drivers/staging/lustre/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.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 Page.
  37  *
  38  *   Author: Nikita Danilov <nikita.danilov@sun.com>
  39  */
  40
  41 #define DEBUG_SUBSYSTEM S_CLASS
  42
  43 #include "../../include/linux/libcfs/libcfs.h"
  44 #include "../include/obd_class.h"
  45 #include "../include/obd_support.h"
  46 #include <linux/list.h>
  47
  48 #include "../include/cl_object.h"
  49 #include "cl_internal.h"
  50
  51 static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg,
  52                             int radix);
  53
  54 # define PASSERT(env, page, expr)                                          \
  55         do {                                                               \
  56                 if (unlikely(!(expr))) {                                   \
  57                         CL_PAGE_DEBUG(D_ERROR, (env), (page), #expr "\n"); \
  58                         LASSERT(0);                                        \
  59                 }                                                          \
  60         } while (0)
  61
  62 # define PINVRNT(env, page, exp) \
  63         ((void)sizeof(env), (void)sizeof(page), (void)sizeof !!(exp))
  64
  65 /**
  66  * Internal version of cl_page_top, it should be called if the page is
  67  * known to be not freed, says with page referenced, or radix tree lock held,
  68  * or page owned.
  69  */
  70 static struct cl_page *cl_page_top_trusted(struct cl_page *page)
  71 {
  72         while (page->cp_parent != NULL)
  73                 page = page->cp_parent;
  74         return page;
  75 }
  76
  77 /**
  78  * Internal version of cl_page_get().
  79  *
  80  * This function can be used to obtain initial reference to previously
  81  * unreferenced cached object. It can be called only if concurrent page
  82  * reclamation is somehow prevented, e.g., by locking page radix-tree
  83  * (cl_object_header::hdr->coh_page_guard), or by keeping a lock on a VM page,
  84  * associated with \a page.
  85  *
  86  * Use with care! Not exported.
  87  */
  88 static void cl_page_get_trust(struct cl_page *page)
  89 {
  90         LASSERT(atomic_read(&page->cp_ref) > 0);
  91         atomic_inc(&page->cp_ref);
  92 }
  93
  94 /**
  95  * Returns a slice within a page, corresponding to the given layer in the
  96  * device stack.
  97  *
  98  * \see cl_lock_at()
  99  */
 100 static const struct cl_page_slice *
 101 cl_page_at_trusted(const struct cl_page *page,
 102                    const struct lu_device_type *dtype)
 103 {
 104         const struct cl_page_slice *slice;
 105
 106         page = cl_page_top_trusted((struct cl_page *)page);
 107         do {
 108                 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
 109                         if (slice->cpl_obj->co_lu.lo_dev->ld_type == dtype)
 110                                 return slice;
 111                 }
 112                 page = page->cp_child;
 113         } while (page != NULL);
 114         return NULL;
 115 }
 116
 117 /**
 118  * Returns a page with given index in the given object, or NULL if no page is
 119  * found. Acquires a reference on \a page.
 120  *
 121  * Locking: called under cl_object_header::coh_page_guard spin-lock.
 122  */
 123 struct cl_page *cl_page_lookup(struct cl_object_header *hdr, pgoff_t index)
 124 {
 125         struct cl_page *page;
 126
 127         assert_spin_locked(&hdr->coh_page_guard);
 128
 129         page = radix_tree_lookup(&hdr->coh_tree, index);
 130         if (page != NULL)
 131                 cl_page_get_trust(page);
 132         return page;
 133 }
 134 EXPORT_SYMBOL(cl_page_lookup);
 135
 136 /**
 137  * Returns a list of pages by a given [start, end] of \a obj.
 138  *
 139  * \param resched If not NULL, then we give up before hogging CPU for too
 140  * long and set *resched = 1, in that case caller should implement a retry
 141  * logic.
 142  *
 143  * Gang tree lookup (radix_tree_gang_lookup()) optimization is absolutely
 144  * crucial in the face of [offset, EOF] locks.
 145  *
 146  * Return at least one page in @queue unless there is no covered page.
 147  */
 148 int cl_page_gang_lookup(const struct lu_env *env, struct cl_object *obj,
 149                         struct cl_io *io, pgoff_t start, pgoff_t end,
 150                         cl_page_gang_cb_t cb, void *cbdata)
 151 {
 152         struct cl_object_header *hdr;
 153         struct cl_page    *page;
 154         struct cl_page   **pvec;
 155         const struct cl_page_slice  *slice;
 156         const struct lu_device_type *dtype;
 157         pgoff_t           idx;
 158         unsigned int         nr;
 159         unsigned int         i;
 160         unsigned int         j;
 161         int                   res = CLP_GANG_OKAY;
 162         int                   tree_lock = 1;
 163
 164         idx = start;
 165         hdr = cl_object_header(obj);
 166         pvec = cl_env_info(env)->clt_pvec;
 167         dtype = cl_object_top(obj)->co_lu.lo_dev->ld_type;
 168         spin_lock(&hdr->coh_page_guard);
 169         while ((nr = radix_tree_gang_lookup(&hdr->coh_tree, (void **)pvec,
 170                                             idx, CLT_PVEC_SIZE)) > 0) {
 171                 int end_of_region = 0;
 172
 173                 idx = pvec[nr - 1]->cp_index + 1;
 174                 for (i = 0, j = 0; i < nr; ++i) {
 175                         page = pvec[i];
 176                         pvec[i] = NULL;
 177
 178                         LASSERT(page->cp_type == CPT_CACHEABLE);
 179                         if (page->cp_index > end) {
 180                                 end_of_region = 1;
 181                                 break;
 182                         }
 183                         if (page->cp_state == CPS_FREEING)
 184                                 continue;
 185
 186                         slice = cl_page_at_trusted(page, dtype);
 187                         /*
 188                          * Pages for lsm-less file has no underneath sub-page
 189                          * for osc, in case of ...
 190                          */
 191                         PASSERT(env, page, slice != NULL);
 192
 193                         page = slice->cpl_page;
 194                         /*
 195                          * Can safely call cl_page_get_trust() under
 196                          * radix-tree spin-lock.
 197                          *
 198                          * XXX not true, because @page is from object another
 199                          * than @hdr and protected by different tree lock.
 200                          */
 201                         cl_page_get_trust(page);
 202                         lu_ref_add_atomic(&page->cp_reference,
 203                                           "gang_lookup", current);
 204                         pvec[j++] = page;
 205                 }
 206
 207                 /*
 208                  * Here a delicate locking dance is performed. Current thread
 209                  * holds a reference to a page, but has to own it before it
 210                  * can be placed into queue. Owning implies waiting, so
 211                  * radix-tree lock is to be released. After a wait one has to
 212                  * check that pages weren't truncated (cl_page_own() returns
 213                  * error in the latter case).
