lustre/obdclass/cl_page.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.gnu.org/licenses/gpl-2.0.html
  19  *
  20  * GPL HEADER END
  21  */
  22 /*
  23  * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Use is subject to license terms.
  25  *
  26  * Copyright (c) 2011, 2017, Intel Corporation.
  27  */
  28 /*
  29  * This file is part of Lustre, http://www.lustre.org/
  30  *
  31  * Client Lustre Page.
  32  *
  33  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  34  *   Author: Jinshan Xiong <jinshan.xiong@intel.com>
  35  */
  36
  37 #define DEBUG_SUBSYSTEM S_CLASS
  38
  39 #include <linux/list.h>
  40 #include <libcfs/libcfs.h>
  41 #include <obd_class.h>
  42 #include <obd_support.h>
  43
  44 #include <cl_object.h>
  45 #include "cl_internal.h"
  46
  47 static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg);
  48 static DEFINE_MUTEX(cl_page_kmem_mutex);
  49
  50 #ifdef LIBCFS_DEBUG
  51 # define PASSERT(env, page, expr)                                       \
  52   do {                                                                    \
  53           if (unlikely(!(expr))) {                                      \
  54                   CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n");    \
  55                   LASSERT(0);                                           \
  56           }                                                             \
  57   } while (0)
  58 #else /* !LIBCFS_DEBUG */
  59 # define PASSERT(env, page, exp) \
  60         ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
  61 #endif /* !LIBCFS_DEBUG */
  62
  63 #ifdef CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK
  64 # define PINVRNT(env, page, expr)                                       \
  65   do {                                                                    \
  66           if (unlikely(!(expr))) {                                      \
  67                   CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n");    \
  68                   LINVRNT(0);                                           \
  69           }                                                             \
  70   } while (0)
  71 #else /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */
  72 # define PINVRNT(env, page, exp) \
  73          ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
  74 #endif /* !CONFIG_LUSTRE_DEBUG_EXPENSIVE_CHECK */
  75
  76 /* Disable page statistic by default due to huge performance penalty. */
  77 static void cs_page_inc(const struct cl_object *obj,
  78                         enum cache_stats_item item)
  79 {
  80 #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
  81         atomic_inc(&cl_object_site(obj)->cs_pages.cs_stats[item]);
  82 #endif
  83 }
  84
  85 static void cs_page_dec(const struct cl_object *obj,
  86                         enum cache_stats_item item)
  87 {
  88 #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
  89         atomic_dec(&cl_object_site(obj)->cs_pages.cs_stats[item]);
  90 #endif
  91 }
  92
  93 static void cs_pagestate_inc(const struct cl_object *obj,
  94                              enum cl_page_state state)
  95 {
  96 #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
  97         atomic_inc(&cl_object_site(obj)->cs_pages_state[state]);
  98 #endif
  99 }
 100
 101 static void cs_pagestate_dec(const struct cl_object *obj,
 102                               enum cl_page_state state)
 103 {
 104 #ifdef CONFIG_DEBUG_PAGESTATE_TRACKING
 105         atomic_dec(&cl_object_site(obj)->cs_pages_state[state]);
 106 #endif
 107 }
 108
 109 /**
 110  * Internal version of cl_page_get().
 111  *
 112  * This function can be used to obtain initial reference to previously
 113  * unreferenced cached object. It can be called only if concurrent page
 114  * reclamation is somehow prevented, e.g., by keeping a lock on a VM page,
 115  * associated with \a page.
 116  *
 117  * Use with care! Not exported.
 118  */
 119 static void cl_page_get_trust(struct cl_page *page)
 120 {
 121         LASSERT(atomic_read(&page->cp_ref) > 0);
 122         atomic_inc(&page->cp_ref);
 123 }
 124
 125 static struct cl_page_slice *
 126 cl_page_slice_get(const struct cl_page *cl_page, int index)
 127 {
 128         if (index < 0 || index >= cl_page->cp_layer_count)
 129                 return NULL;
 130
 131         /* To get the cp_layer_offset values fit under 256 bytes, we
 132          * use the offset beyond the end of struct cl_page.
 133          */
 134         return (struct cl_page_slice *)((char *)cl_page + sizeof(*cl_page) +
 135                                         cl_page->cp_layer_offset[index]);
 136 }
 137
 138 #define cl_page_slice_for_each(cl_page, slice, i)               \
 139         for (i = 0, slice = cl_page_slice_get(cl_page, 0);      \
 140              i < (cl_page)->cp_layer_count;                     \
 141              slice = cl_page_slice_get(cl_page, ++i))
 142
 143 #define cl_page_slice_for_each_reverse(cl_page, slice, i)       \
 144         for (i = (cl_page)->cp_layer_count - 1,                 \
 145              slice = cl_page_slice_get(cl_page, i); i >= 0;     \
 146              slice = cl_page_slice_get(cl_page, --i))
 147
 148 /**
 149  * Returns a slice within a cl_page, corresponding to the given layer in the
 150  * device stack.
