linux/drivers/staging/lustre/lustre/obdclass/cl_object.c
<<
>>
Prefs
   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) 2008, 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 * Client Lustre Object.
  37 *
  38 *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39 */
  40
  41/*
  42 * Locking.
  43 *
  44 *  i_mutex
  45 *      PG_locked
  46 *        ->coh_page_guard
  47 *        ->coh_lock_guard
  48 *        ->coh_attr_guard
  49 *        ->ls_guard
  50 */
  51
  52#define DEBUG_SUBSYSTEM S_CLASS
  53
  54#include <linux/libcfs/libcfs.h>
  55/* class_put_type() */
  56#include <obd_class.h>
  57#include <obd_support.h>
  58#include <lustre_fid.h>
  59#include <linux/list.h>
  60#include <linux/libcfs/libcfs_hash.h> /* for cfs_hash stuff */
  61#include <cl_object.h>
  62#include "cl_internal.h"
  63
  64static struct kmem_cache *cl_env_kmem;
  65
  66/** Lock class of cl_object_header::coh_page_guard */
  67static struct lock_class_key cl_page_guard_class;
  68/** Lock class of cl_object_header::coh_lock_guard */
  69static struct lock_class_key cl_lock_guard_class;
  70/** Lock class of cl_object_header::coh_attr_guard */
  71static struct lock_class_key cl_attr_guard_class;
  72
  73extern __u32 lu_context_tags_default;
  74extern __u32 lu_session_tags_default;
  75/**
  76 * Initialize cl_object_header.
  77 */
  78int cl_object_header_init(struct cl_object_header *h)
  79{
  80        int result;
  81
  82        result = lu_object_header_init(&h->coh_lu);
  83        if (result == 0) {
  84                spin_lock_init(&h->coh_page_guard);
  85                spin_lock_init(&h->coh_lock_guard);
  86                spin_lock_init(&h->coh_attr_guard);
  87                lockdep_set_class(&h->coh_page_guard, &cl_page_guard_class);
  88                lockdep_set_class(&h->coh_lock_guard, &cl_lock_guard_class);
  89                lockdep_set_class(&h->coh_attr_guard, &cl_attr_guard_class);
  90                h->coh_pages = 0;
  91                /* XXX hard coded GFP_* mask. */
  92                INIT_RADIX_TREE(&h->coh_tree, GFP_ATOMIC);
  93                INIT_LIST_HEAD(&h->coh_locks);
  94                h->coh_page_bufsize = ALIGN(sizeof(struct cl_page), 8);
  95        }
  96        return result;
  97}
  98EXPORT_SYMBOL(cl_object_header_init);
  99
 100/**
 101 * Finalize cl_object_header.
 102 */
 103void cl_object_header_fini(struct cl_object_header *h)
 104{
 105        LASSERT(list_empty(&h->coh_locks));
 106        lu_object_header_fini(&h->coh_lu);
 107}
 108EXPORT_SYMBOL(cl_object_header_fini);
 109
 110/**
 111 * Returns a cl_object with a given \a fid.
 112 *
 113 * Returns either cached or newly created object. Additional reference on the
 114 * returned object is acquired.
 115 *
 116 * \see lu_object_find(), cl_page_find(), cl_lock_find()
 117 */
 118struct cl_object *cl_object_find(const struct lu_env *env,
 119                                 struct cl_device *cd, const struct lu_fid *fid,
 120                                 const struct cl_object_conf *c)
 121{
 122        might_sleep();
 123        return lu2cl(lu_object_find_slice(env, cl2lu_dev(cd), fid, &c->coc_lu));
 124}
 125EXPORT_SYMBOL(cl_object_find);
 126
 127/**
 128 * Releases a reference on \a o.
 129 *
 130 * When last reference is released object is returned to the cache, unless
 131 * lu_object_header_flags::LU_OBJECT_HEARD_BANSHEE bit is set in its header.
 132 *
 133 * \see cl_page_put(), cl_lock_put().
 134 */
 135void cl_object_put(const struct lu_env *env, struct cl_object *o)
 136{
 137        lu_object_put(env, &o->co_lu);
 138}
 139EXPORT_SYMBOL(cl_object_put);
 140
 141/**
 142 * Acquire an additional reference to the object \a o.
 143 *
 144 * This can only be used to acquire _additional_ reference, i.e., caller
 145 * already has to possess at least one reference to \a o before calling this.
 146 *
 147 * \see cl_page_get(), cl_lock_get().
 148 */
 149void cl_object_get(struct cl_object *o)
 150{
 151        lu_object_get(&o->co_lu);
 152}
 153EXPORT_SYMBOL(cl_object_get);
 154
 155/**
 156 * Returns the top-object for a given \a o.