 214                  */
 215                 spin_unlock(&hdr->coh_page_guard);
 216                 tree_lock = 0;
 217
 218                 for (i = 0; i < j; ++i) {
 219                         page = pvec[i];
 220                         if (res == CLP_GANG_OKAY)
 221                                 res = (*cb)(env, io, page, cbdata);
 222                         lu_ref_del(&page->cp_reference,
 223                                    "gang_lookup", current);
 224                         cl_page_put(env, page);
 225                 }
 226                 if (nr < CLT_PVEC_SIZE || end_of_region)
 227                         break;
 228
 229                 if (res == CLP_GANG_OKAY && need_resched())
 230                         res = CLP_GANG_RESCHED;
 231                 if (res != CLP_GANG_OKAY)
 232                         break;
 233
 234                 spin_lock(&hdr->coh_page_guard);
 235                 tree_lock = 1;
 236         }
 237         if (tree_lock)
 238                 spin_unlock(&hdr->coh_page_guard);
 239         return res;
 240 }
 241 EXPORT_SYMBOL(cl_page_gang_lookup);
 242
 243 static void cl_page_free(const struct lu_env *env, struct cl_page *page)
 244 {
 245         struct cl_object *obj  = page->cp_obj;
 246
 247         PASSERT(env, page, list_empty(&page->cp_batch));
 248         PASSERT(env, page, page->cp_owner == NULL);
 249         PASSERT(env, page, page->cp_req == NULL);
 250         PASSERT(env, page, page->cp_parent == NULL);
 251         PASSERT(env, page, page->cp_state == CPS_FREEING);
 252
 253         might_sleep();
 254         while (!list_empty(&page->cp_layers)) {
 255                 struct cl_page_slice *slice;
 256
 257                 slice = list_entry(page->cp_layers.next,
 258                                        struct cl_page_slice, cpl_linkage);
 259                 list_del_init(page->cp_layers.next);
 260                 slice->cpl_ops->cpo_fini(env, slice);
 261         }
 262         lu_object_ref_del_at(&obj->co_lu, &page->cp_obj_ref, "cl_page", page);
 263         cl_object_put(env, obj);
 264         lu_ref_fini(&page->cp_reference);
 265         kfree(page);
 266 }
 267
 268 /**
 269  * Helper function updating page state. This is the only place in the code
 270  * where cl_page::cp_state field is mutated.
 271  */
 272 static inline void cl_page_state_set_trust(struct cl_page *page,
 273                                            enum cl_page_state state)
 274 {
 275         /* bypass const. */
 276         *(enum cl_page_state *)&page->cp_state = state;
 277 }
 278
 279 static struct cl_page *cl_page_alloc(const struct lu_env *env,
 280                 struct cl_object *o, pgoff_t ind, struct page *vmpage,
 281                 enum cl_page_type type)
 282 {
 283         struct cl_page    *page;
 284         struct lu_object_header *head;
 285
 286         OBD_ALLOC_GFP(page, cl_object_header(o)->coh_page_bufsize,
 287                         GFP_NOFS);
 288         if (page != NULL) {
 289                 int result = 0;
 290
 291                 atomic_set(&page->cp_ref, 1);
 292                 if (type == CPT_CACHEABLE) /* for radix tree */
 293                         atomic_inc(&page->cp_ref);
 294                 page->cp_obj = o;
 295                 cl_object_get(o);
 296                 lu_object_ref_add_at(&o->co_lu, &page->cp_obj_ref, "cl_page",
 297                                      page);
 298                 page->cp_index = ind;
 299                 cl_page_state_set_trust(page, CPS_CACHED);
 300                 page->cp_type = type;
 301                 INIT_LIST_HEAD(&page->cp_layers);
 302                 INIT_LIST_HEAD(&page->cp_batch);
 303                 INIT_LIST_HEAD(&page->cp_flight);
 304                 mutex_init(&page->cp_mutex);
 305                 lu_ref_init(&page->cp_reference);
 306                 head = o->co_lu.lo_header;
 307                 list_for_each_entry(o, &head->loh_layers,
 308                                         co_lu.lo_linkage) {
 309                         if (o->co_ops->coo_page_init != NULL) {
 310                                 result = o->co_ops->coo_page_init(env, o,
 311                                                                   page, vmpage);
 312                                 if (result != 0) {
 313                                         cl_page_delete0(env, page, 0);
 314                                         cl_page_free(env, page);
 315                                         page = ERR_PTR(result);
 316                                         break;
 317                                 }
 318                         }
 319                 }
 320         } else {
 321                 page = ERR_PTR(-ENOMEM);
 322         }
 323         return page;
 324 }
 325
 326 /**
 327  * Returns a cl_page with index \a idx at the object \a o, and associated with
 328  * the VM page \a vmpage.
 329  *
 330  * This is the main entry point into the cl_page caching interface. First, a
 331  * cache (implemented as a per-object radix tree) is consulted. If page is
 332  * found there, it is returned immediately. Otherwise new page is allocated
 333  * and returned. In any case, additional reference to page is acquired.
 334  *
 335  * \see cl_object_find(), cl_lock_find()
 336  */
 337 static struct cl_page *cl_page_find0(const struct lu_env *env,
 338                                      struct cl_object *o,
 339                                      pgoff_t idx, struct page *vmpage,
 340                                      enum cl_page_type type,
 341                                      struct cl_page *parent)
 342 {
 343         struct cl_page    *page = NULL;
 344         struct cl_page    *ghost = NULL;
 345         struct cl_object_header *hdr;
 346         int err;
 347
 348         LASSERT(type == CPT_CACHEABLE || type == CPT_TRANSIENT);
 349         might_sleep();
 350
 351         hdr = cl_object_header(o);
 352
 353         CDEBUG(D_PAGE, "%lu@"DFID" %p %lx %d\n",
 354                idx, PFID(&hdr->coh_lu.loh_fid), vmpage, vmpage->private, type);
 355         /* fast path. */
 356         if (type == CPT_CACHEABLE) {
 357                 /*
 358                  * vmpage lock is used to protect the child/parent
 359                  * relationship
 360                  */
 361                 KLASSERT(PageLocked(vmpage));
 362                 /*
 363                  * cl_vmpage_page() can be called here without any locks as
 364                  *
 365                  *     - "vmpage" is locked (which prevents ->private from
 366                  *       concurrent updates), and
 367                  *
 368                  *     - "o" cannot be destroyed while current thread holds a
 369                  *       reference on it.
 370                  */
 371                 page = cl_vmpage_page(vmpage, o);
 372                 PINVRNT(env, page,
 373                         ergo(page != NULL,
 374                              cl_page_vmpage(env, page) == vmpage &&
 375                              (void *)radix_tree_lookup(&hdr->coh_tree,
 376                                                        idx) == page));
 377         }
 378
 379         if (page != NULL)
 380                 return page;
 381
 382         /* allocate and initialize cl_page */
 383         page = cl_page_alloc(env, o, idx, vmpage, type);
 384         if (IS_ERR(page))
 385                 return page;
 386
 387         if (type == CPT_TRANSIENT) {
 388                 if (parent) {
 389                         LASSERT(page->cp_parent == NULL);
 390                         page->cp_parent = parent;
 391                         parent->cp_child = page;
 392                 }
 393                 return page;
 394         }
 395
 396         /*
 397          * XXX optimization: use radix_tree_preload() here, and change tree
 398          * gfp mask to GFP_KERNEL in cl_object_header_init().
 399          */
 400         spin_lock(&hdr->coh_page_guard);
 401         err = radix_tree_insert(&hdr->coh_tree, idx, page);
 402         if (err != 0) {
 403                 ghost = page;
 404                 /*
 405                  * Noted by Jay: a lock on \a vmpage protects cl_page_find()
 406                  * from this race, but
 407                  *
 408                  *     0. it's better to have cl_page interface "locally
 409                  *     consistent" so that its correctness can be reasoned
 410                  *     about without appealing to the (obscure world of) VM
 411                  *     locking.
 412                  *
 413                  *     1. handling this race allows ->coh_tree to remain
 414                  *     consistent even when VM locking is somehow busted,
 415                  *     which is very useful during diagnosing and debugging.