 151  *
 152  * \see cl_lock_at()
 153  */
 154 static const struct cl_page_slice *
 155 cl_page_at_trusted(const struct cl_page *cl_page,
 156                    const struct lu_device_type *dtype)
 157 {
 158         const struct cl_page_slice *slice;
 159         int i;
 160
 161         ENTRY;
 162
 163         cl_page_slice_for_each(cl_page, slice, i) {
 164                 if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype)
 165                         RETURN(slice);
 166         }
 167
 168         RETURN(NULL);
 169 }
 170
 171 static void __cl_page_free(struct cl_page *cl_page, unsigned short bufsize)
 172 {
 173         int index = cl_page->cp_kmem_index;
 174
 175         if (index >= 0) {
 176                 LASSERT(index < ARRAY_SIZE(cl_page_kmem_array));
 177                 LASSERT(cl_page_kmem_size_array[index] == bufsize);
 178                 OBD_SLAB_FREE(cl_page, cl_page_kmem_array[index], bufsize);
 179         } else {
 180                 OBD_FREE(cl_page, bufsize);
 181         }
 182 }
 183
 184 static void cl_page_free(const struct lu_env *env, struct cl_page *cl_page,
 185                          struct pagevec *pvec)
 186 {
 187         struct cl_object *obj  = cl_page->cp_obj;
 188         unsigned short bufsize = cl_object_header(obj)->coh_page_bufsize;
 189         struct cl_page_slice *slice;
 190         int i;
 191
 192         ENTRY;
 193         PASSERT(env, cl_page, list_empty(&cl_page->cp_batch));
 194         PASSERT(env, cl_page, cl_page->cp_owner == NULL);
 195         PASSERT(env, cl_page, cl_page->cp_state == CPS_FREEING);
 196
 197         cl_page_slice_for_each(cl_page, slice, i) {
 198                 if (unlikely(slice->cpl_ops->cpo_fini != NULL))
 199                         slice->cpl_ops->cpo_fini(env, slice, pvec);
 200         }
 201         cl_page->cp_layer_count = 0;
 202         cs_page_dec(obj, CS_total);
 203         cs_pagestate_dec(obj, cl_page->cp_state);
 204         lu_object_ref_del_at(&obj->co_lu, &cl_page->cp_obj_ref,
 205                              "cl_page", cl_page);
 206         cl_object_put(env, obj);
 207         lu_ref_fini(&cl_page->cp_reference);
 208         __cl_page_free(cl_page, bufsize);
 209         EXIT;
 210 }
 211
 212 static struct cl_page *__cl_page_alloc(struct cl_object *o)
 213 {
 214         int i = 0;
 215         struct cl_page *cl_page = NULL;
 216         unsigned short bufsize = cl_object_header(o)->coh_page_bufsize;
 217
 218         if (OBD_FAIL_CHECK(OBD_FAIL_LLITE_PAGE_ALLOC))
 219                 return NULL;
 220
 221 check:
 222         /* the number of entries in cl_page_kmem_array is expected to
 223          * only be 2-3 entries, so the lookup overhead should be low.
 224          */
 225         for ( ; i < ARRAY_SIZE(cl_page_kmem_array); i++) {
 226                 if (smp_load_acquire(&cl_page_kmem_size_array[i])
 227                     == bufsize) {
 228                         OBD_SLAB_ALLOC_GFP(cl_page, cl_page_kmem_array[i],
 229                                            bufsize, GFP_NOFS);
 230                         if (cl_page)
 231                                 cl_page->cp_kmem_index = i;
 232                         return cl_page;
 233                 }
 234                 if (cl_page_kmem_size_array[i] == 0)
 235                         break;
 236         }
 237
 238         if (i < ARRAY_SIZE(cl_page_kmem_array)) {
 239                 char cache_name[32];
 240
 241                 mutex_lock(&cl_page_kmem_mutex);
 242                 if (cl_page_kmem_size_array[i]) {
 243                         mutex_unlock(&cl_page_kmem_mutex);
 244                         goto check;
 245                 }
 246                 snprintf(cache_name, sizeof(cache_name),
 247                          "cl_page_kmem-%u", bufsize);
 248                 cl_page_kmem_array[i] =
 249                         kmem_cache_create(cache_name, bufsize,
 250                                           0, 0, NULL);
 251                 if (cl_page_kmem_array[i] == NULL) {
 252                         mutex_unlock(&cl_page_kmem_mutex);
 253                         return NULL;
 254                 }
 255                 smp_store_release(&cl_page_kmem_size_array[i],
 256                                   bufsize);
 257                 mutex_unlock(&cl_page_kmem_mutex);
 258                 goto check;
 259         } else {
 260                 OBD_ALLOC_GFP(cl_page, bufsize, GFP_NOFS);
 261                 if (cl_page)
 262                         cl_page->cp_kmem_index = -1;
 263         }
 264
 265         return cl_page;
 266 }
 267
 268 struct cl_page *cl_page_alloc(const struct lu_env *env, struct cl_object *o,
 269                               pgoff_t ind, struct page *vmpage,
 270                               enum cl_page_type type)
 271 {
 272         struct cl_page *cl_page;
 273         struct cl_object *head;
 274
 275         ENTRY;
 276
 277         cl_page = __cl_page_alloc(o);
 278         if (cl_page != NULL) {
 279                 int result = 0;
 280
 281                 /*
 282                  * Please fix cl_page:cp_state/type declaration if
 283                  * these assertions fail in the future.
 284                  */
 285                 BUILD_BUG_ON((1 << CP_STATE_BITS) < CPS_NR); /* cp_state */
 286                 BUILD_BUG_ON((1 << CP_TYPE_BITS) < CPT_NR); /* cp_type */
 287                 atomic_set(&cl_page->cp_ref, 1);
 288                 cl_page->cp_obj = o;
 289                 cl_object_get(o);
 290                 lu_object_ref_add_at(&o->co_lu, &cl_page->cp_obj_ref,
 291                                      "cl_page", cl_page);
 292                 cl_page->cp_vmpage = vmpage;
 293                 cl_page->cp_state = CPS_CACHED;
 294                 cl_page->cp_type = type;
 295                 cl_page->cp_inode = NULL;
 296                 INIT_LIST_HEAD(&cl_page->cp_batch);
 297                 lu_ref_init(&cl_page->cp_reference);
 298                 head = o;
 299                 cl_object_for_each(o, head) {
 300                         if (o->co_ops->coo_page_init != NULL) {
 301                                 result = o->co_ops->coo_page_init(env, o,
 302                                                         cl_page, ind);
 303                                 if (result != 0) {
 304                                         cl_page_delete0(env, cl_page);
 305                                         cl_page_free(env, cl_page, NULL);
 306                                         cl_page = ERR_PTR(result);
 307                                         break;
 308                                 }
 309                         }
 310                 }
 311                 if (result == 0) {
 312                         cs_page_inc(o, CS_total);
 313                         cs_page_inc(o, CS_create);
 314                         cs_pagestate_dec(o, CPS_CACHED);
 315                 }
 316         } else {
 317                 cl_page = ERR_PTR(-ENOMEM);
 318         }
 319         RETURN(cl_page);
 320 }
 321
 322 /**
 323  * Returns a cl_page with index \a idx at the object \a o, and associated with
 324  * the VM page \a vmpage.
