shrinker lu_site_shrinker = {
1923         .count_objects  = lu_cache_shrink_count,
1924         .scan_objects   = lu_cache_shrink_scan,
1925         .seeks          = DEFAULT_SEEKS,
1926 };
1927
1928 /**
1929  * Initialization of global lu_* data.
1930  */
1931 int lu_global_init(void)
1932 {
1933         int result;
1934
1935         CDEBUG(D_INFO, "Lustre LU module (%p).\n", &lu_keys);
1936
1937         result = lu_ref_global_init();
1938         if (result != 0)
1939                 return result;
1940
1941         LU_CONTEXT_KEY_INIT(&lu_global_key);
1942         result = lu_context_key_register(&lu_global_key);
1943         if (result != 0)
1944                 return result;
1945
1946         /*
1947          * At this level, we don't know what tags are needed, so allocate them
1948          * conservatively. This should not be too bad, because this
1949          * environment is global.
1950          */
1951         mutex_lock(&lu_sites_guard);
1952         result = lu_env_init(&lu_shrink_env, LCT_SHRINKER);
1953         mutex_unlock(&lu_sites_guard);
1954         if (result != 0)
1955                 return result;
1956
1957         /*
1958          * seeks estimation: 3 seeks to read a record from oi, one to read
1959          * inode, one for ea. Unfortunately setting this high value results in
1960          * lu_object/inode cache consuming all the memory.
1961          */
1962         register_shrinker(&lu_site_shrinker);
1963
1964         return result;
1965 }
1966
1967 /**
1968  * Dual to lu_global_init().
1969  */
1970 void lu_global_fini(void)
1971 {
1972         unregister_shrinker(&lu_site_shrinker);
1973         lu_context_key_degister(&lu_global_key);
1974
1975         /*
1976          * Tear shrinker environment down _after_ de-registering
1977          * lu_global_key, because the latter has a value in the former.
1978          */
1979         mutex_lock(&lu_sites_guard);
1980         lu_env_fini(&lu_shrink_env);
1981         mutex_unlock(&lu_sites_guard);
1982
1983         lu_ref_global_fini();
1984 }
1985
1986 static __u32 ls_stats_read(struct lprocfs_stats *stats, int idx)
1987 {
1988 #ifdef LPROCFS
1989         struct lprocfs_counter ret;
1990
1991         lprocfs_stats_collect(stats, idx, &ret);
1992         return (__u32)ret.lc_count;
1993 #else
1994         return 0;
1995 #endif
1996 }
1997
1998 /**
1999  * Output site statistical counters into a buffer. Suitable for
2000  * lprocfs_rd_*()-style functions.
2001  */
2002 int lu_site_stats_print(const struct lu_site *s, struct seq_file *m)
2003 {
2004         lu_site_stats_t stats;
2005
2006         memset(&stats, 0, sizeof(stats));
2007         lu_site_stats_get(s->ls_obj_hash, &stats, 1);
2008
2009         return seq_printf(m, "%d/%d %d/%d %d %d %d %d %d %d %d\n",
2010                         stats.lss_busy,
2011                         stats.lss_total,
2012                         stats.lss_populated,
2013                         CFS_HASH_NHLIST(s->ls_obj_hash),
2014                         stats.lss_max_search,
2015                         ls_stats_read(s->ls_stats, LU_SS_CREATED),
2016                         ls_stats_read(s->ls_stats, LU_SS_CACHE_HIT),
2017                         ls_stats_read(s->ls_stats, LU_SS_CACHE_MISS),
2018                         ls_stats_read(s->ls_stats, LU_SS_CACHE_RACE),
2019                         ls_stats_read(s->ls_stats, LU_SS_CACHE_DEATH_RACE),
2020                         ls_stats_read(s->ls_stats, LU_SS_LRU_PURGED));
2021 }
2022 EXPORT_SYMBOL(lu_site_stats_print);
2023
2024 /**
2025  * Helper function to initialize a number of kmem slab caches at once.
2026  */
2027 int lu_kmem_init(struct lu_kmem_descr *caches)
2028 {
2029         int result;
2030         struct lu_kmem_descr *iter = caches;
2031
2032         for (result = 0; iter->ckd_cache != NULL; ++iter) {
2033                 *iter->ckd_cache = kmem_cache_create(iter->ckd_name,
2034                                                         iter->ckd_size,
2035                                                         0, 0, NULL);
2036                 if (*iter->ckd_cache == NULL) {
2037                         result = -ENOMEM;
2038                         /* free all previously allocated caches */
2039                         lu_kmem_fini(caches);
2040                         break;
2041                 }
2042         }
2043         return result;
2044 }
2045 EXPORT_SYMBOL(lu_kmem_init);
2046
2047 /**
2048  * Helper function to finalize a number of kmem slab cached at once. Dual to
2049  * lu_kmem_init().