 416                  */
 417                 page = ERR_PTR(err);
 418                 CL_PAGE_DEBUG(D_ERROR, env, ghost,
 419                               "fail to insert into radix tree: %d\n", err);
 420         } else {
 421                 if (parent) {
 422                         LASSERT(page->cp_parent == NULL);
 423                         page->cp_parent = parent;
 424                         parent->cp_child = page;
 425                 }
 426                 hdr->coh_pages++;
 427         }
 428         spin_unlock(&hdr->coh_page_guard);
 429
 430         if (unlikely(ghost != NULL)) {
 431                 cl_page_delete0(env, ghost, 0);
 432                 cl_page_free(env, ghost);
 433         }
 434         return page;
 435 }
 436
 437 struct cl_page *cl_page_find(const struct lu_env *env, struct cl_object *o,
 438                              pgoff_t idx, struct page *vmpage,
 439                              enum cl_page_type type)
 440 {
 441         return cl_page_find0(env, o, idx, vmpage, type, NULL);
 442 }
 443 EXPORT_SYMBOL(cl_page_find);
 444
 445
 446 struct cl_page *cl_page_find_sub(const struct lu_env *env, struct cl_object *o,
 447                                  pgoff_t idx, struct page *vmpage,
 448                                  struct cl_page *parent)
 449 {
 450         return cl_page_find0(env, o, idx, vmpage, parent->cp_type, parent);
 451 }
 452 EXPORT_SYMBOL(cl_page_find_sub);
 453
 454 static inline int cl_page_invariant(const struct cl_page *pg)
 455 {
 456         struct cl_object_header *header;
 457         struct cl_page    *parent;
 458         struct cl_page    *child;
 459         struct cl_io        *owner;
 460
 461         /*
 462          * Page invariant is protected by a VM lock.
 463          */
 464         LINVRNT(cl_page_is_vmlocked(NULL, pg));
 465
 466         header = cl_object_header(pg->cp_obj);
 467         parent = pg->cp_parent;
 468         child  = pg->cp_child;
 469         owner  = pg->cp_owner;
 470
 471         return cl_page_in_use(pg) &&
 472                 ergo(parent != NULL, parent->cp_child == pg) &&
 473                 ergo(child != NULL, child->cp_parent == pg) &&
 474                 ergo(child != NULL, pg->cp_obj != child->cp_obj) &&
 475                 ergo(parent != NULL, pg->cp_obj != parent->cp_obj) &&
 476                 ergo(owner != NULL && parent != NULL,
 477                      parent->cp_owner == pg->cp_owner->ci_parent) &&
 478                 ergo(owner != NULL && child != NULL,
 479                      child->cp_owner->ci_parent == owner) &&
 480                 /*
 481                  * Either page is early in initialization (has neither child
 482                  * nor parent yet), or it is in the object radix tree.
 483                  */
 484                 ergo(pg->cp_state < CPS_FREEING && pg->cp_type == CPT_CACHEABLE,
 485                      (void *)radix_tree_lookup(&header->coh_tree,
 486                                                pg->cp_index) == pg ||
 487                      (child == NULL && parent == NULL));
 488 }
 489
 490 static void cl_page_state_set0(const struct lu_env *env,
 491                                struct cl_page *page, enum cl_page_state state)
 492 {
 493         enum cl_page_state old;
 494
 495         /*
 496          * Matrix of allowed state transitions [old][new], for sanity
 497          * checking.
 498          */
 499         static const int allowed_transitions[CPS_NR][CPS_NR] = {
 500                 [CPS_CACHED] = {
 501                         [CPS_CACHED]  = 0,
 502                         [CPS_OWNED]   = 1, /* io finds existing cached page */
 503                         [CPS_PAGEIN]  = 0,
 504                         [CPS_PAGEOUT] = 1, /* write-out from the cache */
 505                         [CPS_FREEING] = 1, /* eviction on the memory pressure */
 506                 },
 507                 [CPS_OWNED] = {
 508                         [CPS_CACHED]  = 1, /* release to the cache */
 509                         [CPS_OWNED]   = 0,
 510                         [CPS_PAGEIN]  = 1, /* start read immediately */
 511                         [CPS_PAGEOUT] = 1, /* start write immediately */
 512                         [CPS_FREEING] = 1, /* lock invalidation or truncate */
 513                 },
 514                 [CPS_PAGEIN] = {
 515                         [CPS_CACHED]  = 1, /* io completion */
 516                         [CPS_OWNED]   = 0,
 517                         [CPS_PAGEIN]  = 0,
 518                         [CPS_PAGEOUT] = 0,
 519                         [CPS_FREEING] = 0,
 520                 },
 521                 [CPS_PAGEOUT] = {
 522                         [CPS_CACHED]  = 1, /* io completion */
 523                         [CPS_OWNED]   = 0,
 524                         [CPS_PAGEIN]  = 0,
 525                         [CPS_PAGEOUT] = 0,
 526                         [CPS_FREEING] = 0,
 527                 },
 528                 [CPS_FREEING] = {
 529                         [CPS_CACHED]  = 0,
 530                         [CPS_OWNED]   = 0,
 531                         [CPS_PAGEIN]  = 0,
 532                         [CPS_PAGEOUT] = 0,
 533                         [CPS_FREEING] = 0,
 534                 }
 535         };
 536
 537         old = page->cp_state;
 538         PASSERT(env, page, allowed_transitions[old][state]);
 539         CL_PAGE_HEADER(D_TRACE, env, page, "%d -> %d\n", old, state);
 540         for (; page != NULL; page = page->cp_child) {
 541                 PASSERT(env, page, page->cp_state == old);
 542                 PASSERT(env, page,
 543                         equi(state == CPS_OWNED, page->cp_owner != NULL));
 544
 545                 cl_page_state_set_trust(page, state);
 546         }
 547 }
 548
 549 static void cl_page_state_set(const struct lu_env *env,
 550                               struct cl_page *page, enum cl_page_state state)
 551 {
 552         cl_page_state_set0(env, page, state);
 553 }
 554
 555 /**
 556  * Acquires an additional reference to a page.
 557  *
 558  * This can be called only by caller already possessing a reference to \a
 559  * page.
 560  *
 561  * \see cl_object_get(), cl_lock_get().
 562  */
 563 void cl_page_get(struct cl_page *page)
 564 {
 565         cl_page_get_trust(page);
 566 }
 567 EXPORT_SYMBOL(cl_page_get);
 568
 569 /**
 570  * Releases a reference to a page.
 571  *
 572  * When last reference is released, page is returned to the cache, unless it
 573  * is in cl_page_state::CPS_FREEING state, in which case it is immediately
 574  * destroyed.
 575  *
 576  * \see cl_object_put(), cl_lock_put().
 577  */
 578 void cl_page_put(const struct lu_env *env, struct cl_page *page)
 579 {
 580         PASSERT(env, page, atomic_read(&page->cp_ref) > !!page->cp_parent);
 581
 582         CL_PAGE_HEADER(D_TRACE, env, page, "%d\n",
 583                        atomic_read(&page->cp_ref));
 584
 585         if (atomic_dec_and_test(&page->cp_ref)) {
 586                 LASSERT(page->cp_state == CPS_FREEING);
 587
 588                 LASSERT(atomic_read(&page->cp_ref) == 0);
 589                 PASSERT(env, page, page->cp_owner == NULL);
 590                 PASSERT(env, page, list_empty(&page->cp_batch));
 591                 /*
 592                  * Page is no longer reachable by other threads. Tear
 593                  * it down.
