linux/fs/cachefiles/rdwr.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-or-later
   2/* Storage object read/write
   3 *
   4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5 * Written by David Howells (dhowells@redhat.com)
   6 */
   7
   8#include <linux/mount.h>
   9#include <linux/slab.h>
  10#include <linux/file.h>
  11#include <linux/swap.h>
  12#include "internal.h"
  13
  14/*
  15 * detect wake up events generated by the unlocking of pages in which we're
  16 * interested
  17 * - we use this to detect read completion of backing pages
  18 * - the caller holds the waitqueue lock
  19 */
  20static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
  21                                  int sync, void *_key)
  22{
  23        struct cachefiles_one_read *monitor =
  24                container_of(wait, struct cachefiles_one_read, monitor);
  25        struct cachefiles_object *object;
  26        struct fscache_retrieval *op = monitor->op;
  27        struct wait_page_key *key = _key;
  28        struct page *page = wait->private;
  29
  30        ASSERT(key);
  31
  32        _enter("{%lu},%u,%d,{%p,%u}",
  33               monitor->netfs_page->index, mode, sync,
  34               key->page, key->bit_nr);
  35
  36        if (key->page != page || key->bit_nr != PG_locked)
  37                return 0;
  38
  39        _debug("--- monitor %p %lx ---", page, page->flags);
  40
  41        if (!PageUptodate(page) && !PageError(page)) {
  42                /* unlocked, not uptodate and not erronous? */
  43                _debug("page probably truncated");
  44        }
  45
  46        /* remove from the waitqueue */
  47        list_del(&wait->entry);
  48
  49        /* move onto the action list and queue for FS-Cache thread pool */
  50        ASSERT(op);
  51
  52        /* We need to temporarily bump the usage count as we don't own a ref
  53         * here otherwise cachefiles_read_copier() may free the op between the
  54         * monitor being enqueued on the op->to_do list and the op getting
  55         * enqueued on the work queue.
  56         */
  57        fscache_get_retrieval(op);
  58
  59        object = container_of(op->op.object, struct cachefiles_object, fscache);
  60        spin_lock(&object->work_lock);
  61        list_add_tail(&monitor->op_link, &op->to_do);
  62        fscache_enqueue_retrieval(op);
  63        spin_unlock(&object->work_lock);
  64
  65        fscache_put_retrieval(op);
  66        return 0;
  67}
  68
  69/*
  70 * handle a probably truncated page
  71 * - check to see if the page is still relevant and reissue the read if
  72 *   possible
  73 * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  74 *   must wait again and 0 if successful
  75 */
  76static int cachefiles_read_reissue(struct cachefiles_object *object,
  77                                   struct cachefiles_one_read *monitor)
  78{
  79        struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
  80        struct page *backpage = monitor->back_page, *backpage2;
  81        int ret;
  82
  83        _enter("{ino=%lx},{%lx,%lx}",
  84               d_backing_inode(object->backer)->i_ino,
  85               backpage->index, backpage->flags);
  86
  87        /* skip if the page was truncated away completely */
  88        if (backpage->mapping != bmapping) {
  89                _leave(" = -ENODATA [mapping]");
  90                return -ENODATA;
  91        }
  92
  93        backpage2 = find_get_page(bmapping, backpage->index);
  94        if (!backpage2) {
  95                _leave(" = -ENODATA [gone]");
  96                return -ENODATA;
  97        }
  98
  99        if (backpage != backpage2) {
 100                put_page(backpage2);
 101                _leave(" = -ENODATA [different]");
 102                return -ENODATA;
 103        }
 104
 105        /* the page is still there and we already have a ref on it, so we don't
 106         * need a second */
 107        put_page(backpage2);
 108
 109        INIT_LIST_HEAD(&monitor->op_link);
 110        add_page_wait_queue(backpage, &monitor->monitor);
 111
 112        if (trylock_page(backpage)) {
 113                ret = -EIO;
 114                if (PageError(backpage))
 115                        goto unlock_discard;
 116                ret = 0;
 117                if (PageUptodate(backpage))
 118                        goto unlock_discard;
 119
 120                _debug("reissue read");
 121                ret = bmapping->a_ops->readpage(NULL, backpage);
 122                if (ret < 0)