 325  *
 326  * This is the main entry point into the cl_page caching interface. First, a
 327  * cache (implemented as a per-object radix tree) is consulted. If page is
 328  * found there, it is returned immediately. Otherwise new page is allocated
 329  * and returned. In any case, additional reference to page is acquired.
 330  *
 331  * \see cl_object_find(), cl_lock_find()
 332  */
 333 struct cl_page *cl_page_find(const struct lu_env *env,
 334                              struct cl_object *o,
 335                              pgoff_t idx, struct page *vmpage,
 336                              enum cl_page_type type)
 337 {
 338         struct cl_page          *page = NULL;
 339         struct cl_object_header *hdr;
 340
 341         LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT);
 342         might_sleep();
 343
 344         ENTRY;
 345
 346         hdr = cl_object_header(o);
 347         cs_page_inc(o, CS_lookup);
 348
 349         CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n",
 350                idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type);
 351         /* fast path. */
 352         if (type == CPT_CACHEABLE) {
 353                 /* vmpage lock is used to protect the child/parent
 354                  * relationship */
 355                 LASSERT(PageLocked(vmpage));
 356                 /*
 357                  * cl_vmpage_page() can be called here without any locks as
 358                  *
 359                  *     - "vmpage" is locked (which prevents ->private from
 360                  *       concurrent updates), and
 361                  *
 362                  *     - "o" cannot be destroyed while current thread holds a
 363                  *       reference on it.
 364                  */
 365                 page = cl_vmpage_page(vmpage, o);
 366                 if (page != NULL) {
 367                         cs_page_inc(o, CS_hit);
 368                         RETURN(page);
 369                 }
 370         }
 371
 372         /* allocate and initialize cl_page */
 373         page = cl_page_alloc(env, o, idx, vmpage, type);
 374         RETURN(page);
 375 }
 376 EXPORT_SYMBOL(cl_page_find);
 377
 378 static inline int cl_page_invariant(const struct cl_page *pg)
 379 {
 380         return cl_page_in_use_noref(pg);
 381 }
 382
 383 static void cl_page_state_set0(const struct lu_env *env,
 384                                struct cl_page *cl_page,
 385                                enum cl_page_state state)
 386 {
 387         enum cl_page_state old;
 388
 389         /*
 390          * Matrix of allowed state transitions [old][new], for sanity
 391          * checking.
 392          */
 393         static const int allowed_transitions[CPS_NR][CPS_NR] = {
 394                 [CPS_CACHED] = {
 395                         [CPS_CACHED]  = 0,
 396                         [CPS_OWNED]   = 1, /* io finds existing cached page */
 397                         [CPS_PAGEIN]  = 0,
 398                         [CPS_PAGEOUT] = 1, /* write-out from the cache */
 399                         [CPS_FREEING] = 1, /* eviction on the memory pressure */
 400                 },
 401                 [CPS_OWNED] = {
 402                         [CPS_CACHED]  = 1, /* release to the cache */
 403                         [CPS_OWNED]   = 0,
 404                         [CPS_PAGEIN]  = 1, /* start read immediately */
 405                         [CPS_PAGEOUT] = 1, /* start write immediately */
 406                         [CPS_FREEING] = 1, /* lock invalidation or truncate */
 407                 },
 408                 [CPS_PAGEIN] = {
 409                         [CPS_CACHED]  = 1, /* io completion */
 410                         [CPS_OWNED]   = 0,
 411                         [CPS_PAGEIN]  = 0,
 412                         [CPS_PAGEOUT] = 0,
 413                         [CPS_FREEING] = 0,
 414                 },
 415                 [CPS_PAGEOUT] = {
 416                         [CPS_CACHED]  = 1, /* io completion */
 417                         [CPS_OWNED]   = 0,
 418                         [CPS_PAGEIN]  = 0,
 419                         [CPS_PAGEOUT] = 0,
 420                         [CPS_FREEING] = 0,
 421                 },
 422                 [CPS_FREEING] = {
 423                         [CPS_CACHED]  = 0,
 424                         [CPS_OWNED]   = 0,
 425                         [CPS_PAGEIN]  = 0,
 426                         [CPS_PAGEOUT] = 0,
 427                         [CPS_FREEING] = 0,
 428                 }
 429         };
 430
 431         ENTRY;
 432         old = cl_page->cp_state;
 433         PASSERT(env, cl_page, allowed_transitions[old][state]);
 434         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d -> %d\n", old, state);
 435         PASSERT(env, cl_page, cl_page->cp_state == old);
 436         PASSERT(env, cl_page, equi(state == CPS_OWNED,
 437                                    cl_page->cp_owner != NULL));
 438
 439         cs_pagestate_dec(cl_page->cp_obj, cl_page->cp_state);
 440         cs_pagestate_inc(cl_page->cp_obj, state);
 441         cl_page->cp_state = state;
 442         EXIT;
 443 }
 444
 445 static void cl_page_state_set(const struct lu_env *env,
 446                               struct cl_page *page, enum cl_page_state state)
 447 {
 448         cl_page_state_set0(env, page, state);
 449 }
 450
 451 /**
 452  * Acquires an additional reference to a page.
 453  *
 454  * This can be called only by caller already possessing a reference to \a
 455  * page.
 456  *
 457  * \see cl_object_get(), cl_lock_get().
 458  */
 459 void cl_page_get(struct cl_page *page)
 460 {
 461         ENTRY;
 462         cl_page_get_trust(page);
 463         EXIT;
 464 }
 465 EXPORT_SYMBOL(cl_page_get);
 466
 467 /**
 468  * Releases a reference to a page, use the pagevec to release the pages
 469  * in batch if provided.
 470  *
 471  * Users need to do a final pagevec_release() to release any trailing pages.