 594                  */
 595                 cl_page_free(env, page);
 596         }
 597 }
 598 EXPORT_SYMBOL(cl_page_put);
 599
 600 /**
 601  * Returns a VM page associated with a given cl_page.
 602  */
 603 struct page *cl_page_vmpage(const struct lu_env *env, struct cl_page *page)
 604 {
 605         const struct cl_page_slice *slice;
 606
 607         /*
 608          * Find uppermost layer with ->cpo_vmpage() method, and return its
 609          * result.
 610          */
 611         page = cl_page_top(page);
 612         do {
 613                 list_for_each_entry(slice, &page->cp_layers, cpl_linkage) {
 614                         if (slice->cpl_ops->cpo_vmpage != NULL)
 615                                 return slice->cpl_ops->cpo_vmpage(env, slice);
 616                 }
 617                 page = page->cp_child;
 618         } while (page != NULL);
 619         LBUG(); /* ->cpo_vmpage() has to be defined somewhere in the stack */
 620 }
 621 EXPORT_SYMBOL(cl_page_vmpage);
 622
 623 /**
 624  * Returns a cl_page associated with a VM page, and given cl_object.
 625  */
 626 struct cl_page *cl_vmpage_page(struct page *vmpage, struct cl_object *obj)
 627 {
 628         struct cl_page *top;
 629         struct cl_page *page;
 630
 631         KLASSERT(PageLocked(vmpage));
 632
 633         /*
 634          * NOTE: absence of races and liveness of data are guaranteed by page
 635          *       lock on a "vmpage". That works because object destruction has
 636          *       bottom-to-top pass.
 637          */
 638
 639         /*
 640          * This loop assumes that ->private points to the top-most page. This
 641          * can be rectified easily.
 642          */
 643         top = (struct cl_page *)vmpage->private;
 644         if (top == NULL)
 645                 return NULL;
 646
 647         for (page = top; page != NULL; page = page->cp_child) {
 648                 if (cl_object_same(page->cp_obj, obj)) {
 649                         cl_page_get_trust(page);
 650                         break;
 651                 }
 652         }
 653         LASSERT(ergo(page, page->cp_type == CPT_CACHEABLE));
 654         return page;
 655 }
 656 EXPORT_SYMBOL(cl_vmpage_page);
 657
 658 /**
 659  * Returns the top-page for a given page.
 660  *
 661  * \see cl_object_top(), cl_io_top()
 662  */
 663 struct cl_page *cl_page_top(struct cl_page *page)
 664 {
 665         return cl_page_top_trusted(page);
 666 }
 667 EXPORT_SYMBOL(cl_page_top);
 668
 669 const struct cl_page_slice *cl_page_at(const struct cl_page *page,
 670                                        const struct lu_device_type *dtype)
 671 {
 672         return cl_page_at_trusted(page, dtype);
 673 }
 674 EXPORT_SYMBOL(cl_page_at);
 675
 676 #define CL_PAGE_OP(opname) offsetof(struct cl_page_operations, opname)
 677
 678 #define CL_PAGE_INVOKE(_env, _page, _op, _proto, ...)              \
 679 ({                                                                    \
 680         const struct lu_env     *__env  = (_env);                   \
 681         struct cl_page       *__page = (_page);            \
 682         const struct cl_page_slice *__scan;                          \
 683         int                      __result;                         \
 684         ptrdiff_t                  __op   = (_op);                   \
 685         int                    (*__method)_proto;                   \
 686                                                                         \
 687         __result = 0;                                              \
 688         __page = cl_page_top(__page);                              \
 689         do {                                                        \
 690                 list_for_each_entry(__scan, &__page->cp_layers,     \
 691                                         cpl_linkage) {            \
 692                         __method = *(void **)((char *)__scan->cpl_ops + \
 693                                               __op);                \
 694                         if (__method != NULL) {                  \
 695                                 __result = (*__method)(__env, __scan,   \
 696                                                        ## __VA_ARGS__); \
 697                                 if (__result != 0)                    \
 698                                         break;                    \
 699                         }                                              \
 700                 }                                                      \
 701                 __page = __page->cp_child;                            \
 702         } while (__page != NULL && __result == 0);                    \
 703         if (__result > 0)                                              \
 704                 __result = 0;                                      \
 705         __result;                                                      \
 706 })
 707
 708 #define CL_PAGE_INVOID(_env, _page, _op, _proto, ...)              \
 709 do {                                                                \
 710         const struct lu_env     *__env  = (_env);                   \
 711         struct cl_page       *__page = (_page);            \
 712         const struct cl_page_slice *__scan;                          \
 713         ptrdiff_t                  __op   = (_op);                   \
 714         void                  (*__method)_proto;                    \
 715                                                                         \
 716         __page = cl_page_top(__page);                              \
 717         do {                                                        \
 718                 list_for_each_entry(__scan, &__page->cp_layers,     \
 719                                         cpl_linkage) {            \
 720                         __method = *(void **)((char *)__scan->cpl_ops + \
 721                                               __op);                \
 722                         if (__method != NULL)                      \
 723                                 (*__method)(__env, __scan,            \
 724                                             ## __VA_ARGS__);        \
 725                 }                                                      \
 726                 __page = __page->cp_child;                            \
 727         } while (__page != NULL);                                      \
 728 } while (0)
 729
 730 #define CL_PAGE_INVOID_REVERSE(_env, _page, _op, _proto, ...)          \
 731 do {                                                                    \
 732         const struct lu_env     *__env  = (_env);                       \
 733         struct cl_page       *__page = (_page);                \
 734         const struct cl_page_slice *__scan;                              \
 735         ptrdiff_t                  __op   = (_op);                       \
 736         void                  (*__method)_proto;                        \
 737                                                                             \
 738         /* get to the bottom page. */                                  \
 739         while (__page->cp_child != NULL)                                    \
 740                 __page = __page->cp_child;                                \
 741         do {                                                            \
 742                 list_for_each_entry_reverse(__scan, &__page->cp_layers, \
 743                                                 cpl_linkage) {        \
 744                         __method = *(void **)((char *)__scan->cpl_ops +     \
 745                                               __op);                    \
 746                         if (__method != NULL)                          \
 747                                 (*__method)(__env, __scan,                \
 748                                             ## __VA_ARGS__);            \
 749                 }                                                          \
 750                 __page = __page->cp_parent;                              \
 751         } while (__page != NULL);                                          \
 752 } while (0)
 753
 754 static int cl_page_invoke(const struct lu_env *env,
 755                           struct cl_io *io, struct cl_page *page, ptrdiff_t op)
 756
 757 {
 758         PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
 759         return CL_PAGE_INVOKE(env, page, op,
 760                               (const struct lu_env *,
 761                                const struct cl_page_slice *, struct cl_io *),
 762                               io);
 763 }
 764
 765 static void cl_page_invoid(const struct lu_env *env,
 766                            struct cl_io *io, struct cl_page *page, ptrdiff_t op)
 767
 768 {
 769         PINVRNT(env, page, cl_object_same(page->cp_obj, io->ci_obj));
 770         CL_PAGE_INVOID(env, page, op,
 771                        (const struct lu_env *,
 772                         const struct cl_page_slice *, struct cl_io *), io);
 773 }
 774
 775 static void cl_page_owner_clear(struct cl_page *page)
 776 {
 777         for (page = cl_page_top(page); page != NULL; page = page->cp_child) {
 778                 if (page->cp_owner != NULL) {
 779                         LASSERT(page->cp_owner->ci_owned_nr > 0);
 780                         page->cp_owner->ci_owned_nr--;
 781                         page->cp_owner = NULL;
 782                         page->cp_task = NULL;
 783                 }
 784         }
 785 }
 786
 787 static void cl_page_owner_set(struct cl_page *page)
 788 {
 789         for (page = cl_page_top(page); page != NULL; page = page->cp_child) {
 790                 LASSERT(page->cp_owner != NULL);
 791                 page->cp_owner->ci_owned_nr++;
 792         }
 793 }
 794
 795 void cl_page_disown0(const struct lu_env *env,
 796                      struct cl_io *io, struct cl_page *pg)
 797 {
 798         enum cl_page_state state;
 799
 800         state = pg->cp_state;
 801         PINVRNT(env, pg, state == CPS_OWNED || state == CPS_FREEING);
 802         PINVRNT(env, pg, cl_page_invariant(pg));
 803         cl_page_owner_clear(pg);
 804
 805         if (state == CPS_OWNED)
 806                 cl_page_state_set(env, pg, CPS_CACHED);
 807         /*
 808          * Completion call-backs are executed in the bottom-up order, so that
 809          * uppermost layer (llite), responsible for VFS/VM interaction runs
 810          * last and can release locks safely.