 123                        goto discard;
 124        }
 125
 126        /* but the page may have been read before the monitor was installed, so
 127         * the monitor may miss the event - so we have to ensure that we do get
 128         * one in such a case */
 129        if (trylock_page(backpage)) {
 130                _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 131                unlock_page(backpage);
 132        }
 133
 134        /* it'll reappear on the todo list */
 135        _leave(" = -EINPROGRESS");
 136        return -EINPROGRESS;
 137
 138unlock_discard:
 139        unlock_page(backpage);
 140discard:
 141        spin_lock_irq(&object->work_lock);
 142        list_del(&monitor->op_link);
 143        spin_unlock_irq(&object->work_lock);
 144        _leave(" = %d", ret);
 145        return ret;
 146}
 147
 148/*
 149 * copy data from backing pages to netfs pages to complete a read operation
 150 * - driven by FS-Cache's thread pool
 151 */
 152static void cachefiles_read_copier(struct fscache_operation *_op)
 153{
 154        struct cachefiles_one_read *monitor;
 155        struct cachefiles_object *object;
 156        struct fscache_retrieval *op;
 157        int error, max;
 158
 159        op = container_of(_op, struct fscache_retrieval, op);
 160        object = container_of(op->op.object,
 161                              struct cachefiles_object, fscache);
 162
 163        _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 164
 165        max = 8;
 166        spin_lock_irq(&object->work_lock);
 167
 168        while (!list_empty(&op->to_do)) {
 169                monitor = list_entry(op->to_do.next,
 170                                     struct cachefiles_one_read, op_link);
 171                list_del(&monitor->op_link);
 172
 173                spin_unlock_irq(&object->work_lock);
 174
 175                _debug("- copy {%lu}", monitor->back_page->index);
 176
 177        recheck:
 178                if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 179                             &object->fscache.cookie->flags)) {
 180                        error = -ESTALE;
 181                } else if (PageUptodate(monitor->back_page)) {
 182                        copy_highpage(monitor->netfs_page, monitor->back_page);
 183                        fscache_mark_page_cached(monitor->op,
 184                                                 monitor->netfs_page);
 185                        error = 0;
 186                } else if (!PageError(monitor->back_page)) {
 187                        /* the page has probably been truncated */
 188                        error = cachefiles_read_reissue(object, monitor);
 189                        if (error == -EINPROGRESS)
 190                                goto next;
 191                        goto recheck;
 192                } else {
 193                        cachefiles_io_error_obj(
 194                                object,
 195                                "Readpage failed on backing file %lx",
 196                                (unsigned long) monitor->back_page->flags);
 197                        error = -EIO;
 198                }
 199
 200                put_page(monitor->back_page);
 201
 202                fscache_end_io(op, monitor->netfs_page, error);
 203                put_page(monitor->netfs_page);
 204                fscache_retrieval_complete(op, 1);
 205                fscache_put_retrieval(op);
 206                kfree(monitor);
 207
 208        next:
 209                /* let the thread pool have some air occasionally */
 210                max--;
 211                if (max < 0 || need_resched()) {
 212                        if (!list_empty(&op->to_do))
 213                                fscache_enqueue_retrieval(op);
 214                        _leave(" [maxed out]");
 215                        return;
 216                }
 217
 218                spin_lock_irq(&object->work_lock);
 219        }
 220
 221        spin_unlock_irq(&object->work_lock);
 222        _leave("");
 223}
 224
 225/*
 226 * read the corresponding page to the given set from the backing file
 227 * - an uncertain page is simply discarded, to be tried again another time
 228 */
 229static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 230                                            struct fscache_retrieval *op,
 231                                            struct page *netpage)
 232{
 233        struct cachefiles_one_read *monitor;
 234        struct address_space *bmapping;