2050  */
2051 void lu_kmem_fini(struct lu_kmem_descr *caches)
2052 {
2053         for (; caches->ckd_cache != NULL; ++caches) {
2054                 if (*caches->ckd_cache != NULL) {
2055                         kmem_cache_destroy(*caches->ckd_cache);
2056                         *caches->ckd_cache = NULL;
2057                 }
2058         }
2059 }
2060 EXPORT_SYMBOL(lu_kmem_fini);
2061
2062 /**
2063  * Temporary solution to be able to assign fid in ->do_create()
2064  * till we have fully-functional OST fids
2065  */
2066 void lu_object_assign_fid(const struct lu_env *env, struct lu_object *o,
2067                           const struct lu_fid *fid)
2068 {
2069         struct lu_site          *s = o->lo_dev->ld_site;
2070         struct lu_fid           *old = &o->lo_header->loh_fid;
2071         struct lu_site_bkt_data *bkt;
2072         struct lu_object        *shadow;
2073         wait_queue_t             waiter;
2074         cfs_hash_t              *hs;
2075         cfs_hash_bd_t            bd;
2076         __u64                    version = 0;
2077
2078         LASSERT(fid_is_zero(old));
2079
2080         hs = s->ls_obj_hash;
2081         cfs_hash_bd_get_and_lock(hs, (void *)fid, &bd, 1);
2082         shadow = htable_lookup(s, &bd, fid, &waiter, &version);
2083         /* supposed to be unique */
2084         LASSERT(IS_ERR(shadow) && PTR_ERR(shadow) == -ENOENT);
2085         *old = *fid;
2086         bkt = cfs_hash_bd_extra_get(hs, &bd);
2087         cfs_hash_bd_add_locked(hs, &bd, &o->lo_header->loh_hash);
2088         bkt->lsb_busy++;
2089         cfs_hash_bd_unlock(hs, &bd, 1);
2090 }
2091 EXPORT_SYMBOL(lu_object_assign_fid);
2092
2093 /**
2094  * allocates object with 0 (non-assiged) fid
2095  * XXX: temporary solution to be able to assign fid in ->do_create()
2096  *      till we have fully-functional OST fids
2097  */
2098 struct lu_object *lu_object_anon(const struct lu_env *env,
2099                                  struct lu_device *dev,
2100                                  const struct lu_object_conf *conf)
2101 {
2102         struct lu_fid     fid;
2103         struct lu_object *o;
2104
2105         fid_zero(&fid);
2106         o = lu_object_alloc(env, dev, &fid, conf);
2107
2108         return o;
2109 }
2110 EXPORT_SYMBOL(lu_object_anon);
2111
2112 struct lu_buf LU_BUF_NULL = {
2113         .lb_buf = NULL,
2114         .lb_len = 0
2115 };
2116 EXPORT_SYMBOL(LU_BUF_NULL);
2117
2118 void lu_buf_free(struct lu_buf *buf)
2119 {
2120         LASSERT(buf);
2121         if (buf->lb_buf) {
2122                 LASSERT(buf->lb_len > 0);
2123                 OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
2124                 buf->lb_buf = NULL;
2125                 buf->lb_len = 0;
2126         }
2127 }
2128 EXPORT_SYMBOL(lu_buf_free);
2129
2130 void lu_buf_alloc(struct lu_buf *buf, int size)
2131 {
2132         LASSERT(buf);
2133         LASSERT(buf->lb_buf == NULL);
2134         LASSERT(buf->lb_len == 0);
2135         OBD_ALLOC_LARGE(buf->lb_buf, size);
2136         if (likely(buf->lb_buf))
2137                 buf->lb_len = size;
2138 }
2139 EXPORT_SYMBOL(lu_buf_alloc);
2140
2141 void lu_buf_realloc(struct lu_buf *buf, int size)
2142 {
2143         lu_buf_free(buf);
2144         lu_buf_alloc(buf, size);
2145 }
2146 EXPORT_SYMBOL(lu_buf_realloc);
2147
2148 struct lu_buf *lu_buf_check_and_alloc(struct lu_buf *buf, int len)
2149 {
2150         if (buf->lb_buf == NULL && buf->lb_len == 0)
2151                 lu_buf_alloc(buf, len);
2152
2153         if ((len > buf->lb_len) && (buf->lb_buf != NULL))
2154                 lu_buf_realloc(buf, len);
2155
2156         return buf;
2157 }
2158 EXPORT_SYMBOL(lu_buf_check_and_alloc);
2159
2160 /**
2161  * Increase the size of the \a buf.
2162  * preserves old data in buffer
2163  * old buffer remains unchanged on error
2164  * \retval 0 or -ENOMEM
2165  */
2166 int lu_buf_check_and_grow(struct lu_buf *buf, int len)
2167 {
2168         char *ptr;
2169
2170         if (len <= buf->lb_len)
2171                 return 0;
2172
2173         OBD_ALLOC_LARGE(ptr, len);
2174         if (ptr == NULL)
2175                 return -ENOMEM;
2176
2177         /* Free the old buf */
2178         if (buf->lb_buf != NULL) {
2179                 memcpy(ptr, buf->lb_buf, buf->lb_len);
2180                 OBD_FREE_LARGE(buf->lb_buf, buf->lb_len);
2181         }
2182
2183         buf->lb_buf = ptr;
2184         buf->lb_len = len;
2185         return 0;
2186 }
2187 EXPORT_SYMBOL(lu_buf_check_and_grow);