 811          */
 812         CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_disown),
 813                                (const struct lu_env *,
 814                                 const struct cl_page_slice *, struct cl_io *),
 815                                io);
 816 }
 817
 818 /**
 819  * returns true, iff page is owned by the given io.
 820  */
 821 int cl_page_is_owned(const struct cl_page *pg, const struct cl_io *io)
 822 {
 823         LINVRNT(cl_object_same(pg->cp_obj, io->ci_obj));
 824         return pg->cp_state == CPS_OWNED && pg->cp_owner == io;
 825 }
 826 EXPORT_SYMBOL(cl_page_is_owned);
 827
 828 /**
 829  * Try to own a page by IO.
 830  *
 831  * Waits until page is in cl_page_state::CPS_CACHED state, and then switch it
 832  * into cl_page_state::CPS_OWNED state.
 833  *
 834  * \pre  !cl_page_is_owned(pg, io)
 835  * \post result == 0 iff cl_page_is_owned(pg, io)
 836  *
 837  * \retval 0   success
 838  *
 839  * \retval -ve failure, e.g., page was destroyed (and landed in
 840  *           cl_page_state::CPS_FREEING instead of cl_page_state::CPS_CACHED).
 841  *           or, page was owned by another thread, or in IO.
 842  *
 843  * \see cl_page_disown()
 844  * \see cl_page_operations::cpo_own()
 845  * \see cl_page_own_try()
 846  * \see cl_page_own
 847  */
 848 static int cl_page_own0(const struct lu_env *env, struct cl_io *io,
 849                         struct cl_page *pg, int nonblock)
 850 {
 851         int result;
 852
 853         PINVRNT(env, pg, !cl_page_is_owned(pg, io));
 854
 855         pg = cl_page_top(pg);
 856         io = cl_io_top(io);
 857
 858         if (pg->cp_state == CPS_FREEING) {
 859                 result = -ENOENT;
 860         } else {
 861                 result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(cpo_own),
 862                                         (const struct lu_env *,
 863                                          const struct cl_page_slice *,
 864                                          struct cl_io *, int),
 865                                         io, nonblock);
 866                 if (result == 0) {
 867                         PASSERT(env, pg, pg->cp_owner == NULL);
 868                         PASSERT(env, pg, pg->cp_req == NULL);
 869                         pg->cp_owner = io;
 870                         pg->cp_task  = current;
 871                         cl_page_owner_set(pg);
 872                         if (pg->cp_state != CPS_FREEING) {
 873                                 cl_page_state_set(env, pg, CPS_OWNED);
 874                         } else {
 875                                 cl_page_disown0(env, io, pg);
 876                                 result = -ENOENT;
 877                         }
 878                 }
 879         }
 880         PINVRNT(env, pg, ergo(result == 0, cl_page_invariant(pg)));
 881         return result;
 882 }
 883
 884 /**
 885  * Own a page, might be blocked.
 886  *
 887  * \see cl_page_own0()
 888  */
 889 int cl_page_own(const struct lu_env *env, struct cl_io *io, struct cl_page *pg)
 890 {
 891         return cl_page_own0(env, io, pg, 0);
 892 }
 893 EXPORT_SYMBOL(cl_page_own);
 894
 895 /**
 896  * Nonblock version of cl_page_own().
 897  *
 898  * \see cl_page_own0()
 899  */
 900 int cl_page_own_try(const struct lu_env *env, struct cl_io *io,
 901                     struct cl_page *pg)
 902 {
 903         return cl_page_own0(env, io, pg, 1);
 904 }
 905 EXPORT_SYMBOL(cl_page_own_try);
 906
 907
 908 /**
 909  * Assume page ownership.
 910  *
 911  * Called when page is already locked by the hosting VM.
 912  *
 913  * \pre !cl_page_is_owned(pg, io)
 914  * \post cl_page_is_owned(pg, io)
 915  *
 916  * \see cl_page_operations::cpo_assume()
 917  */
 918 void cl_page_assume(const struct lu_env *env,
 919                     struct cl_io *io, struct cl_page *pg)
 920 {
 921         PINVRNT(env, pg, cl_object_same(pg->cp_obj, io->ci_obj));
 922
 923         pg = cl_page_top(pg);
 924         io = cl_io_top(io);
 925
 926         cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_assume));
 927         PASSERT(env, pg, pg->cp_owner == NULL);
 928         pg->cp_owner = io;
 929         pg->cp_task = current;
 930         cl_page_owner_set(pg);
 931         cl_page_state_set(env, pg, CPS_OWNED);
 932 }
 933 EXPORT_SYMBOL(cl_page_assume);
 934
 935 /**
 936  * Releases page ownership without unlocking the page.
 937  *
 938  * Moves page into cl_page_state::CPS_CACHED without releasing a lock on the
 939  * underlying VM page (as VM is supposed to do this itself).
 940  *
 941  * \pre   cl_page_is_owned(pg, io)
 942  * \post !cl_page_is_owned(pg, io)
 943  *
 944  * \see cl_page_assume()
 945  */
 946 void cl_page_unassume(const struct lu_env *env,
 947                       struct cl_io *io, struct cl_page *pg)
 948 {
 949         PINVRNT(env, pg, cl_page_is_owned(pg, io));
 950         PINVRNT(env, pg, cl_page_invariant(pg));
 951
 952         pg = cl_page_top(pg);
 953         io = cl_io_top(io);
 954         cl_page_owner_clear(pg);
 955         cl_page_state_set(env, pg, CPS_CACHED);
 956         CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(cpo_unassume),
 957                                (const struct lu_env *,
 958                                 const struct cl_page_slice *, struct cl_io *),
 959                                io);
 960 }
 961 EXPORT_SYMBOL(cl_page_unassume);
 962
 963 /**
 964  * Releases page ownership.
 965  *
 966  * Moves page into cl_page_state::CPS_CACHED.
 967  *
 968  * \pre   cl_page_is_owned(pg, io)
 969  * \post !cl_page_is_owned(pg, io)
 970  *
 971  * \see cl_page_own()
 972  * \see cl_page_operations::cpo_disown()
 973  */
 974 void cl_page_disown(const struct lu_env *env,
 975                     struct cl_io *io, struct cl_page *pg)
 976 {
 977         PINVRNT(env, pg, cl_page_is_owned(pg, io));
 978
 979         pg = cl_page_top(pg);
 980         io = cl_io_top(io);
 981         cl_page_disown0(env, io, pg);
 982 }
 983 EXPORT_SYMBOL(cl_page_disown);
 984
 985 /**
 986  * Called when page is to be removed from the object, e.g., as a result of
 987  * truncate.