 235        struct page *newpage, *backpage;
 236        int ret;
 237
 238        _enter("");
 239
 240        _debug("read back %p{%lu,%d}",
 241               netpage, netpage->index, page_count(netpage));
 242
 243        monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 244        if (!monitor)
 245                goto nomem;
 246
 247        monitor->netfs_page = netpage;
 248        monitor->op = fscache_get_retrieval(op);
 249
 250        init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 251
 252        /* attempt to get hold of the backing page */
 253        bmapping = d_backing_inode(object->backer)->i_mapping;
 254        newpage = NULL;
 255
 256        for (;;) {
 257                backpage = find_get_page(bmapping, netpage->index);
 258                if (backpage)
 259                        goto backing_page_already_present;
 260
 261                if (!newpage) {
 262                        newpage = __page_cache_alloc(cachefiles_gfp);
 263                        if (!newpage)
 264                                goto nomem_monitor;
 265                }
 266
 267                ret = add_to_page_cache_lru(newpage, bmapping,
 268                                            netpage->index, cachefiles_gfp);
 269                if (ret == 0)
 270                        goto installed_new_backing_page;
 271                if (ret != -EEXIST)
 272                        goto nomem_page;
 273        }
 274
 275        /* we've installed a new backing page, so now we need to start
 276         * it reading */
 277installed_new_backing_page:
 278        _debug("- new %p", newpage);
 279
 280        backpage = newpage;
 281        newpage = NULL;
 282
 283read_backing_page:
 284        ret = bmapping->a_ops->readpage(NULL, backpage);
 285        if (ret < 0)
 286                goto read_error;
 287
 288        /* set the monitor to transfer the data across */
 289monitor_backing_page:
 290        _debug("- monitor add");
 291
 292        /* install the monitor */
 293        get_page(monitor->netfs_page);
 294        get_page(backpage);
 295        monitor->back_page = backpage;
 296        monitor->monitor.private = backpage;
 297        add_page_wait_queue(backpage, &monitor->monitor);
 298        monitor = NULL;
 299
 300        /* but the page may have been read before the monitor was installed, so
 301         * the monitor may miss the event - so we have to ensure that we do get
 302         * one in such a case */
 303        if (trylock_page(backpage)) {
 304                _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 305                unlock_page(backpage);
 306        }
 307        goto success;
 308
 309        /* if the backing page is already present, it can be in one of
 310         * three states: read in progress, read failed or read okay */
 311backing_page_already_present:
 312        _debug("- present");
 313
 314        if (newpage) {
 315                put_page(newpage);
 316                newpage = NULL;
 317        }
 318
 319        if (PageError(backpage))
 320                goto io_error;
 321
 322        if (PageUptodate(backpage))
 323                goto backing_page_already_uptodate;
 324
 325        if (!trylock_page(backpage))
 326                goto monitor_backing_page;
 327        _debug("read %p {%lx}", backpage, backpage->flags);
 328        goto read_backing_page;
 329
 330        /* the backing page is already up to date, attach the netfs
 331         * page to the pagecache and LRU and copy the data across */
 332backing_page_already_uptodate:
 333        _debug("- uptodate");
 334
 335        fscache_mark_page_cached(op, netpage);
 336
 337        copy_highpage(netpage, backpage);
 338        fscache_end_io(op, netpage, 0);
 339        fscache_retrieval_complete(op, 1);
 340
 341success:
 342        _debug("success");
 343        ret = 0;
 344
 345out:
 346        if (backpage)
 347                put_page(backpage);
 348        if (monitor) {
 349                fscache_put_retrieval(monitor->op);
 350                kfree(monitor);
 351        }
 352        _leave(" = %d", ret);
 353        return ret;
 354
 355read_error:
 356        _debug("read error %d", ret);
 357        if (ret == -ENOMEM) {
 358                fscache_retrieval_complete(op, 1);
 359                goto out;
 360        }
 361io_error:
 362        cachefiles_io_error_obj(object, "Page read error on backing file");
 363        fscache_retrieval_complete(op, 1);
 364        ret = -ENOBUFS;
 365        goto out;
 366
 367nomem_page:
 368        put_page(newpage);
 369nomem_monitor:
 370        fscache_put_retrieval(monitor->op);
 371        kfree(monitor);
 372nomem:
 373        fscache_retrieval_complete(op, 1);
 374        _leave(" = -ENOMEM");
 375        return -ENOMEM;
 376}
 377
 378/*
 379 * read a page from the cache or allocate a block in which to store it
 380 * - cache withdrawal is prevented by the caller
 381 * - returns -EINTR if interrupted
 382 * - returns -ENOMEM if ran out of memory
 383 * - returns -ENOBUFS if no buffers can be made available
 384 * - returns -ENOBUFS if page is beyond EOF
 385 * - if the page is backed by a block in the cache:
 386 *   - a read will be started which will call the callback on completion
 387 *   - 0 will be returned
 388 * - else if the page is unbacked:
 389 *   - the metadata will be retained
 390 *   - -ENODATA will be returned
 391 */
 392int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 393                                  struct page *page,
 394                                  gfp_t gfp)
 395{
 396        struct cachefiles_object *object;
 397        struct cachefiles_cache *cache;
 398        struct inode *inode;
 399        sector_t block;
 400        unsigned shift;
 401        int ret, ret2;
 402
 403        object = container_of(op->op.object,
 404                              struct cachefiles_object, fscache);
 405        cache = container_of(object->fscache.cache,
 406                             struct cachefiles_cache, cache);
 407
 408        _enter("{%p},{%lx},,,", object, page->index);
 409
 410        if (!object->backer)
 411                goto enobufs;
 412
 413        inode = d_backing_inode(object->backer);
 414        ASSERT(S_ISREG(inode->i_mode));
 415
 416        /* calculate the shift required to use bmap */
 417        shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 418
 419        op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 420        op->op.flags |= FSCACHE_OP_ASYNC;
 421        op->op.processor = cachefiles_read_copier;
 422
 423        /* we assume the absence or presence of the first block is a good
 424         * enough indication for the page as a whole
 425         * - TODO: don't use bmap() for this as it is _not_ actually good
 426         *   enough for this as it doesn't indicate errors, but it's all we've
 427         *   got for the moment
 428         */
 429        block = page->index;
 430        block <<= shift;
 431
 432        ret2 = bmap(inode, &block);
 433        ASSERT(ret2 == 0);
 434
 435        _debug("%llx -> %llx",
 436               (unsigned long long) (page->index << shift),
 437               (unsigned long long) block);
 438
 439        if (block) {
 440                /* submit the apparently valid page to the backing fs to be
 441                 * read from disk */
 442                ret = cachefiles_read_backing_file_one(object, op, page);
 443        } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 444                /* there's space in the cache we can use */
 445                fscache_mark_page_cached(op, page);
 446                fscache_retrieval_complete(op, 1);
 447                ret = -ENODATA;
 448        } else {
 449                goto enobufs;
 450        }
 451
 452        _leave(" = %d", ret);
 453        return ret;
 454
 455enobufs:
 456        fscache_retrieval_complete(op, 1);
 457        _leave(" = -ENOBUFS");
 458        return -ENOBUFS;
 459}
 460
 461/*
 462 * read the corresponding pages to the given set from the backing file
 463 * - any uncertain pages are simply discarded, to be tried again another time
 464 */
 465static int cachefiles_read_backing_file(struct cachefiles_object *object,
 466                                        struct fscache_retrieval *op,
 467                                        struct list_head *list)
 468{
 469        struct cachefiles_one_read *monitor = NULL;
 470        struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 471        struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 472        int ret = 0;
 473
 474        _enter("");
 475
 476        list_for_each_entry_safe(netpage, _n, list, lru) {
 477                list_del(&netpage->lru);
 478
 479                _debug("read back %p{%lu,%d}",
 480                       netpage, netpage->index, page_count(netpage));
 481
 482                if (!monitor) {
 483                        monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 484                        if (!monitor)
 485                                goto nomem;