 472  */
 473 void cl_pagevec_put(const struct lu_env *env, struct cl_page *page,
 474                   struct pagevec *pvec)
 475 {
 476         ENTRY;
 477         CL_PAGE_HEADER(D_TRACE, env, page, "%d\n",
 478                        atomic_read(&page->cp_ref));
 479
 480         if (atomic_dec_and_test(&page->cp_ref)) {
 481                 LASSERT(page->cp_state == CPS_FREEING);
 482
 483                 LASSERT(atomic_read(&page->cp_ref) == 0);
 484                 PASSERT(env, page, page->cp_owner == NULL);
 485                 PASSERT(env, page, list_empty(&page->cp_batch));
 486                 /*
 487                  * Page is no longer reachable by other threads. Tear
 488                  * it down.
 489                  */
 490                 cl_page_free(env, page, pvec);
 491         }
 492
 493         EXIT;
 494 }
 495 EXPORT_SYMBOL(cl_pagevec_put);
 496
 497 /**
 498  * Releases a reference to a page, wrapper to cl_pagevec_put
 499  *
 500  * When last reference is released, page is returned to the cache, unless it
 501  * is in cl_page_state::CPS_FREEING state, in which case it is immediately
 502  * destroyed.
 503  *
 504  * \see cl_object_put(), cl_lock_put().
 505  */
 506 void cl_page_put(const struct lu_env *env, struct cl_page *page)
 507 {
 508         cl_pagevec_put(env, page, NULL);
 509 }
 510 EXPORT_SYMBOL(cl_page_put);
 511
 512 /**
 513  * Returns a cl_page associated with a VM page, and given cl_object.
 514  */
 515 struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj)
 516 {
 517         struct cl_page *page;
 518
 519         ENTRY;
 520         LASSERT(PageLocked(vmpage));
 521
 522         /*
 523          * NOTE: absence of races and liveness of data are guaranteed by page
 524          *       lock on a "vmpage". That works because object destruction has
 525          *       bottom-to-top pass.
 526          */
 527
 528         page = (struct cl_page *)vmpage->private;
 529         if (page != NULL) {
 530                 cl_page_get_trust(page);
 531                 LASSERT(page->cp_type == CPT_CACHEABLE);
 532         }
 533         RETURN(page);
 534 }
 535 EXPORT_SYMBOL(cl_vmpage_page);
 536
 537 const struct cl_page_slice *cl_page_at(const struct cl_page *page,
 538                                        const struct lu_device_type *dtype)
 539 {
 540         return cl_page_at_trusted(page, dtype);
 541 }
 542 EXPORT_SYMBOL(cl_page_at);
 543
 544 static void cl_page_owner_clear(struct cl_page *page)
 545 {
 546         ENTRY;
 547         if (page->cp_owner != NULL) {
 548                 LASSERT(page->cp_owner->ci_owned_nr > 0);
 549                 page->cp_owner->ci_owned_nr--;
 550                 page->cp_owner = NULL;
 551         }
 552         EXIT;
 553 }
 554
 555 static void cl_page_owner_set(struct cl_page *page)
 556 {
 557         ENTRY;
 558         LASSERT(page->cp_owner != NULL);
 559         page->cp_owner->ci_owned_nr++;
 560         EXIT;
 561 }
 562
 563 void cl_page_disown0(const struct lu_env *env,
 564                      struct cl_io *io, struct cl_page *cl_page)
 565 {
 566         const struct cl_page_slice *slice;
 567         enum cl_page_state state;
 568         int i;
 569
 570         ENTRY;
 571         state = cl_page->cp_state;
 572         PINVRNT(env, cl_page, state == CPS_OWNED ||
 573                 state == CPS_FREEING);
 574         PINVRNT(env, cl_page, cl_page_invariant(cl_page) ||
 575                 state == CPS_FREEING);
 576         cl_page_owner_clear(cl_page);
 577
 578         if (state == CPS_OWNED)
 579                 cl_page_state_set(env, cl_page, CPS_CACHED);
 580         /*
 581          * Completion call-backs are executed in the bottom-up order, so that
 582          * uppermost layer (llite), responsible for VFS/VM interaction runs
 583          * last and can release locks safely.
 584          */
 585         cl_page_slice_for_each_reverse(cl_page, slice, i) {
 586                 if (slice->cpl_ops->cpo_disown != NULL)
 587                         (*slice->cpl_ops->cpo_disown)(env, slice, io);
 588         }
 589
 590         EXIT;
 591 }
 592
 593 /**
 594  * returns true, iff page is owned by the given io.
 595  */
 596 int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io)
 597 {
 598         struct cl_io *top = cl_io_top((struct cl_io *)io);
 599         LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj));
 600         ENTRY;
 601         RETURN(pg->cp_state == CPS_OWNED && pg->cp_owner == top);
 602 }
 603 EXPORT_SYMBOL(cl_page_is_owned);
 604
 605 /**
 606  * Try to own a page by IO.
 607  *
 608  * Waits until page is in cl_page_state::CPS_CACHED state, and then switch it
 609  * into cl_page_state::CPS_OWNED state.
 610  *
 611  * \pre  !cl_page_is_owned(cl_page, io)
 612  * \post result == 0 iff cl_page_is_owned(cl_page, io)
 613  *
 614  * \retval 0   success
 615  *
 616  * \retval -ve failure, e.g., cl_page was destroyed (and landed in
 617  *             cl_page_state::CPS_FREEING instead of cl_page_state::CPS_CACHED).
 618  *             or, page was owned by another thread, or in IO.