 988  *
 989  * Calls cl_page_operations::cpo_discard() top-to-bottom.
 990  *
 991  * \pre cl_page_is_owned(pg, io)
 992  *
 993  * \see cl_page_operations::cpo_discard()
 994  */
 995 void cl_page_discard(const struct lu_env *env,
 996                      struct cl_io *io, struct cl_page *pg)
 997 {
 998         PINVRNT(env, pg, cl_page_is_owned(pg, io));
 999         PINVRNT(env, pg, cl_page_invariant(pg));
1000
1001         cl_page_invoid(env, io, pg, CL_PAGE_OP(cpo_discard));
1002 }
1003 EXPORT_SYMBOL(cl_page_discard);
1004
1005 /**
1006  * Version of cl_page_delete() that can be called for not fully constructed
1007  * pages, e.g,. in a error handling cl_page_find()->cl_page_delete0()
1008  * path. Doesn't check page invariant.
1009  */
1010 static void cl_page_delete0(const struct lu_env *env, struct cl_page *pg,
1011                             int radix)
1012 {
1013         struct cl_page *tmp = pg;
1014
1015         PASSERT(env, pg, pg == cl_page_top(pg));
1016         PASSERT(env, pg, pg->cp_state != CPS_FREEING);
1017
1018         /*
1019          * Severe all ways to obtain new pointers to @pg.
1020          */
1021         cl_page_owner_clear(pg);
1022
1023         /*
1024          * unexport the page firstly before freeing it so that
1025          * the page content is considered to be invalid.
1026          * We have to do this because a CPS_FREEING cl_page may
1027          * be NOT under the protection of a cl_lock.
1028          * Afterwards, if this page is found by other threads, then this
1029          * page will be forced to reread.
1030          */
1031         cl_page_export(env, pg, 0);
1032         cl_page_state_set0(env, pg, CPS_FREEING);
1033
1034         CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_delete),
1035                        (const struct lu_env *, const struct cl_page_slice *));
1036
1037         if (tmp->cp_type == CPT_CACHEABLE) {
1038                 if (!radix)
1039                         /* !radix means that @pg is not yet in the radix tree,
1040                          * skip removing it.
1041                          */
1042                         tmp = pg->cp_child;
1043                 for (; tmp != NULL; tmp = tmp->cp_child) {
1044                         void                *value;
1045                         struct cl_object_header *hdr;
1046
1047                         hdr = cl_object_header(tmp->cp_obj);
1048                         spin_lock(&hdr->coh_page_guard);
1049                         value = radix_tree_delete(&hdr->coh_tree,
1050                                                   tmp->cp_index);
1051                         PASSERT(env, tmp, value == tmp);
1052                         PASSERT(env, tmp, hdr->coh_pages > 0);
1053                         hdr->coh_pages--;
1054                         spin_unlock(&hdr->coh_page_guard);
1055                         cl_page_put(env, tmp);
1056                 }
1057         }
1058 }
1059
1060 /**
1061  * Called when a decision is made to throw page out of memory.
1062  *
1063  * Notifies all layers about page destruction by calling
1064  * cl_page_operations::cpo_delete() method top-to-bottom.
1065  *
1066  * Moves page into cl_page_state::CPS_FREEING state (this is the only place
1067  * where transition to this state happens).
1068  *
1069  * Eliminates all venues through which new references to the page can be
1070  * obtained:
1071  *
1072  *     - removes page from the radix trees,
1073  *
1074  *     - breaks linkage from VM page to cl_page.
1075  *
1076  * Once page reaches cl_page_state::CPS_FREEING, all remaining references will
1077  * drain after some time, at which point page will be recycled.
1078  *
1079  * \pre  pg == cl_page_top(pg)
1080  * \pre  VM page is locked
1081  * \post pg->cp_state == CPS_FREEING
1082  *
1083  * \see cl_page_operations::cpo_delete()
1084  */
1085 void cl_page_delete(const struct lu_env *env, struct cl_page *pg)
1086 {
1087         PINVRNT(env, pg, cl_page_invariant(pg));
1088         cl_page_delete0(env, pg, 1);
1089 }
1090 EXPORT_SYMBOL(cl_page_delete);
1091
1092 /**
1093  * Unmaps page from user virtual memory.
1094  *
1095  * Calls cl_page_operations::cpo_unmap() through all layers top-to-bottom. The
1096  * layer responsible for VM interaction has to unmap page from user space
1097  * virtual memory.
1098  *
1099  * \see cl_page_operations::cpo_unmap()
1100  */
1101 int cl_page_unmap(const struct lu_env *env,
1102                   struct cl_io *io, struct cl_page *pg)
1103 {
1104         PINVRNT(env, pg, cl_page_is_owned(pg, io));
1105         PINVRNT(env, pg, cl_page_invariant(pg));
1106
1107         return cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_unmap));
1108 }
1109 EXPORT_SYMBOL(cl_page_unmap);
1110
1111 /**
1112  * Marks page up-to-date.
1113  *
1114  * Call cl_page_operations::cpo_export() through all layers top-to-bottom. The
1115  * layer responsible for VM interaction has to mark/clear page as up-to-date
1116  * by the \a uptodate argument.
1117  *
1118  * \see cl_page_operations::cpo_export()
1119  */
1120 void cl_page_export(const struct lu_env *env, struct cl_page *pg, int uptodate)
1121 {
1122         PINVRNT(env, pg, cl_page_invariant(pg));
1123         CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_export),
1124                        (const struct lu_env *,
1125                         const struct cl_page_slice *, int), uptodate);
1126 }
1127 EXPORT_SYMBOL(cl_page_export);
1128
1129 /**
1130  * Returns true, iff \a pg is VM locked in a suitable sense by the calling
1131  * thread.
1132  */
1133 int cl_page_is_vmlocked(const struct lu_env *env, const struct cl_page *pg)
1134 {
1135         int result;
1136         const struct cl_page_slice *slice;
1137
1138         pg = cl_page_top_trusted((struct cl_page *)pg);
1139         slice = container_of(pg->cp_layers.next,
1140                              const struct cl_page_slice, cpl_linkage);
1141         PASSERT(env, pg, slice->cpl_ops->cpo_is_vmlocked != NULL);
1142         /*
1143          * Call ->cpo_is_vmlocked() directly instead of going through
1144          * CL_PAGE_INVOKE(), because cl_page_is_vmlocked() is used by
1145          * cl_page_invariant().
1146          */
1147         result = slice->cpl_ops->cpo_is_vmlocked(env, slice);
1148         PASSERT(env, pg, result == -EBUSY || result == -ENODATA);
1149         return result == -EBUSY;
1150 }
1151 EXPORT_SYMBOL(cl_page_is_vmlocked);
1152
1153 static enum cl_page_state cl_req_type_state(enum cl_req_type crt)
1154 {
1155         return crt == CRT_WRITE ? CPS_PAGEOUT : CPS_PAGEIN;
1156 }
1157
1158 static void cl_page_io_start(const struct lu_env *env,
1159                              struct cl_page *pg, enum cl_req_type crt)
1160 {
1161         /*
1162          * Page is queued for IO, change its state.