 486
 487                        monitor->op = fscache_get_retrieval(op);
 488                        init_waitqueue_func_entry(&monitor->monitor,
 489                                                  cachefiles_read_waiter);
 490                }
 491
 492                for (;;) {
 493                        backpage = find_get_page(bmapping, netpage->index);
 494                        if (backpage)
 495                                goto backing_page_already_present;
 496
 497                        if (!newpage) {
 498                                newpage = __page_cache_alloc(cachefiles_gfp);
 499                                if (!newpage)
 500                                        goto nomem;
 501                        }
 502
 503                        ret = add_to_page_cache_lru(newpage, bmapping,
 504                                                    netpage->index,
 505                                                    cachefiles_gfp);
 506                        if (ret == 0)
 507                                goto installed_new_backing_page;
 508                        if (ret != -EEXIST)
 509                                goto nomem;
 510                }
 511
 512                /* we've installed a new backing page, so now we need
 513                 * to start it reading */
 514        installed_new_backing_page:
 515                _debug("- new %p", newpage);
 516
 517                backpage = newpage;
 518                newpage = NULL;
 519
 520        reread_backing_page:
 521                ret = bmapping->a_ops->readpage(NULL, backpage);
 522                if (ret < 0)
 523                        goto read_error;
 524
 525                /* add the netfs page to the pagecache and LRU, and set the
 526                 * monitor to transfer the data across */
 527        monitor_backing_page:
 528                _debug("- monitor add");
 529
 530                ret = add_to_page_cache_lru(netpage, op->mapping,
 531                                            netpage->index, cachefiles_gfp);
 532                if (ret < 0) {
 533                        if (ret == -EEXIST) {
 534                                put_page(backpage);
 535                                backpage = NULL;
 536                                put_page(netpage);
 537                                netpage = NULL;
 538                                fscache_retrieval_complete(op, 1);
 539                                continue;
 540                        }
 541                        goto nomem;
 542                }
 543
 544                /* install a monitor */
 545                get_page(netpage);
 546                monitor->netfs_page = netpage;
 547
 548                get_page(backpage);
 549                monitor->back_page = backpage;
 550                monitor->monitor.private = backpage;
 551                add_page_wait_queue(backpage, &monitor->monitor);
 552                monitor = NULL;
 553
 554                /* but the page may have been read before the monitor was
 555                 * installed, so the monitor may miss the event - so we have to
 556                 * ensure that we do get one in such a case */
 557                if (trylock_page(backpage)) {
 558                        _debug("2unlock %p {%lx}", backpage, backpage->flags);
 559                        unlock_page(backpage);
 560                }
 561
 562                put_page(backpage);
 563                backpage = NULL;
 564
 565                put_page(netpage);
 566                netpage = NULL;
 567                continue;
 568
 569                /* if the backing page is already present, it can be in one of
 570                 * three states: read in progress, read failed or read okay */
 571        backing_page_already_present:
 572                _debug("- present %p", backpage);
 573
 574                if (PageError(backpage))
 575                        goto io_error;
 576
 577                if (PageUptodate(backpage))
 578                        goto backing_page_already_uptodate;
 579
 580                _debug("- not ready %p{%lx}", backpage, backpage->flags);
 581
 582                if (!trylock_page(backpage))
 583                        goto monitor_backing_page;
 584
 585                if (PageError(backpage)) {
 586                        _debug("error %lx", backpage->flags);
 587                        unlock_page(backpage);
 588                        goto io_error;
 589                }
 590
 591                if (PageUptodate(backpage))