 157 *
 158 * \see cl_page_top(), cl_io_top()
 159 */
 160struct cl_object *cl_object_top(struct cl_object *o)
 161{
 162        struct cl_object_header *hdr = cl_object_header(o);
 163        struct cl_object *top;
 164
 165        while (hdr->coh_parent != NULL)
 166                hdr = hdr->coh_parent;
 167
 168        top = lu2cl(lu_object_top(&hdr->coh_lu));
 169        CDEBUG(D_TRACE, "%p -> %p\n", o, top);
 170        return top;
 171}
 172EXPORT_SYMBOL(cl_object_top);
 173
 174/**
 175 * Returns pointer to the lock protecting data-attributes for the given object
 176 * \a o.
 177 *
 178 * Data-attributes are protected by the cl_object_header::coh_attr_guard
 179 * spin-lock in the top-object.
 180 *
 181 * \see cl_attr, cl_object_attr_lock(), cl_object_operations::coo_attr_get().
 182 */
 183static spinlock_t *cl_object_attr_guard(struct cl_object *o)
 184{
 185        return &cl_object_header(cl_object_top(o))->coh_attr_guard;
 186}
 187
 188/**
 189 * Locks data-attributes.
 190 *
 191 * Prevents data-attributes from changing, until lock is released by
 192 * cl_object_attr_unlock(). This has to be called before calls to
 193 * cl_object_attr_get(), cl_object_attr_set().
 194 */
 195void cl_object_attr_lock(struct cl_object *o)
 196{
 197        spin_lock(cl_object_attr_guard(o));
 198}
 199EXPORT_SYMBOL(cl_object_attr_lock);
 200
 201/**
 202 * Releases data-attributes lock, acquired by cl_object_attr_lock().
 203 */
 204void cl_object_attr_unlock(struct cl_object *o)
 205{
 206        spin_unlock(cl_object_attr_guard(o));
 207}
 208EXPORT_SYMBOL(cl_object_attr_unlock);
 209
 210/**
 211 * Returns data-attributes of an object \a obj.
 212 *
 213 * Every layer is asked (by calling cl_object_operations::coo_attr_get())
 214 * top-to-bottom to fill in parts of \a attr that this layer is responsible
 215 * for.
 216 */
 217int cl_object_attr_get(const struct lu_env *env, struct cl_object *obj,
 218                       struct cl_attr *attr)
 219{
 220        struct lu_object_header *top;
 221        int result;
 222
 223        LASSERT(spin_is_locked(cl_object_attr_guard(obj)));
 224
 225        top = obj->co_lu.lo_header;
 226        result = 0;
 227        list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
 228                if (obj->co_ops->coo_attr_get != NULL) {
 229                        result = obj->co_ops->coo_attr_get(env, obj, attr);
 230                        if (result != 0) {
 231                                if (result > 0)
 232                                        result = 0;
 233                                break;
 234                        }
 235                }
 236        }
 237        return result;
 238}
 239EXPORT_SYMBOL(cl_object_attr_get);
 240
 241/**
 242 * Updates data-attributes of an object \a obj.
 243 *
 244 * Only attributes, mentioned in a validness bit-mask \a v are
 245 * updated. Calls cl_object_operations::coo_attr_set() on every layer, bottom
 246 * to top.
 247 */
 248int cl_object_attr_set(const struct lu_env *env, struct cl_object *obj,
 249                       const struct cl_attr *attr, unsigned v)
 250{
 251        struct lu_object_header *top;
 252        int result;
 253
 254        LASSERT(spin_is_locked(cl_object_attr_guard(obj)));
 255
 256        top = obj->co_lu.lo_header;
 257        result = 0;
 258        list_for_each_entry_reverse(obj, &top->loh_layers,
 259                                        co_lu.lo_linkage) {
 260                if (obj->co_ops->coo_attr_set != NULL) {
 261                        result = obj->co_ops->coo_attr_set(env, obj, attr, v);
 262                        if (result != 0) {
 263                                if (result > 0)
 264                                        result = 0;
 265                                break;
 266                        }
 267                }
 268        }
 269        return result;
 270}
 271EXPORT_SYMBOL(cl_object_attr_set);
 272
 273/**
 274 * Notifies layers (bottom-to-top) that glimpse AST was received.
 275 *
 276 * Layers have to fill \a lvb fields with information that will be shipped
 277 * back to glimpse issuer.
 278 *
 279 * \see cl_lock_operations::clo_glimpse()
 280 */
 281int cl_object_glimpse(const struct lu_env *env, struct cl_object *obj,
 282                      struct ost_lvb *lvb)
 283{
 284        struct lu_object_header *top;
 285        int result;
 286
 287        top = obj->co_lu.lo_header;
 288        result = 0;
 289        list_for_each_entry_reverse(obj, &top->loh_layers,
 290                                        co_lu.lo_linkage) {
 291                if (obj->co_ops->coo_glimpse != NULL) {
 292                        result = obj->co_ops->coo_glimpse(env, obj, lvb);
 293                        if (result != 0)
 294                                break;
 295                }
 296        }
 297        LU_OBJECT_HEADER(D_DLMTRACE, env, lu_object_top(top),
 298                         "size: "LPU64" mtime: "LPU64" atime: "LPU64" "
 299                         "ctime: "LPU64" blocks: "LPU64"\n",
 300                         lvb->lvb_size, lvb->lvb_mtime, lvb->lvb_atime,
 301                         lvb->lvb_ctime, lvb->lvb_blocks);
 302        return result;
 303}
 304EXPORT_SYMBOL(cl_object_glimpse);
 305
 306/**
 307 * Updates a configuration of an object \a obj.