1163          */
1164         cl_page_owner_clear(pg);
1165         cl_page_state_set(env, pg, cl_req_type_state(crt));
1166 }
1167
1168 /**
1169  * Prepares page for immediate transfer. cl_page_operations::cpo_prep() is
1170  * called top-to-bottom. Every layer either agrees to submit this page (by
1171  * returning 0), or requests to omit this page (by returning -EALREADY). Layer
1172  * handling interactions with the VM also has to inform VM that page is under
1173  * transfer now.
1174  */
1175 int cl_page_prep(const struct lu_env *env, struct cl_io *io,
1176                  struct cl_page *pg, enum cl_req_type crt)
1177 {
1178         int result;
1179
1180         PINVRNT(env, pg, cl_page_is_owned(pg, io));
1181         PINVRNT(env, pg, cl_page_invariant(pg));
1182         PINVRNT(env, pg, crt < CRT_NR);
1183
1184         /*
1185          * XXX this has to be called bottom-to-top, so that llite can set up
1186          * PG_writeback without risking other layers deciding to skip this
1187          * page.
1188          */
1189         if (crt >= CRT_NR)
1190                 return -EINVAL;
1191         result = cl_page_invoke(env, io, pg, CL_PAGE_OP(io[crt].cpo_prep));
1192         if (result == 0)
1193                 cl_page_io_start(env, pg, crt);
1194
1195         KLASSERT(ergo(crt == CRT_WRITE && pg->cp_type == CPT_CACHEABLE,
1196                       equi(result == 0,
1197                            PageWriteback(cl_page_vmpage(env, pg)))));
1198         CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1199         return result;
1200 }
1201 EXPORT_SYMBOL(cl_page_prep);
1202
1203 /**
1204  * Notify layers about transfer completion.
1205  *
1206  * Invoked by transfer sub-system (which is a part of osc) to notify layers
1207  * that a transfer, of which this page is a part of has completed.
1208  *
1209  * Completion call-backs are executed in the bottom-up order, so that
1210  * uppermost layer (llite), responsible for the VFS/VM interaction runs last
1211  * and can release locks safely.
1212  *
1213  * \pre  pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
1214  * \post pg->cp_state == CPS_CACHED
1215  *
1216  * \see cl_page_operations::cpo_completion()
1217  */
1218 void cl_page_completion(const struct lu_env *env,
1219                         struct cl_page *pg, enum cl_req_type crt, int ioret)
1220 {
1221         struct cl_sync_io *anchor = pg->cp_sync_io;
1222
1223         PASSERT(env, pg, crt < CRT_NR);
1224         /* cl_page::cp_req already cleared by the caller (osc_completion()) */
1225         PASSERT(env, pg, pg->cp_req == NULL);
1226         PASSERT(env, pg, pg->cp_state == cl_req_type_state(crt));
1227
1228         CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, ioret);
1229         if (crt == CRT_READ && ioret == 0) {
1230                 PASSERT(env, pg, !(pg->cp_flags & CPF_READ_COMPLETED));
1231                 pg->cp_flags |= CPF_READ_COMPLETED;
1232         }
1233
1234         cl_page_state_set(env, pg, CPS_CACHED);
1235         if (crt >= CRT_NR)
1236                 return;
1237         CL_PAGE_INVOID_REVERSE(env, pg, CL_PAGE_OP(io[crt].cpo_completion),
1238                                (const struct lu_env *,
1239                                 const struct cl_page_slice *, int), ioret);
1240         if (anchor) {
1241                 LASSERT(cl_page_is_vmlocked(env, pg));
1242                 LASSERT(pg->cp_sync_io == anchor);
1243                 pg->cp_sync_io = NULL;
1244         }
1245         /*
1246          * As page->cp_obj is pinned by a reference from page->cp_req, it is
1247          * safe to call cl_page_put() without risking object destruction in a
1248          * non-blocking context.
1249          */
1250         cl_page_put(env, pg);
1251
1252         if (anchor)
1253                 cl_sync_io_note(anchor, ioret);
1254 }
1255 EXPORT_SYMBOL(cl_page_completion);
1256
1257 /**
1258  * Notify layers that transfer formation engine decided to yank this page from
1259  * the cache and to make it a part of a transfer.
1260  *
1261  * \pre  pg->cp_state == CPS_CACHED
1262  * \post pg->cp_state == CPS_PAGEIN || pg->cp_state == CPS_PAGEOUT
1263  *
1264  * \see cl_page_operations::cpo_make_ready()
1265  */
1266 int cl_page_make_ready(const struct lu_env *env, struct cl_page *pg,
1267                        enum cl_req_type crt)
1268 {
1269         int result;
1270
1271         PINVRNT(env, pg, crt < CRT_NR);
1272
1273         if (crt >= CRT_NR)
1274                 return -EINVAL;
1275         result = CL_PAGE_INVOKE(env, pg, CL_PAGE_OP(io[crt].cpo_make_ready),
1276                                 (const struct lu_env *,
1277                                  const struct cl_page_slice *));
1278         if (result == 0) {
1279                 PASSERT(env, pg, pg->cp_state == CPS_CACHED);
1280                 cl_page_io_start(env, pg, crt);
1281         }
1282         CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1283         return result;
1284 }
1285 EXPORT_SYMBOL(cl_page_make_ready);
1286
1287 /**
1288  * Notify layers that high level io decided to place this page into a cache
1289  * for future transfer.
1290  *
1291  * The layer implementing transfer engine (osc) has to register this page in
1292  * its queues.
1293  *
1294  * \pre  cl_page_is_owned(pg, io)
1295  * \post cl_page_is_owned(pg, io)
1296  *
1297  * \see cl_page_operations::cpo_cache_add()
1298  */
1299 int cl_page_cache_add(const struct lu_env *env, struct cl_io *io,
1300                       struct cl_page *pg, enum cl_req_type crt)
1301 {
1302         const struct cl_page_slice *scan;
1303         int result = 0;
1304
1305         PINVRNT(env, pg, crt < CRT_NR);
1306         PINVRNT(env, pg, cl_page_is_owned(pg, io));
1307         PINVRNT(env, pg, cl_page_invariant(pg));
1308
1309         if (crt >= CRT_NR)
1310                 return -EINVAL;
1311
1312         list_for_each_entry(scan, &pg->cp_layers, cpl_linkage) {
1313                 if (scan->cpl_ops->io[crt].cpo_cache_add == NULL)
1314                         continue;
1315
1316                 result = scan->cpl_ops->io[crt].cpo_cache_add(env, scan, io);
1317                 if (result != 0)
1318                         break;
1319         }
1320         CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", crt, result);
1321         return result;
1322 }
1323 EXPORT_SYMBOL(cl_page_cache_add);
1324
1325 /**
1326  * Called if a pge is being written back by kernel's intention.
1327  *
1328  * \pre  cl_page_is_owned(pg, io)
1329  * \post ergo(result == 0, pg->cp_state == CPS_PAGEOUT)
1330  *
1331  * \see cl_page_operations::cpo_flush()
1332  */
1333 int cl_page_flush(const struct lu_env *env, struct cl_io *io,
1334                   struct cl_page *pg)
1335 {
1336         int result;
1337
1338         PINVRNT(env, pg, cl_page_is_owned(pg, io));
1339         PINVRNT(env, pg, cl_page_invariant(pg));
1340
1341         result = cl_page_invoke(env, io, pg, CL_PAGE_OP(cpo_flush));
1342
1343         CL_PAGE_HEADER(D_TRACE, env, pg, "%d\n", result);
1344         return result;
1345 }
1346 EXPORT_SYMBOL(cl_page_flush);
1347
1348 /**
1349  * Checks whether page is protected by any extent lock is at least required
1350  * mode.