 592                        goto backing_page_already_uptodate_unlock;
 593
 594                /* we've locked a page that's neither up to date nor erroneous,
 595                 * so we need to attempt to read it again */
 596                goto reread_backing_page;
 597
 598                /* the backing page is already up to date, attach the netfs
 599                 * page to the pagecache and LRU and copy the data across */
 600        backing_page_already_uptodate_unlock:
 601                _debug("uptodate %lx", backpage->flags);
 602                unlock_page(backpage);
 603        backing_page_already_uptodate:
 604                _debug("- uptodate");
 605
 606                ret = add_to_page_cache_lru(netpage, op->mapping,
 607                                            netpage->index, cachefiles_gfp);
 608                if (ret < 0) {
 609                        if (ret == -EEXIST) {
 610                                put_page(backpage);
 611                                backpage = NULL;
 612                                put_page(netpage);
 613                                netpage = NULL;
 614                                fscache_retrieval_complete(op, 1);
 615                                continue;
 616                        }
 617                        goto nomem;
 618                }
 619
 620                copy_highpage(netpage, backpage);
 621
 622                put_page(backpage);
 623                backpage = NULL;
 624
 625                fscache_mark_page_cached(op, netpage);
 626
 627                /* the netpage is unlocked and marked up to date here */
 628                fscache_end_io(op, netpage, 0);
 629                put_page(netpage);
 630                netpage = NULL;
 631                fscache_retrieval_complete(op, 1);
 632                continue;
 633        }
 634
 635        netpage = NULL;
 636
 637        _debug("out");
 638
 639out:
 640        /* tidy up */
 641        if (newpage)
 642                put_page(newpage);
 643        if (netpage)
 644                put_page(netpage);
 645        if (backpage)
 646                put_page(backpage);
 647        if (monitor) {
 648                fscache_put_retrieval(op);
 649                kfree(monitor);
 650        }
 651
 652        list_for_each_entry_safe(netpage, _n, list, lru) {
 653                list_del(&netpage->lru);
 654                put_page(netpage);
 655                fscache_retrieval_complete(op, 1);
 656        }
 657
 658        _leave(" = %d", ret);
 659        return ret;
 660
 661nomem:
 662        _debug("nomem");
 663        ret = -ENOMEM;
 664        goto record_page_complete;
 665
 666read_error:
 667        _debug("read error %d", ret);
 668        if (ret == -ENOMEM)
 669                goto record_page_complete;
 670io_error:
 671        cachefiles_io_error_obj(object, "Page read error on backing file");
 672        ret = -ENOBUFS;
 673record_page_complete:
 674        fscache_retrieval_complete(op, 1);
 675        goto out;
 676}
 677
 678/*
 679 * read a list of pages from the cache or allocate blocks in which to store
 680 * them
 681 */
 682int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 683                                   struct list_head *pages,
 684                                   unsigned *nr_pages,
 685                                   gfp_t gfp)
 686{
 687        struct cachefiles_object *object;
 688        struct cachefiles_cache *cache;
 689        struct list_head backpages;
 690        struct pagevec pagevec;
 691        struct inode *inode;
 692        struct page *page, *_n;
 693        unsigned shift, nrbackpages;
 694        int ret, ret2, space;
 695
 696        object = container_of(op->op.object,
 697                              struct cachefiles_object, fscache);
 698        cache = container_of(object->fscache.cache,
 699                             struct cachefiles_cache, cache);
 700
 701        _enter("{OBJ%x,%d},,%d,,",
 702               object->fscache.debug_id, atomic_read(&op->op.usage),
 703               *nr_pages);
 704
 705        if (!object->backer)
 706                goto all_enobufs;
 707
 708        space = 1;
 709        if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 710                space = 0;
 711
 712        inode = d_backing_inode(object->backer);
 713        ASSERT(S_ISREG(inode->i_mode));
 714
 715        /* calculate the shift required to use bmap */
 716        shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 717
 718        pagevec_init(&pagevec);
 719