 308 */
 309int cl_conf_set(const struct lu_env *env, struct cl_object *obj,
 310                const struct cl_object_conf *conf)
 311{
 312        struct lu_object_header *top;
 313        int result;
 314
 315        top = obj->co_lu.lo_header;
 316        result = 0;
 317        list_for_each_entry(obj, &top->loh_layers, co_lu.lo_linkage) {
 318                if (obj->co_ops->coo_conf_set != NULL) {
 319                        result = obj->co_ops->coo_conf_set(env, obj, conf);
 320                        if (result != 0)
 321                                break;
 322                }
 323        }
 324        return result;
 325}
 326EXPORT_SYMBOL(cl_conf_set);
 327
 328/**
 329 * Helper function removing all object locks, and marking object for
 330 * deletion. All object pages must have been deleted at this point.
 331 *
 332 * This is called by cl_inode_fini() and lov_object_delete() to destroy top-
 333 * and sub- objects respectively.
 334 */
 335void cl_object_kill(const struct lu_env *env, struct cl_object *obj)
 336{
 337        struct cl_object_header *hdr;
 338
 339        hdr = cl_object_header(obj);
 340        LASSERT(hdr->coh_tree.rnode == NULL);
 341        LASSERT(hdr->coh_pages == 0);
 342
 343        set_bit(LU_OBJECT_HEARD_BANSHEE, &hdr->coh_lu.loh_flags);
 344        /*
 345         * Destroy all locks. Object destruction (including cl_inode_fini())
 346         * cannot cancel the locks, because in the case of a local client,
 347         * where client and server share the same thread running
 348         * prune_icache(), this can dead-lock with ldlm_cancel_handler()
 349         * waiting on __wait_on_freeing_inode().
 350         */
 351        cl_locks_prune(env, obj, 0);
 352}
 353EXPORT_SYMBOL(cl_object_kill);
 354
 355/**
 356 * Prunes caches of pages and locks for this object.
 357 */
 358void cl_object_prune(const struct lu_env *env, struct cl_object *obj)
 359{
 360        cl_pages_prune(env, obj);
 361        cl_locks_prune(env, obj, 1);
 362}
 363EXPORT_SYMBOL(cl_object_prune);
 364
 365/**
 366 * Check if the object has locks.
 367 */
 368int cl_object_has_locks(struct cl_object *obj)
 369{
 370        struct cl_object_header *head = cl_object_header(obj);
 371        int has;
 372
 373        spin_lock(&head->coh_lock_guard);
 374        has = list_empty(&head->coh_locks);
 375        spin_unlock(&head->coh_lock_guard);
 376
 377        return (has == 0);
 378}
 379EXPORT_SYMBOL(cl_object_has_locks);
 380
 381void cache_stats_init(struct cache_stats *cs, const char *name)
 382{
 383        int i;
 384
 385        cs->cs_name = name;
 386        for (i = 0; i < CS_NR; i++)
 387                atomic_set(&cs->cs_stats[i], 0);
 388}
 389
 390int cache_stats_print(const struct cache_stats *cs, struct seq_file *m, int h)
 391{
 392        int i;
 393        /*
 394         *   lookup    hit    total  cached create
 395         * env: ...... ...... ...... ...... ......
 396         */
 397        if (h) {
 398                const char *names[CS_NR] = CS_NAMES;
 399
 400                seq_printf(m, "%6s", " ");
 401                for (i = 0; i < CS_NR; i++)
 402                        seq_printf(m, "%8s", names[i]);
 403                seq_printf(m, "\n");
 404        }
 405
 406        seq_printf(m, "%5.5s:", cs->cs_name);
 407        for (i = 0; i < CS_NR; i++)
 408                seq_printf(m, "%8u", atomic_read(&cs->cs_stats[i]));
 409        return 0;
 410}
 411
 412/**
 413 * Initialize client site.
 414 *
 415 * Perform common initialization (lu_site_init()), and initialize statistical
 416 * counters. Also perform global initializations on the first call.
 417 */
 418int cl_site_init(struct cl_site *s, struct cl_device *d)
 419{
 420        int i;
 421        int result;
 422
 423        result = lu_site_init(&s->cs_lu, &d->cd_lu_dev);
 424        if (result == 0) {
 425                cache_stats_init(&s->cs_pages, "pages");
 426                cache_stats_init(&s->cs_locks, "locks");
 427                for (i = 0; i < ARRAY_SIZE(s->cs_pages_state); ++i)
 428                        atomic_set(&s->cs_pages_state[0], 0);
 429                for (i = 0; i < ARRAY_SIZE(s->cs_locks_state); ++i)
 430                        atomic_set(&s->cs_locks_state[i], 0);
 431        }
 432        return result;
 433}
 434EXPORT_SYMBOL(cl_site_init);
 435
 436/**
 437 * Finalize client site. Dual to cl_site_init().