1351  *
1352  * \return the same as in cl_page_operations::cpo_is_under_lock() method.
1353  * \see cl_page_operations::cpo_is_under_lock()
1354  */
1355 int cl_page_is_under_lock(const struct lu_env *env, struct cl_io *io,
1356                           struct cl_page *page)
1357 {
1358         int rc;
1359
1360         PINVRNT(env, page, cl_page_invariant(page));
1361
1362         rc = CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_is_under_lock),
1363                             (const struct lu_env *,
1364                              const struct cl_page_slice *, struct cl_io *),
1365                             io);
1366         PASSERT(env, page, rc != 0);
1367         return rc;
1368 }
1369 EXPORT_SYMBOL(cl_page_is_under_lock);
1370
1371 static int page_prune_cb(const struct lu_env *env, struct cl_io *io,
1372                          struct cl_page *page, void *cbdata)
1373 {
1374         cl_page_own(env, io, page);
1375         cl_page_unmap(env, io, page);
1376         cl_page_discard(env, io, page);
1377         cl_page_disown(env, io, page);
1378         return CLP_GANG_OKAY;
1379 }
1380
1381 /**
1382  * Purges all cached pages belonging to the object \a obj.
1383  */
1384 int cl_pages_prune(const struct lu_env *env, struct cl_object *clobj)
1385 {
1386         struct cl_thread_info   *info;
1387         struct cl_object        *obj = cl_object_top(clobj);
1388         struct cl_io        *io;
1389         int                   result;
1390
1391         info  = cl_env_info(env);
1392         io    = &info->clt_io;
1393
1394         /*
1395          * initialize the io. This is ugly since we never do IO in this
1396          * function, we just make cl_page_list functions happy. -jay
1397          */
1398         io->ci_obj = obj;
1399         io->ci_ignore_layout = 1;
1400         result = cl_io_init(env, io, CIT_MISC, obj);
1401         if (result != 0) {
1402                 cl_io_fini(env, io);
1403                 return io->ci_result;
1404         }
1405
1406         do {
1407                 result = cl_page_gang_lookup(env, obj, io, 0, CL_PAGE_EOF,
1408                                              page_prune_cb, NULL);
1409                 if (result == CLP_GANG_RESCHED)
1410                         cond_resched();
1411         } while (result != CLP_GANG_OKAY);
1412
1413         cl_io_fini(env, io);
1414         return result;
1415 }
1416 EXPORT_SYMBOL(cl_pages_prune);
1417
1418 /**
1419  * Tells transfer engine that only part of a page is to be transmitted.
1420  *
1421  * \see cl_page_operations::cpo_clip()
1422  */
1423 void cl_page_clip(const struct lu_env *env, struct cl_page *pg,
1424                   int from, int to)
1425 {
1426         PINVRNT(env, pg, cl_page_invariant(pg));
1427
1428         CL_PAGE_HEADER(D_TRACE, env, pg, "%d %d\n", from, to);
1429         CL_PAGE_INVOID(env, pg, CL_PAGE_OP(cpo_clip),
1430                        (const struct lu_env *,
1431                         const struct cl_page_slice *, int, int),
1432                        from, to);
1433 }
1434 EXPORT_SYMBOL(cl_page_clip);
1435
1436 /**
1437  * Prints human readable representation of \a pg to the \a f.
1438  */
1439 void cl_page_header_print(const struct lu_env *env, void *cookie,
1440                           lu_printer_t printer, const struct cl_page *pg)
1441 {
1442         (*printer)(env, cookie,
1443                    "page@%p[%d %p:%lu ^%p_%p %d %d %d %p %p %#x]\n",
1444                    pg, atomic_read(&pg->cp_ref), pg->cp_obj,
1445                    pg->cp_index, pg->cp_parent, pg->cp_child,
1446                    pg->cp_state, pg->cp_error, pg->cp_type,
1447                    pg->cp_owner, pg->cp_req, pg->cp_flags);
1448 }
1449 EXPORT_SYMBOL(cl_page_header_print);
1450
1451 /**
1452  * Prints human readable representation of \a pg to the \a f.
1453  */
1454 void cl_page_print(const struct lu_env *env, void *cookie,
1455                    lu_printer_t printer, const struct cl_page *pg)
1456 {
1457         struct cl_page *scan;
1458
1459         for (scan = cl_page_top((struct cl_page *)pg);
1460              scan != NULL; scan = scan->cp_child)
1461                 cl_page_header_print(env, cookie, printer, scan);
1462         CL_PAGE_INVOKE(env, (struct cl_page *)pg, CL_PAGE_OP(cpo_print),
1463                        (const struct lu_env *env,
1464                         const struct cl_page_slice *slice,
1465                         void *cookie, lu_printer_t p), cookie, printer);
1466         (*printer)(env, cookie, "end page@%p\n", pg);
1467 }
1468 EXPORT_SYMBOL(cl_page_print);
1469
1470 /**
1471  * Cancel a page which is still in a transfer.
1472  */
1473 int cl_page_cancel(const struct lu_env *env, struct cl_page *page)
1474 {
1475         return CL_PAGE_INVOKE(env, page, CL_PAGE_OP(cpo_cancel),
1476                               (const struct lu_env *,
1477                                const struct cl_page_slice *));
1478 }
1479 EXPORT_SYMBOL(cl_page_cancel);
1480
1481 /**
1482  * Converts a byte offset within object \a obj into a page index.
1483  */
1484 loff_t cl_offset(const struct cl_object *obj, pgoff_t idx)
1485 {
1486         /*
1487          * XXX for now.
1488          */
1489         return (loff_t)idx << PAGE_CACHE_SHIFT;
1490 }
1491 EXPORT_SYMBOL(cl_offset);
1492
1493 /**
1494  * Converts a page index into a byte offset within object \a obj.
1495  */
1496 pgoff_t cl_index(const struct cl_object *obj, loff_t offset)
1497 {
1498         /*
1499          * XXX for now.
1500          */
1501         return offset >> PAGE_CACHE_SHIFT;
1502 }
1503 EXPORT_SYMBOL(cl_index);
1504
1505 int cl_page_size(const struct cl_object *obj)
1506 {
1507         return 1 << PAGE_CACHE_SHIFT;
1508 }
1509 EXPORT_SYMBOL(cl_page_size);
1510
1511 /**
1512  * Adds page slice to the compound page.
1513  *
1514  * This is called by cl_object_operations::coo_page_init() methods to add a
1515  * per-layer state to the page. New state is added at the end of
1516  * cl_page::cp_layers list, that is, it is at the bottom of the stack.
1517  *
1518  * \see cl_lock_slice_add(), cl_req_slice_add(), cl_io_slice_add()
1519  */
1520 void cl_page_slice_add(struct cl_page *page, struct cl_page_slice *slice,
1521                        struct cl_object *obj,
1522                        const struct cl_page_operations *ops)
1523 {
1524         list_add_tail(&slice->cpl_linkage, &page->cp_layers);
1525         slice->cpl_obj  = obj;
1526         slice->cpl_ops  = ops;
1527         slice->cpl_page = page;
1528 }
1529 EXPORT_SYMBOL(cl_page_slice_add);
1530
1531 int  cl_page_init(void)
1532 {
1533         return 0;
1534 }
1535
1536 void cl_page_fini(void)
1537 {
1538 }