fs/cachefiles/rdwr.c

   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  */
  20 static 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_bit_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->flags, key->bit_nr);
  35
  36         if (key->flags != &page->flags ||
  37             key->bit_nr != PG_locked)
  38                 return 0;
  39
  40         _debug("--- monitor %p %lx ---", page, page->flags);
  41
  42         if (!PageUptodate(page) && !PageError(page)) {
  43                 /* unlocked, not uptodate and not erronous? */
  44                 _debug("page probably truncated");
  45         }
  46
  47         /* remove from the waitqueue */
  48         list_del(&wait->entry);
  49
  50         /* move onto the action list and queue for FS-Cache thread pool */
  51         ASSERT(op);
  52
  53         /* We need to temporarily bump the usage count as we don't own a ref
  54          * here otherwise cachefiles_read_copier() may free the op between the
  55          * monitor being enqueued on the op->to_do list and the op getting
  56          * enqueued on the work queue.
  57          */
  58         fscache_get_retrieval(op);
  59
  60         object = container_of(op->op.object, struct cachefiles_object, fscache);
  61         spin_lock(&object->work_lock);
  62         list_add_tail(&monitor->op_link, &op->to_do);
  63         fscache_enqueue_retrieval(op);
  64         spin_unlock(&object->work_lock);
  65
  66         fscache_put_retrieval(op);
  67         return 0;
  68 }
  69
  70 /*
  71  * handle a probably truncated page
  72  * - check to see if the page is still relevant and reissue the read if
  73  *   possible
  74  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  75  *   must wait again and 0 if successful
  76  */
  77 static int cachefiles_read_reissue(struct cachefiles_object *object,
  78                                    struct cachefiles_one_read *monitor)
  79 {
  80         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
  81         struct page *backpage = monitor->back_page, *backpage2;
  82         int ret;
  83
  84         _enter("{ino=%lx},{%lx,%lx}",
  85                d_backing_inode(object->backer)->i_ino,
  86                backpage->index, backpage->flags);
  87
  88         /* skip if the page was truncated away completely */
  89         if (backpage->mapping != bmapping) {
  90                 _leave(" = -ENODATA [mapping]");
  91                 return -ENODATA;
  92         }
  93
  94         backpage2 = find_get_page(bmapping, backpage->index);
  95         if (!backpage2) {
  96                 _leave(" = -ENODATA [gone]");
  97                 return -ENODATA;
  98         }
  99
 100         if (backpage != backpage2) {
 101                 put_page(backpage2);
 102                 _leave(" = -ENODATA [different]");
 103                 return -ENODATA;
 104         }
 105
 106         /* the page is still there and we already have a ref on it, so we don't
 107          * need a second */
 108         put_page(backpage2);
 109
 110         INIT_LIST_HEAD(&monitor->op_link);
 111         add_page_wait_queue(backpage, &monitor->monitor);
 112
 113         if (trylock_page(backpage)) {
 114                 ret = -EIO;
 115                 if (PageError(backpage))
 116                         goto unlock_discard;
 117                 ret = 0;
 118                 if (PageUptodate(backpage))
 119                         goto unlock_discard;
 120
 121                 _debug("reissue read");
 122                 ret = bmapping->a_ops->readpage(NULL, backpage);
 123                 if (ret < 0)
 124                         goto discard;
 125         }
 126
 127         /* but the page may have been read before the monitor was installed, so
 128          * the monitor may miss the event - so we have to ensure that we do get
 129          * one in such a case */
 130         if (trylock_page(backpage)) {
 131                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 132                 unlock_page(backpage);
 133         }
 134
 135         /* it'll reappear on the todo list */
 136         _leave(" = -EINPROGRESS");
 137         return -EINPROGRESS;
 138
 139 unlock_discard:
 140         unlock_page(backpage);
 141 discard:
 142         spin_lock_irq(&object->work_lock);
 143         list_del(&monitor->op_link);
 144         spin_unlock_irq(&object->work_lock);
 145         _leave(" = %d", ret);
 146         return ret;
 147 }
 148
 149 /*
 150  * copy data from backing pages to netfs pages to complete a read operation
 151  * - driven by FS-Cache's thread pool
 152  */
 153 static void cachefiles_read_copier(struct fscache_operation *_op)
 154 {
 155         struct cachefiles_one_read *monitor;