 438 */
 439void cl_site_fini(struct cl_site *s)
 440{
 441        lu_site_fini(&s->cs_lu);
 442}
 443EXPORT_SYMBOL(cl_site_fini);
 444
 445static struct cache_stats cl_env_stats = {
 446        .cs_name    = "envs",
 447        .cs_stats = { ATOMIC_INIT(0), }
 448};
 449
 450/**
 451 * Outputs client site statistical counters into a buffer. Suitable for
 452 * ll_rd_*()-style functions.
 453 */
 454int cl_site_stats_print(const struct cl_site *site, struct seq_file *m)
 455{
 456        int i;
 457        static const char *pstate[] = {
 458                [CPS_CACHED]  = "c",
 459                [CPS_OWNED]   = "o",
 460                [CPS_PAGEOUT] = "w",
 461                [CPS_PAGEIN]  = "r",
 462                [CPS_FREEING] = "f"
 463        };
 464        static const char *lstate[] = {
 465                [CLS_NEW]       = "n",
 466                [CLS_QUEUING]   = "q",
 467                [CLS_ENQUEUED]  = "e",
 468                [CLS_HELD]      = "h",
 469                [CLS_INTRANSIT] = "t",
 470                [CLS_CACHED]    = "c",
 471                [CLS_FREEING]   = "f"
 472        };
 473/*
 474       lookup    hit  total   busy create
 475pages: ...... ...... ...... ...... ...... [...... ...... ...... ......]
 476locks: ...... ...... ...... ...... ...... [...... ...... ...... ...... ......]
 477  env: ...... ...... ...... ...... ......
 478 */
 479        lu_site_stats_print(&site->cs_lu, m);
 480        cache_stats_print(&site->cs_pages, m, 1);
 481        seq_printf(m, " [");
 482        for (i = 0; i < ARRAY_SIZE(site->cs_pages_state); ++i)
 483                seq_printf(m, "%s: %u ", pstate[i],
 484                                atomic_read(&site->cs_pages_state[i]));
 485        seq_printf(m, "]\n");
 486        cache_stats_print(&site->cs_locks, m, 0);
 487        seq_printf(m, " [");
 488        for (i = 0; i < ARRAY_SIZE(site->cs_locks_state); ++i)
 489                seq_printf(m, "%s: %u ", lstate[i],
 490                                atomic_read(&site->cs_locks_state[i]));
 491        seq_printf(m, "]\n");
 492        cache_stats_print(&cl_env_stats, m, 0);
 493        seq_printf(m, "\n");
 494        return 0;
 495}
 496EXPORT_SYMBOL(cl_site_stats_print);
 497
 498/*****************************************************************************
 499 *
 500 * lu_env handling on client.
 501 *
 502 */
 503
 504/**
 505 * The most efficient way is to store cl_env pointer in task specific
 506 * structures. On Linux, it wont' be easy to use task_struct->journal_info
 507 * because Lustre code may call into other fs which has certain assumptions
 508 * about journal_info. Currently following fields in task_struct are identified
 509 * can be used for this purpose:
 510 *  - cl_env: for liblustre.
 511 *  - tux_info: ony on RedHat kernel.
 512 *  - ...
 513 * \note As long as we use task_struct to store cl_env, we assume that once
 514 * called into Lustre, we'll never call into the other part of the kernel
 515 * which will use those fields in task_struct without explicitly exiting
 516 * Lustre.
 517 *
 518 * If there's no space in task_struct is available, hash will be used.
 519 * bz20044, bz22683.
 520 */
 521
 522struct cl_env {
 523        void         *ce_magic;
 524        struct lu_env     ce_lu;
 525        struct lu_context ce_ses;
 526
 527        /**
 528         * This allows cl_env to be entered into cl_env_hash which implements
 529         * the current thread -> client environment lookup.
 530         */
 531        struct hlist_node  ce_node;
 532        /**
 533         * Owner for the current cl_env.
 534         *
 535         * If LL_TASK_CL_ENV is defined, this point to the owning current,
 536         * only for debugging purpose ;
 537         * Otherwise hash is used, and this is the key for cfs_hash.
 538         * Now current thread pid is stored. Note using thread pointer would
 539         * lead to unbalanced hash because of its specific allocation locality
 540         * and could be varied for different platforms and OSes, even different
 541         * OS versions.
 542         */
 543        void         *ce_owner;
 544
 545        /*
 546         * Linkage into global list of all client environments. Used for
 547         * garbage collection.
 548         */
 549        struct list_head        ce_linkage;
 550        /*
 551         *
 552         */
 553        int            ce_ref;
 554        /*
 555         * Debugging field: address of the caller who made original
 556         * allocation.