 619  *
 620  * \see cl_page_disown()
 621  * \see cl_page_operations::cpo_own()
 622  * \see cl_page_own_try()
 623  * \see cl_page_own
 624  */
 625 static int cl_page_own0(const struct lu_env *env, struct cl_io *io,
 626                         struct cl_page *cl_page, int nonblock)
 627 {
 628         const struct cl_page_slice *slice;
 629         int result = 0;
 630         int i;
 631
 632         ENTRY;
 633         PINVRNT(env, cl_page, !cl_page_is_owned(cl_page, io));
 634         io = cl_io_top(io);
 635
 636         if (cl_page->cp_state == CPS_FREEING) {
 637                 result = -ENOENT;
 638                 goto out;
 639         }
 640
 641         cl_page_slice_for_each(cl_page, slice, i) {
 642                 if (slice->cpl_ops->cpo_own)
 643                         result = (*slice->cpl_ops->cpo_own)(env, slice,
 644                                                             io, nonblock);
 645                 if (result != 0)
 646                         break;
 647         }
 648         if (result > 0)
 649                 result = 0;
 650
 651         if (result == 0) {
 652                 PASSERT(env, cl_page, cl_page->cp_owner == NULL);
 653                 cl_page->cp_owner = cl_io_top(io);
 654                 cl_page_owner_set(cl_page);
 655                 if (cl_page->cp_state != CPS_FREEING) {
 656                         cl_page_state_set(env, cl_page, CPS_OWNED);
 657                 } else {
 658                         cl_page_disown0(env, io, cl_page);
 659                         result = -ENOENT;
 660                 }
 661         }
 662
 663 out:
 664         PINVRNT(env, cl_page, ergo(result == 0,
 665                 cl_page_invariant(cl_page)));
 666         RETURN(result);
 667 }
 668
 669 /**
 670  * Own a page, might be blocked.
 671  *
 672  * \see cl_page_own0()
 673  */
 674 int cl_page_own(const struct lu_env *env, struct cl_io *io, struct cl_page *pg)
 675 {
 676         return cl_page_own0(env, io, pg, 0);
 677 }
 678 EXPORT_SYMBOL(cl_page_own);
 679
 680 /**
 681  * Nonblock version of cl_page_own().
 682  *
 683  * \see cl_page_own0()
 684  */
 685 int cl_page_own_try(const struct lu_env *env, struct cl_io *io,
 686                     struct cl_page *pg)
 687 {
 688         return cl_page_own0(env, io, pg, 1);
 689 }
 690 EXPORT_SYMBOL(cl_page_own_try);
 691
 692
 693 /**
 694  * Assume page ownership.
 695  *
 696  * Called when page is already locked by the hosting VM.
 697  *
 698  * \pre !cl_page_is_owned(cl_page, io)
 699  * \post cl_page_is_owned(cl_page, io)
 700  *
 701  * \see cl_page_operations::cpo_assume()
 702  */
 703 void cl_page_assume(const struct lu_env *env,
 704                     struct cl_io *io, struct cl_page *cl_page)
 705 {
 706         const struct cl_page_slice *slice;
 707         int i;
 708
 709         ENTRY;
 710
 711         PINVRNT(env, cl_page,
 712                 cl_object_same(cl_page->cp_obj, io->ci_obj));
 713         io = cl_io_top(io);
 714
 715         cl_page_slice_for_each(cl_page, slice, i) {
 716                 if (slice->cpl_ops->cpo_assume != NULL)
 717                         (*slice->cpl_ops->cpo_assume)(env, slice, io);
 718         }
 719
 720         PASSERT(env, cl_page, cl_page->cp_owner == NULL);
 721         cl_page->cp_owner = cl_io_top(io);
 722         cl_page_owner_set(cl_page);
 723         cl_page_state_set(env, cl_page, CPS_OWNED);
 724         EXIT;
 725 }
 726 EXPORT_SYMBOL(cl_page_assume);
 727
 728 /**
 729  * Releases page ownership without unlocking the page.
 730  *
 731  * Moves cl_page into cl_page_state::CPS_CACHED without releasing a lock
 732  * on the underlying VM page (as VM is supposed to do this itself).
 733  *
 734  * \pre   cl_page_is_owned(cl_page, io)
 735  * \post !cl_page_is_owned(cl_page, io)
 736  *
 737  * \see cl_page_assume()
 738  */
 739 void cl_page_unassume(const struct lu_env *env,
 740                       struct cl_io *io, struct cl_page *cl_page)
 741 {
 742         const struct cl_page_slice *slice;
 743         int i;
 744
 745         ENTRY;
 746         PINVRNT(env, cl_page, cl_page_is_owned(cl_page, io));
 747         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
 748
 749         io = cl_io_top(io);
 750         cl_page_owner_clear(cl_page);
 751         cl_page_state_set(env, cl_page, CPS_CACHED);
 752
 753         cl_page_slice_for_each_reverse(cl_page, slice, i) {
 754                 if (slice->cpl_ops->cpo_unassume != NULL)
 755                         (*slice->cpl_ops->cpo_unassume)(env, slice, io);
 756         }
 757
 758         EXIT;
 759 }
 760 EXPORT_SYMBOL(cl_page_unassume);
 761
 762 /**
 763  * Releases page ownership.
 764  *
 765  * Moves page into cl_page_state::CPS_CACHED.
 766  *
 767  * \pre   cl_page_is_owned(pg, io)
 768  * \post !cl_page_is_owned(pg, io)
 769  *
 770  * \see cl_page_own()
 771  * \see cl_page_operations::cpo_disown()
 772  */
 773 void cl_page_disown(const struct lu_env *env,
 774                     struct cl_io *io, struct cl_page *pg)
 775 {
 776         PINVRNT(env, pg, cl_page_is_owned(pg, io) ||
 777                 pg->cp_state == CPS_FREEING);
 778
 779         ENTRY;
 780         io = cl_io_top(io);
 781         cl_page_disown0(env, io, pg);
 782         EXIT;
 783 }
 784 EXPORT_SYMBOL(cl_page_disown);
 785
 786 /**
 787  * Called when cl_page is to be removed from the object, e.g.,
 788  * as a result of truncate.
 789  *
 790  * Calls cl_page_operations::cpo_discard() top-to-bottom.
 791  *
 792  * \pre cl_page_is_owned(cl_page, io)
 793  *
 794  * \see cl_page_operations::cpo_discard()
 795  */
 796 void cl_page_discard(const struct lu_env *env,
 797                      struct cl_io *io, struct cl_page *cl_page)
 798 {
 799         const struct cl_page_slice *slice;
 800         int i;
 801
 802         PINVRNT(env, cl_page, cl_page_is_owned(cl_page, io));
 803         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
 804
 805         cl_page_slice_for_each(cl_page, slice, i) {
 806                 if (slice->cpl_ops->cpo_discard != NULL)
 807                         (*slice->cpl_ops->cpo_discard)(env, slice, io);
 808         }
 809 }
 810 EXPORT_SYMBOL(cl_page_discard);
 811
 812 /**
 813  * Version of cl_page_delete() that can be called for not fully constructed
 814  * cl_pages, e.g. in an error handling cl_page_find()->cl_page_delete0()
 815  * path. Doesn't check cl_page invariant.