 156         struct cachefiles_object *object;
 157         struct fscache_retrieval *op;
 158         int error, max;
 159
 160         op = container_of(_op, struct fscache_retrieval, op);
 161         object = container_of(op->op.object,
 162                               struct cachefiles_object, fscache);
 163
 164         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 165
 166         max = 8;
 167         spin_lock_irq(&object->work_lock);
 168
 169         while (!list_empty(&op->to_do)) {
 170                 monitor = list_entry(op->to_do.next,
 171                                      struct cachefiles_one_read, op_link);
 172                 list_del(&monitor->op_link);
 173
 174                 spin_unlock_irq(&object->work_lock);
 175
 176                 _debug("- copy {%lu}", monitor->back_page->index);
 177
 178         recheck:
 179                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 180                              &object->fscache.cookie->flags)) {
 181                         error = -ESTALE;
 182                 } else if (PageUptodate(monitor->back_page)) {
 183                         copy_highpage(monitor->netfs_page, monitor->back_page);
 184                         fscache_mark_page_cached(monitor->op,
 185                                                  monitor->netfs_page);
 186                         error = 0;
 187                 } else if (!PageError(monitor->back_page)) {
 188                         /* the page has probably been truncated */
 189                         error = cachefiles_read_reissue(object, monitor);
 190                         if (error == -EINPROGRESS)
 191                                 goto next;
 192                         goto recheck;
 193                 } else {
 194                         cachefiles_io_error_obj(
 195                                 object,
 196                                 "Readpage failed on backing file %lx",
 197                                 (unsigned long) monitor->back_page->flags);
 198                         error = -EIO;
 199                 }
 200
 201                 put_page(monitor->back_page);
 202
 203                 fscache_end_io(op, monitor->netfs_page, error);
 204                 put_page(monitor->netfs_page);
 205                 fscache_retrieval_complete(op, 1);
 206                 fscache_put_retrieval(op);
 207                 kfree(monitor);
 208
 209         next:
 210                 /* let the thread pool have some air occasionally */
 211                 max--;
 212                 if (max < 0 || need_resched()) {
 213                         if (!list_empty(&op->to_do))
 214                                 fscache_enqueue_retrieval(op);
 215                         _leave(" [maxed out]");
 216                         return;
 217                 }
 218
 219                 spin_lock_irq(&object->work_lock);
 220         }
 221
 222         spin_unlock_irq(&object->work_lock);
 223         _leave("");
 224 }
 225
 226 /*
 227  * read the corresponding page to the given set from the backing file
 228  * - an uncertain page is simply discarded, to be tried again another time
 229  */
 230 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 231                                             struct fscache_retrieval *op,
 232                                             struct page *netpage)
 233 {
 234         struct cachefiles_one_read *monitor;
 235         struct address_space *bmapping;
 236         struct page *newpage, *backpage;
 237         int ret;
 238
 239         _enter("");
 240
 241         _debug("read back %p{%lu,%d}",
 242                netpage, netpage->index, page_count(netpage));
 243
 244         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 245         if (!monitor)
 246                 goto nomem;
 247
 248         monitor->netfs_page = netpage;
 249         monitor->op = fscache_get_retrieval(op);
 250
 251         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 252
 253         /* attempt to get hold of the backing page */
 254         bmapping = d_backing_inode(object->backer)->i_mapping;
 255         newpage = NULL;
 256
 257         for (;;) {
 258                 backpage = find_get_page(bmapping, netpage->index);
 259                 if (backpage)
 260                         goto backing_page_already_present;
 261
 262                 if (!newpage) {
 263                         newpage = __page_cache_alloc(cachefiles_gfp);
 264                         if (!newpage)
 265                                 goto nomem_monitor;
 266                 }
 267
 268                 ret = add_to_page_cache_lru(newpage, bmapping,
 269                                             netpage->index, cachefiles_gfp);
 270                 if (ret == 0)