 557         */
 558        void         *ce_debug;
 559};
 560
 561#define CL_ENV_INC(counter)
 562#define CL_ENV_DEC(counter)
 563
 564static void cl_env_init0(struct cl_env *cle, void *debug)
 565{
 566        LASSERT(cle->ce_ref == 0);
 567        LASSERT(cle->ce_magic == &cl_env_init0);
 568        LASSERT(cle->ce_debug == NULL && cle->ce_owner == NULL);
 569
 570        cle->ce_ref = 1;
 571        cle->ce_debug = debug;
 572        CL_ENV_INC(busy);
 573}
 574
 575
 576/*
 577 * The implementation of using hash table to connect cl_env and thread
 578 */
 579
 580static cfs_hash_t *cl_env_hash;
 581
 582static unsigned cl_env_hops_hash(cfs_hash_t *lh,
 583                                 const void *key, unsigned mask)
 584{
 585#if BITS_PER_LONG == 64
 586        return cfs_hash_u64_hash((__u64)key, mask);
 587#else
 588        return cfs_hash_u32_hash((__u32)key, mask);
 589#endif
 590}
 591
 592static void *cl_env_hops_obj(struct hlist_node *hn)
 593{
 594        struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node);
 595        LASSERT(cle->ce_magic == &cl_env_init0);
 596        return (void *)cle;
 597}
 598
 599static int cl_env_hops_keycmp(const void *key, struct hlist_node *hn)
 600{
 601        struct cl_env *cle = cl_env_hops_obj(hn);
 602
 603        LASSERT(cle->ce_owner != NULL);
 604        return (key == cle->ce_owner);
 605}
 606
 607static void cl_env_hops_noop(cfs_hash_t *hs, struct hlist_node *hn)
 608{
 609        struct cl_env *cle = hlist_entry(hn, struct cl_env, ce_node);
 610        LASSERT(cle->ce_magic == &cl_env_init0);
 611}
 612
 613static cfs_hash_ops_t cl_env_hops = {
 614        .hs_hash        = cl_env_hops_hash,
 615        .hs_key  = cl_env_hops_obj,
 616        .hs_keycmp      = cl_env_hops_keycmp,
 617        .hs_object      = cl_env_hops_obj,
 618        .hs_get  = cl_env_hops_noop,
 619        .hs_put_locked  = cl_env_hops_noop,
 620};
 621
 622static inline struct cl_env *cl_env_fetch(void)
 623{
 624        struct cl_env *cle;
 625
 626        cle = cfs_hash_lookup(cl_env_hash, (void *) (long) current->pid);
 627        LASSERT(ergo(cle, cle->ce_magic == &cl_env_init0));
 628        return cle;
 629}
 630
 631static inline void cl_env_attach(struct cl_env *cle)
 632{
 633        if (cle) {
 634                int rc;
 635
 636                LASSERT(cle->ce_owner == NULL);
 637                cle->ce_owner = (void *) (long) current->pid;
 638                rc = cfs_hash_add_unique(cl_env_hash, cle->ce_owner,
 639                                         &cle->ce_node);
 640                LASSERT(rc == 0);
 641        }
 642}
 643
 644static inline void cl_env_do_detach(struct cl_env *cle)
 645{
 646        void *cookie;
 647
 648        LASSERT(cle->ce_owner == (void *) (long) current->pid);
 649        cookie = cfs_hash_del(cl_env_hash, cle->ce_owner,
 650                              &cle->ce_node);
 651        LASSERT(cookie == cle);
 652        cle->ce_owner = NULL;
 653}
 654
 655static int cl_env_store_init(void) {
 656        cl_env_hash = cfs_hash_create("cl_env",
 657                                      HASH_CL_ENV_BITS, HASH_CL_ENV_BITS,
 658                                      HASH_CL_ENV_BKT_BITS, 0,
 659                                      CFS_HASH_MIN_THETA,
 660                                      CFS_HASH_MAX_THETA,
 661                                      &cl_env_hops,
 662                                      CFS_HASH_RW_BKTLOCK);
 663        return cl_env_hash != NULL ? 0 :-ENOMEM;
 664}
 665
 666static void cl_env_store_fini(void) {
 667        cfs_hash_putref(cl_env_hash);
 668}
 669
 670
 671static inline struct cl_env *cl_env_detach(struct cl_env *cle)
 672{
 673        if (cle == NULL)
 674                cle = cl_env_fetch();
 675
 676        if (cle && cle->ce_owner)
 677                cl_env_do_detach(cle);
 678
 679        return cle;
 680}
 681
 682static struct lu_env *cl_env_new(__u32 ctx_tags, __u32 ses_tags, void *debug)
 683{
 684        struct lu_env *env;
 685        struct cl_env *cle;
 686
 687        OBD_SLAB_ALLOC_PTR_GFP(cle, cl_env_kmem, __GFP_IO);
 688        if (cle != NULL) {
 689                int rc;
 690
 691                INIT_LIST_HEAD(&cle->ce_linkage);
 692                cle->ce_magic = &cl_env_init0;
 693                env = &cle->ce_lu;
 694                rc = lu_env_init(env, LCT_CL_THREAD|ctx_tags);
 695                if (rc == 0) {
 696                        rc = lu_context_init(&cle->ce_ses,
 697                                             LCT_SESSION | ses_tags);
 698                        if (rc == 0) {
 699                                lu_context_enter(&cle->ce_ses);
 700                                env->le_ses = &cle->ce_ses;