 720        op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 721        op->op.flags |= FSCACHE_OP_ASYNC;
 722        op->op.processor = cachefiles_read_copier;
 723
 724        INIT_LIST_HEAD(&backpages);
 725        nrbackpages = 0;
 726
 727        ret = space ? -ENODATA : -ENOBUFS;
 728        list_for_each_entry_safe(page, _n, pages, lru) {
 729                sector_t block;
 730
 731                /* we assume the absence or presence of the first block is a
 732                 * good enough indication for the page as a whole
 733                 * - TODO: don't use bmap() for this as it is _not_ actually
 734                 *   good enough for this as it doesn't indicate errors, but
 735                 *   it's all we've got for the moment
 736                 */
 737                block = page->index;
 738                block <<= shift;
 739
 740                ret2 = bmap(inode, &block);
 741                ASSERT(ret2 == 0);
 742
 743                _debug("%llx -> %llx",
 744                       (unsigned long long) (page->index << shift),
 745                       (unsigned long long) block);
 746
 747                if (block) {
 748                        /* we have data - add it to the list to give to the
 749                         * backing fs */
 750                        list_move(&page->lru, &backpages);
 751                        (*nr_pages)--;
 752                        nrbackpages++;
 753                } else if (space && pagevec_add(&pagevec, page) == 0) {
 754                        fscache_mark_pages_cached(op, &pagevec);
 755                        fscache_retrieval_complete(op, 1);
 756                        ret = -ENODATA;
 757                } else {
 758                        fscache_retrieval_complete(op, 1);
 759                }
 760        }
 761
 762        if (pagevec_count(&pagevec) > 0)
 763                fscache_mark_pages_cached(op, &pagevec);
 764
 765        if (list_empty(pages))
 766                ret = 0;
 767
 768        /* submit the apparently valid pages to the backing fs to be read from
 769         * disk */
 770        if (nrbackpages > 0) {
 771                ret2 = cachefiles_read_backing_file(object, op, &backpages);
 772                if (ret2 == -ENOMEM || ret2 == -EINTR)
 773                        ret = ret2;
 774        }
 775
 776        _leave(" = %d [nr=%u%s]",
 777               ret, *nr_pages, list_empty(pages) ? " empty" : "");
 778        return ret;
 779
 780all_enobufs:
 781        fscache_retrieval_complete(op, *nr_pages);
 782        return -ENOBUFS;
 783}
 784
 785/*
 786 * allocate a block in the cache in which to store a page
 787 * - cache withdrawal is prevented by the caller
 788 * - returns -EINTR if interrupted
 789 * - returns -ENOMEM if ran out of memory
 790 * - returns -ENOBUFS if no buffers can be made available
 791 * - returns -ENOBUFS if page is beyond EOF
 792 * - otherwise:
 793 *   - the metadata will be retained
 794 *   - 0 will be returned
 795 */
 796int cachefiles_allocate_page(struct fscache_retrieval *op,
 797                             struct page *page,
 798                             gfp_t gfp)
 799{
 800        struct cachefiles_object *object;
 801        struct cachefiles_cache *cache;
 802        int ret;
 803
 804        object = container_of(op->op.object,
 805                              struct cachefiles_object, fscache);
 806        cache = container_of(object->fscache.cache,
 807                             struct cachefiles_cache, cache);
 808
 809        _enter("%p,{%lx},", object, page->index);
 810
 811        ret = cachefiles_has_space(cache, 0, 1);
 812        if (ret == 0)
 813                fscache_mark_page_cached(op, page);
 814        else
 815                ret = -ENOBUFS;
 816
 817        fscache_retrieval_complete(op, 1);
 818        _leave(" = %d", ret);
 819        return ret;
 820}
 821
 822/*
 823 * allocate blocks in the cache in which to store a set of pages
 824 * - cache withdrawal is prevented by the caller
 825 * - returns -EINTR if interrupted
 826 * - returns -ENOMEM if ran out of memory
 827 * - returns -ENOBUFS if some buffers couldn't be made available
 828 * - returns -ENOBUFS if some pages are beyond EOF
 829 * - otherwise:
 830 *   - -ENODATA will be returned
 831 * - metadata will be retained for any page marked
 832 */
 833int cachefiles_allocate_pages(struct fscache_retrieval *op,
 834                              struct list_head *pages,
 835                              unsigned *nr_pages,