 271                         goto installed_new_backing_page;
 272                 if (ret != -EEXIST)
 273                         goto nomem_page;
 274         }
 275
 276         /* we've installed a new backing page, so now we need to start
 277          * it reading */
 278 installed_new_backing_page:
 279         _debug("- new %p", newpage);
 280
 281         backpage = newpage;
 282         newpage = NULL;
 283
 284 read_backing_page:
 285         ret = bmapping->a_ops->readpage(NULL, backpage);
 286         if (ret < 0)
 287                 goto read_error;
 288
 289         /* set the monitor to transfer the data across */
 290 monitor_backing_page:
 291         _debug("- monitor add");
 292
 293         /* install the monitor */
 294         get_page(monitor->netfs_page);
 295         get_page(backpage);
 296         monitor->back_page = backpage;
 297         monitor->monitor.private = backpage;
 298         add_page_wait_queue(backpage, &monitor->monitor);
 299         monitor = NULL;
 300
 301         /* but the page may have been read before the monitor was installed, so
 302          * the monitor may miss the event - so we have to ensure that we do get
 303          * one in such a case */
 304         if (trylock_page(backpage)) {
 305                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 306                 unlock_page(backpage);
 307         }
 308         goto success;
 309
 310         /* if the backing page is already present, it can be in one of
 311          * three states: read in progress, read failed or read okay */
 312 backing_page_already_present:
 313         _debug("- present");
 314
 315         if (newpage) {
 316                 put_page(newpage);
 317                 newpage = NULL;
 318         }
 319
 320         if (PageError(backpage))
 321                 goto io_error;
 322
 323         if (PageUptodate(backpage))
 324                 goto backing_page_already_uptodate;
 325
 326         if (!trylock_page(backpage))
 327                 goto monitor_backing_page;
 328         _debug("read %p {%lx}", backpage, backpage->flags);
 329         goto read_backing_page;
 330
 331         /* the backing page is already up to date, attach the netfs
 332          * page to the pagecache and LRU and copy the data across */
 333 backing_page_already_uptodate:
 334         _debug("- uptodate");
 335
 336         fscache_mark_page_cached(op, netpage);
 337
 338         copy_highpage(netpage, backpage);
 339         fscache_end_io(op, netpage, 0);
 340         fscache_retrieval_complete(op, 1);
 341
 342 success:
 343         _debug("success");
 344         ret = 0;
 345
 346 out:
 347         if (backpage)
 348                 put_page(backpage);
 349         if (monitor) {
 350                 fscache_put_retrieval(monitor->op);
 351                 kfree(monitor);
 352         }
 353         _leave(" = %d", ret);
 354         return ret;
 355
 356 read_error:
 357         _debug("read error %d", ret);
 358         if (ret == -ENOMEM) {
 359                 fscache_retrieval_complete(op, 1);
 360                 goto out;
 361         }
 362 io_error:
 363         cachefiles_io_error_obj(object, "Page read error on backing file");
 364         fscache_retrieval_complete(op, 1);
 365         ret = -ENOBUFS;
 366         goto out;
 367
 368 nomem_page:
 369         put_page(newpage);
 370 nomem_monitor:
 371         fscache_put_retrieval(monitor->op);
 372         kfree(monitor);
 373 nomem:
 374         fscache_retrieval_complete(op, 1);
 375         _leave(" = -ENOMEM");
 376         return -ENOMEM;
 377 }
 378
 379 /*
 380  * read a page from the cache or allocate a block in which to store it
 381  * - cache withdrawal is prevented by the caller
 382  * - returns -EINTR if interrupted
 383  * - returns -ENOMEM if ran out of memory
 384  * - returns -ENOBUFS if no buffers can be made available
 385  * - returns -ENOBUFS if page is beyond EOF
 386  * - if the page is backed by a block in the cache:
 387  *   - a read will be started which will call the callback on completion
 388  *   - 0 will be returned
 389  * - else if the page is unbacked:
 390  *   - the metadata will be retained
 391  *   - -ENODATA will be returned
 392  */
 393 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 394                                   struct page *page,
 395                                   gfp_t gfp)
 396 {
 397         struct cachefiles_object *object;
 398         struct cachefiles_cache *cache;
 399         struct inode *inode;
 400         sector_t block;
 401         unsigned shift;
 402         int ret, ret2;
 403
 404         object = container_of(op->op.object,
 405                               struct cachefiles_object, fscache);