 701                                cl_env_init0(cle, debug);
 702                        } else
 703                                lu_env_fini(env);
 704                }
 705                if (rc != 0) {
 706                        OBD_SLAB_FREE_PTR(cle, cl_env_kmem);
 707                        env = ERR_PTR(rc);
 708                } else {
 709                        CL_ENV_INC(create);
 710                        CL_ENV_INC(total);
 711                }
 712        } else
 713                env = ERR_PTR(-ENOMEM);
 714        return env;
 715}
 716
 717static void cl_env_fini(struct cl_env *cle)
 718{
 719        CL_ENV_DEC(total);
 720        lu_context_fini(&cle->ce_lu.le_ctx);
 721        lu_context_fini(&cle->ce_ses);
 722        OBD_SLAB_FREE_PTR(cle, cl_env_kmem);
 723}
 724
 725static inline struct cl_env *cl_env_container(struct lu_env *env)
 726{
 727        return container_of(env, struct cl_env, ce_lu);
 728}
 729
 730struct lu_env *cl_env_peek(int *refcheck)
 731{
 732        struct lu_env *env;
 733        struct cl_env *cle;
 734
 735        CL_ENV_INC(lookup);
 736
 737        /* check that we don't go far from untrusted pointer */
 738        CLASSERT(offsetof(struct cl_env, ce_magic) == 0);
 739
 740        env = NULL;
 741        cle = cl_env_fetch();
 742        if (cle != NULL) {
 743                CL_ENV_INC(hit);
 744                env = &cle->ce_lu;
 745                *refcheck = ++cle->ce_ref;
 746        }
 747        CDEBUG(D_OTHER, "%d@%p\n", cle ? cle->ce_ref : 0, cle);
 748        return env;
 749}
 750EXPORT_SYMBOL(cl_env_peek);
 751
 752/**
 753 * Returns lu_env: if there already is an environment associated with the
 754 * current thread, it is returned, otherwise, new environment is allocated.
 755 *
 756 * \param refcheck pointer to a counter used to detect environment leaks. In
 757 * the usual case cl_env_get() and cl_env_put() are called in the same lexical
 758 * scope and pointer to the same integer is passed as \a refcheck. This is
 759 * used to detect missed cl_env_put().
 760 *
 761 * \see cl_env_put()
 762 */
 763struct lu_env *cl_env_get(int *refcheck)
 764{
 765        struct lu_env *env;
 766
 767        env = cl_env_peek(refcheck);
 768        if (env == NULL) {
 769                env = cl_env_new(lu_context_tags_default,
 770                                 lu_session_tags_default,
 771                                 __builtin_return_address(0));
 772
 773                if (!IS_ERR(env)) {
 774                        struct cl_env *cle;
 775
 776                        cle = cl_env_container(env);
 777                        cl_env_attach(cle);
 778                        *refcheck = cle->ce_ref;
 779                        CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
 780                }
 781        }
 782        return env;
 783}
 784EXPORT_SYMBOL(cl_env_get);
 785
 786/**
 787 * Forces an allocation of a fresh environment with given tags.
 788 *
 789 * \see cl_env_get()
 790 */
 791struct lu_env *cl_env_alloc(int *refcheck, __u32 tags)
 792{
 793        struct lu_env *env;
 794
 795        LASSERT(cl_env_peek(refcheck) == NULL);
 796        env = cl_env_new(tags, tags, __builtin_return_address(0));
 797        if (!IS_ERR(env)) {
 798                struct cl_env *cle;
 799
 800                cle = cl_env_container(env);
 801                *refcheck = cle->ce_ref;
 802                CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
 803        }
 804        return env;
 805}
 806EXPORT_SYMBOL(cl_env_alloc);
 807
 808static void cl_env_exit(struct cl_env *cle)
 809{
 810        LASSERT(cle->ce_owner == NULL);
 811        lu_context_exit(&cle->ce_lu.le_ctx);
 812        lu_context_exit(&cle->ce_ses);
 813}
 814
 815/**
 816 * Release an environment.
 817 *
 818 * Decrement \a env reference counter. When counter drops to 0, nothing in
 819 * this thread is using environment and it is returned to the allocation
 820 * cache, or freed straight away, if cache is large enough.
 821 */
 822void cl_env_put(struct lu_env *env, int *refcheck)
 823{
 824        struct cl_env *cle;
 825
 826        cle = cl_env_container(env);
 827
 828        LASSERT(cle->ce_ref > 0);
 829        LASSERT(ergo(refcheck != NULL, cle->ce_ref == *refcheck));
 830
 831        CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
 832        if (--cle->ce_ref == 0) {
 833                CL_ENV_DEC(busy);
 834                cl_env_detach(cle);
 835                cle->ce_debug = NULL;
 836                cl_env_exit(cle);
 837                cl_env_fini(cle);
 838        }
 839}
 840EXPORT_SYMBOL(cl_env_put);
 841
 842/**
 843 * Declares a point of re-entrancy.