 816  */
 817 static void cl_page_delete0(const struct lu_env *env,
 818                             struct cl_page *cl_page)
 819 {
 820         const struct cl_page_slice *slice;
 821         int i;
 822
 823         ENTRY;
 824
 825         PASSERT(env, cl_page, cl_page->cp_state != CPS_FREEING);
 826
 827         /*
 828          * Severe all ways to obtain new pointers to @pg.
 829          */
 830         cl_page_owner_clear(cl_page);
 831         cl_page_state_set0(env, cl_page, CPS_FREEING);
 832
 833         cl_page_slice_for_each_reverse(cl_page, slice, i) {
 834                 if (slice->cpl_ops->cpo_delete != NULL)
 835                         (*slice->cpl_ops->cpo_delete)(env, slice);
 836         }
 837
 838         EXIT;
 839 }
 840
 841 /**
 842  * Called when a decision is made to throw page out of memory.
 843  *
 844  * Notifies all layers about page destruction by calling
 845  * cl_page_operations::cpo_delete() method top-to-bottom.
 846  *
 847  * Moves page into cl_page_state::CPS_FREEING state (this is the only place
 848  * where transition to this state happens).
 849  *
 850  * Eliminates all venues through which new references to the page can be
 851  * obtained:
 852  *
 853  *     - removes page from the radix trees,
 854  *
 855  *     - breaks linkage from VM page to cl_page.
 856  *
 857  * Once page reaches cl_page_state::CPS_FREEING, all remaining references will
 858  * drain after some time, at which point page will be recycled.
 859  *
 860  * \pre  VM page is locked
 861  * \post pg->cp_state == CPS_FREEING
 862  *
 863  * \see cl_page_operations::cpo_delete()
 864  */
 865 void cl_page_delete(const struct lu_env *env, struct cl_page *pg)
 866 {
 867         PINVRNT(env, pg, cl_page_invariant(pg));
 868         ENTRY;
 869         cl_page_delete0(env, pg);
 870         EXIT;
 871 }
 872 EXPORT_SYMBOL(cl_page_delete);
 873
 874 /**
 875  * Marks page up-to-date.
 876  *
 877  * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The
 878  * layer responsible for VM interaction has to mark/clear page as up-to-date
 879  * by the \a uptodate argument.
 880  *
 881  * \see cl_page_operations::cpo_export()
 882  */
 883 void cl_page_export(const struct lu_env *env, struct cl_page *cl_page,
 884                     int uptodate)
 885 {
 886         const struct cl_page_slice *slice;
 887         int i;
 888
 889         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
 890
 891         cl_page_slice_for_each(cl_page, slice, i) {
 892                 if (slice->cpl_ops->cpo_export != NULL)
 893                         (*slice->cpl_ops->cpo_export)(env, slice, uptodate);
 894         }
 895 }
 896 EXPORT_SYMBOL(cl_page_export);
 897
 898 /**
 899  * Returns true, if \a page is VM locked in a suitable sense by the calling
 900  * thread.
 901  */
 902 int cl_page_is_vmlocked(const struct lu_env *env,
 903                         const struct cl_page *cl_page)
 904 {
 905         const struct cl_page_slice *slice;
 906         int result;
 907
 908         ENTRY;
 909         slice = cl_page_slice_get(cl_page, 0);
 910         PASSERT(env, cl_page, slice->cpl_ops->cpo_is_vmlocked != NULL);
 911         /*
 912          * Call ->cpo_is_vmlocked() directly instead of going through
 913          * CL_PAGE_INVOKE(), because cl_page_is_vmlocked() is used by
 914          * cl_page_invariant().
 915          */
 916         result = slice->cpl_ops->cpo_is_vmlocked(env, slice);
 917         PASSERT(env, cl_page, result == -EBUSY || result == -ENODATA);
 918
 919         RETURN(result == -EBUSY);
 920 }
 921 EXPORT_SYMBOL(cl_page_is_vmlocked);
 922
 923 void cl_page_touch(const struct lu_env *env,
 924                    const struct cl_page *cl_page, size_t to)
 925 {
 926         const struct cl_page_slice *slice;
 927         int i;
 928
 929         ENTRY;
 930
 931         cl_page_slice_for_each(cl_page, slice, i) {
 932                 if (slice->cpl_ops->cpo_page_touch != NULL)
 933                         (*slice->cpl_ops->cpo_page_touch)(env, slice, to);
 934         }
 935
 936         EXIT;
 937 }
 938 EXPORT_SYMBOL(cl_page_touch);
 939
 940 static enum cl_page_state cl_req_type_state(enum cl_req_type crt)
 941 {
 942         ENTRY;
 943         RETURN(crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN);
 944 }
 945
 946 static void cl_page_io_start(const struct lu_env *env,
 947                              struct cl_page *pg, enum cl_req_type crt)
 948 {
 949         /*
 950          * Page is queued for IO, change its state.
 951          */
 952         ENTRY;
 953         cl_page_owner_clear(pg);
 954         cl_page_state_set(env, pg, cl_req_type_state(crt));
 955         EXIT;
 956 }
 957
 958 /**
 959  * Prepares page for immediate transfer. cl_page_operations::cpo_prep() is
 960  * called top-to-bottom. Every layer either agrees to submit this page (by
 961  * returning 0), or requests to omit this page (by returning -EALREADY). Layer
 962  * handling interactions with the VM also has to inform VM that page is under
 963  * transfer now.