 836                              gfp_t gfp)
 837{
 838        struct cachefiles_object *object;
 839        struct cachefiles_cache *cache;
 840        struct pagevec pagevec;
 841        struct page *page;
 842        int ret;
 843
 844        object = container_of(op->op.object,
 845                              struct cachefiles_object, fscache);
 846        cache = container_of(object->fscache.cache,
 847                             struct cachefiles_cache, cache);
 848
 849        _enter("%p,,,%d,", object, *nr_pages);
 850
 851        ret = cachefiles_has_space(cache, 0, *nr_pages);
 852        if (ret == 0) {
 853                pagevec_init(&pagevec);
 854
 855                list_for_each_entry(page, pages, lru) {
 856                        if (pagevec_add(&pagevec, page) == 0)
 857                                fscache_mark_pages_cached(op, &pagevec);
 858                }
 859
 860                if (pagevec_count(&pagevec) > 0)
 861                        fscache_mark_pages_cached(op, &pagevec);
 862                ret = -ENODATA;
 863        } else {
 864                ret = -ENOBUFS;
 865        }
 866
 867        fscache_retrieval_complete(op, *nr_pages);
 868        _leave(" = %d", ret);
 869        return ret;
 870}
 871
 872/*
 873 * request a page be stored in the cache
 874 * - cache withdrawal is prevented by the caller
 875 * - this request may be ignored if there's no cache block available, in which
 876 *   case -ENOBUFS will be returned
 877 * - if the op is in progress, 0 will be returned
 878 */
 879int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 880{
 881        struct cachefiles_object *object;
 882        struct cachefiles_cache *cache;
 883        struct file *file;
 884        struct path path;
 885        loff_t pos, eof;
 886        size_t len;
 887        void *data;
 888        int ret = -ENOBUFS;
 889
 890        ASSERT(op != NULL);
 891        ASSERT(page != NULL);
 892
 893        object = container_of(op->op.object,
 894                              struct cachefiles_object, fscache);
 895
 896        _enter("%p,%p{%lx},,,", object, page, page->index);
 897
 898        if (!object->backer) {
 899                _leave(" = -ENOBUFS");
 900                return -ENOBUFS;
 901        }
 902
 903        ASSERT(d_is_reg(object->backer));
 904
 905        cache = container_of(object->fscache.cache,
 906                             struct cachefiles_cache, cache);
 907
 908        pos = (loff_t)page->index << PAGE_SHIFT;
 909
 910        /* We mustn't write more data than we have, so we have to beware of a
 911         * partial page at EOF.
 912         */
 913        eof = object->fscache.store_limit_l;
 914        if (pos >= eof)
 915                goto error;
 916
 917        /* write the page to the backing filesystem and let it store it in its
 918         * own time */
 919        path.mnt = cache->mnt;
 920        path.dentry = object->backer;
 921        file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 922        if (IS_ERR(file)) {
 923                ret = PTR_ERR(file);
 924                goto error_2;
 925        }
 926
 927        len = PAGE_SIZE;
 928        if (eof & ~PAGE_MASK) {
 929                if (eof - pos < PAGE_SIZE) {
 930                        _debug("cut short %llx to %llx",
 931                               pos, eof);
 932                        len = eof - pos;
 933                        ASSERTCMP(pos + len, ==, eof);
 934                }
 935        }
 936
 937        data = kmap(page);
 938        ret = kernel_write(file, data, len, &pos);
 939        kunmap(page);
 940        fput(file);
 941        if (ret != len)
 942                goto error_eio;
 943
 944        _leave(" = 0");
 945        return 0;
 946
 947error_eio:
 948        ret = -EIO;
 949error_2:
 950        if (ret == -EIO)
 951                cachefiles_io_error_obj(object,
 952                                        "Write page to backing file failed");
 953error:
 954        _leave(" = -ENOBUFS [%d]", ret);
 955        return -ENOBUFS;
 956}
 957
 958/*
 959 * detach a backing block from a page
 960 * - cache withdrawal is prevented by the caller
 961 */
 962void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 963        __releases(&object->fscache.cookie->lock)
 964{
 965        struct cachefiles_object *object;
 966
 967        object = container_of(_object, struct cachefiles_object, fscache);
 968
 969        _enter("%p,{%lu}", object, page->index);
 970
 971        spin_unlock(&object->fscache.cookie->lock);
 972}
 973