 844 *
 845 * \see cl_env_reexit()
 846 */
 847void *cl_env_reenter(void)
 848{
 849        return cl_env_detach(NULL);
 850}
 851EXPORT_SYMBOL(cl_env_reenter);
 852
 853/**
 854 * Exits re-entrancy.
 855 */
 856void cl_env_reexit(void *cookie)
 857{
 858        cl_env_detach(NULL);
 859        cl_env_attach(cookie);
 860}
 861EXPORT_SYMBOL(cl_env_reexit);
 862
 863/**
 864 * Setup user-supplied \a env as a current environment. This is to be used to
 865 * guaranteed that environment exists even when cl_env_get() fails. It is up
 866 * to user to ensure proper concurrency control.
 867 *
 868 * \see cl_env_unplant()
 869 */
 870void cl_env_implant(struct lu_env *env, int *refcheck)
 871{
 872        struct cl_env *cle = cl_env_container(env);
 873
 874        LASSERT(cle->ce_ref > 0);
 875
 876        cl_env_attach(cle);
 877        cl_env_get(refcheck);
 878        CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
 879}
 880EXPORT_SYMBOL(cl_env_implant);
 881
 882/**
 883 * Detach environment installed earlier by cl_env_implant().
 884 */
 885void cl_env_unplant(struct lu_env *env, int *refcheck)
 886{
 887        struct cl_env *cle = cl_env_container(env);
 888
 889        LASSERT(cle->ce_ref > 1);
 890
 891        CDEBUG(D_OTHER, "%d@%p\n", cle->ce_ref, cle);
 892
 893        cl_env_detach(cle);
 894        cl_env_put(env, refcheck);
 895}
 896EXPORT_SYMBOL(cl_env_unplant);
 897
 898struct lu_env *cl_env_nested_get(struct cl_env_nest *nest)
 899{
 900        struct lu_env *env;
 901
 902        nest->cen_cookie = NULL;
 903        env = cl_env_peek(&nest->cen_refcheck);
 904        if (env != NULL) {
 905                if (!cl_io_is_going(env))
 906                        return env;
 907                else {
 908                        cl_env_put(env, &nest->cen_refcheck);
 909                        nest->cen_cookie = cl_env_reenter();
 910                }
 911        }
 912        env = cl_env_get(&nest->cen_refcheck);
 913        if (IS_ERR(env)) {
 914                cl_env_reexit(nest->cen_cookie);
 915                return env;
 916        }
 917
 918        LASSERT(!cl_io_is_going(env));
 919        return env;
 920}
 921EXPORT_SYMBOL(cl_env_nested_get);
 922
 923void cl_env_nested_put(struct cl_env_nest *nest, struct lu_env *env)
 924{
 925        cl_env_put(env, &nest->cen_refcheck);
 926        cl_env_reexit(nest->cen_cookie);
 927}
 928EXPORT_SYMBOL(cl_env_nested_put);
 929
 930/**
 931 * Converts struct cl_attr to struct ost_lvb.
 932 *
 933 * \see cl_lvb2attr
 934 */
 935void cl_attr2lvb(struct ost_lvb *lvb, const struct cl_attr *attr)
 936{
 937        lvb->lvb_size   = attr->cat_size;
 938        lvb->lvb_mtime  = attr->cat_mtime;
 939        lvb->lvb_atime  = attr->cat_atime;
 940        lvb->lvb_ctime  = attr->cat_ctime;
 941        lvb->lvb_blocks = attr->cat_blocks;
 942}
 943EXPORT_SYMBOL(cl_attr2lvb);
 944
 945/**
 946 * Converts struct ost_lvb to struct cl_attr.
 947 *
 948 * \see cl_attr2lvb
 949 */
 950void cl_lvb2attr(struct cl_attr *attr, const struct ost_lvb *lvb)
 951{
 952        attr->cat_size   = lvb->lvb_size;
 953        attr->cat_mtime  = lvb->lvb_mtime;
 954        attr->cat_atime  = lvb->lvb_atime;
 955        attr->cat_ctime  = lvb->lvb_ctime;
 956        attr->cat_blocks = lvb->lvb_blocks;
 957}
 958EXPORT_SYMBOL(cl_lvb2attr);
 959
 960/*****************************************************************************
 961 *
 962 * Temporary prototype thing: mirror obd-devices into cl devices.
 963 *
 964 */
 965
 966struct cl_device *cl_type_setup(const struct lu_env *env, struct lu_site *site,
 967                                struct lu_device_type *ldt,
 968                                struct lu_device *next)
 969{
 970        const char       *typename;
 971        struct lu_device *d;
 972
 973        LASSERT(ldt != NULL);
 974
 975        typename = ldt->ldt_name;
 976        d = ldt->ldt_ops->ldto_device_alloc(env, ldt, NULL);
 977        if (!IS_ERR(d)) {
 978                int rc;
 979
 980                if (site != NULL)
 981                        d->ld_site = site;
 982                rc = ldt->ldt_ops->ldto_device_init(env, d, typename, next);
 983                if (rc == 0) {
 984                        lu_device_get(d);
 985                        lu_ref_add(&d->ld_reference,
 986                                   "lu-stack", &lu_site_init);
 987                } else {
 988                        ldt->ldt_ops->ldto_device_free(env, d);
 989                        CERROR("can't init device '%s', %d\n", typename, rc);
 990                        d = ERR_PTR(rc);
 991                }
 992        } else
 993                CERROR("Cannot allocate device: '%s'\n", typename);
 994        return lu2cl_dev(d);
 995}
 996EXPORT_SYMBOL(cl_type_setup);
 997
 998/**
 999 * Finalize device stack by calling lu_stack_fini().