 964  */
 965 int cl_page_prep(const struct lu_env *env, struct cl_io *io,
 966                  struct cl_page *cl_page, enum cl_req_type crt)
 967 {
 968         const struct cl_page_slice *slice;
 969         int result = 0;
 970         int i;
 971
 972         PINVRNT(env, cl_page, cl_page_is_owned(cl_page, io));
 973         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
 974         PINVRNT(env, cl_page, crt < CRT_NR);
 975
 976         /*
 977          * this has to be called bottom-to-top, so that llite can set up
 978          * PG_writeback without risking other layers deciding to skip this
 979          * page.
 980          */
 981         if (crt >= CRT_NR)
 982                 return -EINVAL;
 983
 984         cl_page_slice_for_each(cl_page, slice, i) {
 985                 if (slice->cpl_ops->cpo_own)
 986                         result = (*slice->cpl_ops->io[crt].cpo_prep)(env,
 987                                                                      slice,
 988                                                                      io);
 989                 if (result != 0)
 990                         break;
 991         }
 992
 993         if (result >= 0) {
 994                 result = 0;
 995                 cl_page_io_start(env, cl_page, crt);
 996         }
 997
 998         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", crt, result);
 999         return result;
1000 }
1001 EXPORT_SYMBOL(cl_page_prep);
1002
1003 /**
1004  * Notify layers about transfer completion.
1005  *
1006  * Invoked by transfer sub-system (which is a part of osc) to notify layers
1007  * that a transfer, of which this page is a part of has completed.
1008  *
1009  * Completion call-backs are executed in the bottom-up order, so that
1010  * uppermost layer (llite), responsible for the VFS/VM interaction runs last
1011  * and can release locks safely.
1012  *
1013  * \pre  cl_page->cp_state == CPS_PAGEIN || cl_page->cp_state == CPS_PAGEOUT
1014  * \post cl_page->cl_page_state == CPS_CACHED
1015  *
1016  * \see cl_page_operations::cpo_completion()
1017  */
1018 void cl_page_completion(const struct lu_env *env,
1019                         struct cl_page *cl_page, enum cl_req_type crt,
1020                         int ioret)
1021 {
1022         const struct cl_page_slice *slice;
1023         struct cl_sync_io *anchor = cl_page->cp_sync_io;
1024         int i;
1025
1026         ENTRY;
1027         PASSERT(env, cl_page, crt < CRT_NR);
1028         PASSERT(env, cl_page, cl_page->cp_state == cl_req_type_state(crt));
1029
1030         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", crt, ioret);
1031         cl_page_state_set(env, cl_page, CPS_CACHED);
1032         if (crt >= CRT_NR)
1033                 return;
1034
1035         cl_page_slice_for_each_reverse(cl_page, slice, i) {
1036                 if (slice->cpl_ops->io[crt].cpo_completion != NULL)
1037                         (*slice->cpl_ops->io[crt].cpo_completion)(env, slice,
1038                                                                   ioret);
1039         }
1040
1041         if (anchor != NULL) {
1042                 LASSERT(cl_page->cp_sync_io == anchor);
1043                 cl_page->cp_sync_io = NULL;
1044                 cl_sync_io_note(env, anchor, ioret);
1045         }
1046         EXIT;
1047 }
1048 EXPORT_SYMBOL(cl_page_completion);
1049
1050 /**
1051  * Notify layers that transfer formation engine decided to yank this page from
1052  * the cache and to make it a part of a transfer.
1053  *
1054  * \pre  cl_page->cp_state == CPS_CACHED
1055  * \post cl_page->cp_state == CPS_PAGEIN || cl_page->cp_state == CPS_PAGEOUT
1056  *
1057  * \see cl_page_operations::cpo_make_ready()
1058  */
1059 int cl_page_make_ready(const struct lu_env *env, struct cl_page *cl_page,
1060                        enum cl_req_type crt)
1061 {
1062         const struct cl_page_slice *slice;
1063         int result = 0;
1064         int i;
1065
1066         ENTRY;
1067         PINVRNT(env, cl_page, crt < CRT_NR);
1068         if (crt >= CRT_NR)
1069                 RETURN(-EINVAL);
1070
1071         cl_page_slice_for_each(cl_page, slice, i) {
1072                 if (slice->cpl_ops->io[crt].cpo_make_ready != NULL)
1073                         result = (*slice->cpl_ops->io[crt].cpo_make_ready)(env, slice);
1074                 if (result != 0)
1075                         break;
1076         }
1077
1078         if (result >= 0) {
1079                 result = 0;
1080                 PASSERT(env, cl_page, cl_page->cp_state == CPS_CACHED);
1081                 cl_page_io_start(env, cl_page, crt);
1082         }
1083         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", crt, result);
1084
1085         RETURN(result);
1086 }
1087 EXPORT_SYMBOL(cl_page_make_ready);
1088
1089 /**
1090  * Called if a page is being written back by kernel's intention.
1091  *
1092  * \pre  cl_page_is_owned(cl_page, io)
1093  * \post ergo(result == 0, cl_page->cp_state == CPS_PAGEOUT)
1094  *
1095  * \see cl_page_operations::cpo_flush()
1096  */
1097 int cl_page_flush(const struct lu_env *env, struct cl_io *io,
1098                   struct cl_page *cl_page)
1099 {
1100         const struct cl_page_slice *slice;
1101         int result = 0;
1102         int i;
1103
1104         ENTRY;
1105         PINVRNT(env, cl_page, cl_page_is_owned(cl_page, io));
1106         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
1107
1108         cl_page_slice_for_each(cl_page, slice, i) {
1109                 if (slice->cpl_ops->cpo_flush != NULL)
1110                         result = (*slice->cpl_ops->cpo_flush)(env, slice, io);
1111                 if (result != 0)
1112                         break;
1113         }
1114         if (result > 0)
1115                 result = 0;
1116
1117         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d\n", result);
1118         RETURN(result);
1119 }
1120 EXPORT_SYMBOL(cl_page_flush);
1121
1122 /**
1123  * Tells transfer engine that only part of a page is to be transmitted.