1000 */
1001void cl_stack_fini(const struct lu_env *env, struct cl_device *cl)
1002{
1003        lu_stack_fini(env, cl2lu_dev(cl));
1004}
1005EXPORT_SYMBOL(cl_stack_fini);
1006
1007int  cl_lock_init(void);
1008void cl_lock_fini(void);
1009
1010int  cl_page_init(void);
1011void cl_page_fini(void);
1012
1013static struct lu_context_key cl_key;
1014
1015struct cl_thread_info *cl_env_info(const struct lu_env *env)
1016{
1017        return lu_context_key_get(&env->le_ctx, &cl_key);
1018}
1019
1020/* defines cl0_key_{init,fini}() */
1021LU_KEY_INIT_FINI(cl0, struct cl_thread_info);
1022
1023static void *cl_key_init(const struct lu_context *ctx,
1024                         struct lu_context_key *key)
1025{
1026        struct cl_thread_info *info;
1027
1028        info = cl0_key_init(ctx, key);
1029        if (!IS_ERR(info)) {
1030                int i;
1031
1032                for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i)
1033                        lu_ref_init(&info->clt_counters[i].ctc_locks_locked);
1034        }
1035        return info;
1036}
1037
1038static void cl_key_fini(const struct lu_context *ctx,
1039                        struct lu_context_key *key, void *data)
1040{
1041        struct cl_thread_info *info;
1042        int i;
1043
1044        info = data;
1045        for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i)
1046                lu_ref_fini(&info->clt_counters[i].ctc_locks_locked);
1047        cl0_key_fini(ctx, key, data);
1048}
1049
1050static void cl_key_exit(const struct lu_context *ctx,
1051                        struct lu_context_key *key, void *data)
1052{
1053        struct cl_thread_info *info = data;
1054        int i;
1055
1056        for (i = 0; i < ARRAY_SIZE(info->clt_counters); ++i) {
1057                LASSERT(info->clt_counters[i].ctc_nr_held == 0);
1058                LASSERT(info->clt_counters[i].ctc_nr_used == 0);
1059                LASSERT(info->clt_counters[i].ctc_nr_locks_acquired == 0);
1060                LASSERT(info->clt_counters[i].ctc_nr_locks_locked == 0);
1061                lu_ref_fini(&info->clt_counters[i].ctc_locks_locked);
1062                lu_ref_init(&info->clt_counters[i].ctc_locks_locked);
1063        }
1064}
1065
1066static struct lu_context_key cl_key = {
1067        .lct_tags = LCT_CL_THREAD,
1068        .lct_init = cl_key_init,
1069        .lct_fini = cl_key_fini,
1070        .lct_exit = cl_key_exit
1071};
1072
1073static struct lu_kmem_descr cl_object_caches[] = {
1074        {
1075                .ckd_cache = &cl_env_kmem,
1076                .ckd_name  = "cl_env_kmem",
1077                .ckd_size  = sizeof (struct cl_env)
1078        },
1079        {
1080                .ckd_cache = NULL
1081        }
1082};
1083
1084/**
1085 * Global initialization of cl-data. Create kmem caches, register
1086 * lu_context_key's, etc.
1087 *
1088 * \see cl_global_fini()
1089 */
1090int cl_global_init(void)
1091{
1092        int result;
1093
1094        result = cl_env_store_init();
1095        if (result)
1096                return result;
1097
1098        result = lu_kmem_init(cl_object_caches);
1099        if (result)
1100                goto out_store;
1101
1102        LU_CONTEXT_KEY_INIT(&cl_key);
1103        result = lu_context_key_register(&cl_key);
1104        if (result)
1105                goto out_kmem;
1106
1107        result = cl_lock_init();
1108        if (result)
1109                goto out_context;
1110
1111        result = cl_page_init();
1112        if (result)
1113                goto out_lock;
1114
1115        return 0;
1116out_lock:
1117        cl_lock_fini();
1118out_context:
1119        lu_context_key_degister(&cl_key);
1120out_kmem:
1121        lu_kmem_fini(cl_object_caches);
1122out_store:
1123        cl_env_store_fini();
1124        return result;
1125}
1126
1127/**
1128 * Finalization of global cl-data. Dual to cl_global_init().
1129 */
1130void cl_global_fini(void)
1131{
1132        cl_lock_fini();
1133        cl_page_fini();
1134        lu_context_key_degister(&cl_key);
1135        lu_kmem_fini(cl_object_caches);
1136        cl_env_store_fini();
1137}
1138