1124  *
1125  * \see cl_page_operations::cpo_clip()
1126  */
1127 void cl_page_clip(const struct lu_env *env, struct cl_page *cl_page,
1128                   int from, int to)
1129 {
1130         const struct cl_page_slice *slice;
1131         int i;
1132
1133         PINVRNT(env, cl_page, cl_page_invariant(cl_page));
1134
1135         CL_PAGE_HEADER(D_TRACE, env, cl_page, "%d %d\n", from, to);
1136         cl_page_slice_for_each(cl_page, slice, i) {
1137                 if (slice->cpl_ops->cpo_clip != NULL)
1138                         (*slice->cpl_ops->cpo_clip)(env, slice, from, to);
1139         }
1140 }
1141 EXPORT_SYMBOL(cl_page_clip);
1142
1143 /**
1144  * Prints human readable representation of \a pg to the \a f.
1145  */
1146 void cl_page_header_print(const struct lu_env *env, void *cookie,
1147                           lu_printer_t printer, const struct cl_page *pg)
1148 {
1149         (*printer)(env, cookie,
1150                    "page@%p[%d %p %d %d %p]\n",
1151                    pg, atomic_read(&pg->cp_ref), pg->cp_obj,
1152                    pg->cp_state, pg->cp_type,
1153                    pg->cp_owner);
1154 }
1155 EXPORT_SYMBOL(cl_page_header_print);
1156
1157 /**
1158  * Prints human readable representation of \a cl_page to the \a f.
1159  */
1160 void cl_page_print(const struct lu_env *env, void *cookie,
1161                    lu_printer_t printer, const struct cl_page *cl_page)
1162 {
1163         const struct cl_page_slice *slice;
1164         int result = 0;
1165         int i;
1166
1167         cl_page_header_print(env, cookie, printer, cl_page);
1168         cl_page_slice_for_each(cl_page, slice, i) {
1169                 if (slice->cpl_ops->cpo_print != NULL)
1170                         result = (*slice->cpl_ops->cpo_print)(env, slice,
1171                                                              cookie, printer);
1172                 if (result != 0)
1173                         break;
1174         }
1175         (*printer)(env, cookie, "end page@%p\n", cl_page);
1176 }
1177 EXPORT_SYMBOL(cl_page_print);
1178
1179 /**
1180  * Converts a byte offset within object \a obj into a page index.
1181  */
1182 loff_t cl_offset(const struct cl_object *obj, pgoff_t idx)
1183 {
1184         return (loff_t)idx << PAGE_SHIFT;
1185 }
1186 EXPORT_SYMBOL(cl_offset);
1187
1188 /**
1189  * Converts a page index into a byte offset within object \a obj.
1190  */
1191 pgoff_t cl_index(const struct cl_object *obj, loff_t offset)
1192 {
1193         return offset >> PAGE_SHIFT;
1194 }
1195 EXPORT_SYMBOL(cl_index);
1196
1197 size_t cl_page_size(const struct cl_object *obj)
1198 {
1199         return 1UL << PAGE_SHIFT;
1200 }
1201 EXPORT_SYMBOL(cl_page_size);
1202
1203 /**
1204  * Adds page slice to the compound page.
1205  *
1206  * This is called by cl_object_operations::coo_page_init() methods to add a
1207  * per-layer state to the page. New state is added at the end of
1208  * cl_page::cp_layers list, that is, it is at the bottom of the stack.
1209  *
1210  * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add()
1211  */
1212 void cl_page_slice_add(struct cl_page *cl_page, struct cl_page_slice *slice,
1213                        struct cl_object *obj,
1214                        const struct cl_page_operations *ops)
1215 {
1216         unsigned int offset = (char *)slice -
1217                         ((char *)cl_page + sizeof(*cl_page));
1218
1219         ENTRY;
1220         LASSERT(cl_page->cp_layer_count < CP_MAX_LAYER);
1221         LASSERT(offset < (1 << sizeof(cl_page->cp_layer_offset[0]) * 8));
1222         cl_page->cp_layer_offset[cl_page->cp_layer_count++] = offset;
1223         slice->cpl_obj  = obj;
1224         slice->cpl_ops  = ops;
1225         slice->cpl_page = cl_page;
1226
1227         EXIT;
1228 }
1229 EXPORT_SYMBOL(cl_page_slice_add);
1230
1231 /**
1232  * Allocate and initialize cl_cache, called by ll_init_sbi().
1233  */
1234 struct cl_client_cache *cl_cache_init(unsigned long lru_page_max)
1235 {
1236         struct cl_client_cache  *cache = NULL;
1237
1238         ENTRY;
1239         OBD_ALLOC(cache, sizeof(*cache));
1240         if (cache == NULL)
1241                 RETURN(NULL);
1242
1243         /* Initialize cache data */
1244         atomic_set(&cache->ccc_users, 1);
1245         cache->ccc_lru_max = lru_page_max;
1246         atomic_long_set(&cache->ccc_lru_left, lru_page_max);
1247         spin_lock_init(&cache->ccc_lru_lock);
1248         INIT_LIST_HEAD(&cache->ccc_lru);
1249
1250         /* turn unstable check off by default as it impacts performance */
1251         cache->ccc_unstable_check = 0;
1252         atomic_long_set(&cache->ccc_unstable_nr, 0);
1253         init_waitqueue_head(&cache->ccc_unstable_waitq);
1254         mutex_init(&cache->ccc_max_cache_mb_lock);
1255
1256         RETURN(cache);
1257 }
1258 EXPORT_SYMBOL(cl_cache_init);
1259
1260 /**
1261  * Increase cl_cache refcount
1262  */
1263 void cl_cache_incref(struct cl_client_cache *cache)
1264 {
1265         atomic_inc(&cache->ccc_users);
1266 }
1267 EXPORT_SYMBOL(cl_cache_incref);
1268
1269 /**
1270  * Decrease cl_cache refcount and free the cache if refcount=0.
1271  * Since llite, lov and osc all hold cl_cache refcount,
1272  * the free will not cause race. (LU-6173)
1273  */
1274 void cl_cache_decref(struct cl_client_cache *cache)
1275 {
1276         if (atomic_dec_and_test(&cache->ccc_users))
1277                 OBD_FREE(cache, sizeof(*cache));
1278 }
1279 EXPORT_SYMBOL(cl_cache_decref);