 406         cache = container_of(object->fscache.cache,
 407                              struct cachefiles_cache, cache);
 408
 409         _enter("{%p},{%lx},,,", object, page->index);
 410
 411         if (!object->backer)
 412                 goto enobufs;
 413
 414         inode = d_backing_inode(object->backer);
 415         ASSERT(S_ISREG(inode->i_mode));
 416         ASSERT(inode->i_mapping->a_ops->readpages);
 417
 418         /* calculate the shift required to use bmap */
 419         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 420
 421         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 422         op->op.flags |= FSCACHE_OP_ASYNC;
 423         op->op.processor = cachefiles_read_copier;
 424
 425         /* we assume the absence or presence of the first block is a good
 426          * enough indication for the page as a whole
 427          * - TODO: don't use bmap() for this as it is _not_ actually good
 428          *   enough for this as it doesn't indicate errors, but it's all we've
 429          *   got for the moment
 430          */
 431         block = page->index;
 432         block <<= shift;
 433
 434         ret2 = bmap(inode, &block);
 435         ASSERT(ret2 == 0);
 436
 437         _debug("%llx -> %llx",
 438                (unsigned long long) (page->index << shift),
 439                (unsigned long long) block);
 440
 441         if (block) {
 442                 /* submit the apparently valid page to the backing fs to be
 443                  * read from disk */
 444                 ret = cachefiles_read_backing_file_one(object, op, page);
 445         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 446                 /* there's space in the cache we can use */
 447                 fscache_mark_page_cached(op, page);
 448                 fscache_retrieval_complete(op, 1);
 449                 ret = -ENODATA;
 450         } else {
 451                 goto enobufs;
 452         }
 453
 454         _leave(" = %d", ret);
 455         return ret;
 456
 457 enobufs:
 458         fscache_retrieval_complete(op, 1);
 459         _leave(" = -ENOBUFS");
 460         return -ENOBUFS;
 461 }
 462
 463 /*
 464  * read the corresponding pages to the given set from the backing file
 465  * - any uncertain pages are simply discarded, to be tried again another time
 466  */
 467 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 468                                         struct fscache_retrieval *op,
 469                                         struct list_head *list)
 470 {
 471         struct cachefiles_one_read *monitor = NULL;
 472         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 473         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 474         int ret = 0;
 475
 476         _enter("");
 477
 478         list_for_each_entry_safe(netpage, _n, list, lru) {
 479                 list_del(&netpage->lru);
 480
 481                 _debug("read back %p{%lu,%d}",
 482                        netpage, netpage->index, page_count(netpage));
 483
 484                 if (!monitor) {
 485                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 486                         if (!monitor)
 487                                 goto nomem;
 488
 489                         monitor->op = fscache_get_retrieval(op);
 490                         init_waitqueue_func_entry(&monitor->monitor,
 491                                                   cachefiles_read_waiter);
 492                 }
 493
 494                 for (;;) {
 495                         backpage = find_get_page(bmapping, netpage->index);
 496                         if (backpage)
 497                                 goto backing_page_already_present;
 498
 499                         if (!newpage) {
 500                                 newpage = __page_cache_alloc(cachefiles_gfp);
 501                                 if (!newpage)
 502                                         goto nomem;
 503                         }
 504
 505                         ret = add_to_page_cache_lru(newpage, bmapping,
 506                                                     netpage->index,
 507                                                     cachefiles_gfp);
 508                         if (ret == 0)
 509                                 goto installed_new_backing_page;
 510                         if (ret != -EEXIST)
 511                                 goto nomem;
 512                 }
 513
 514                 /* we've installed a new backing page, so now we need
 515                  * to start it reading */
 516         installed_new_backing_page:
 517                 _debug("- new %p", newpage);
 518
 519                 backpage = newpage;
 520                 newpage = NULL;
 521
 522         reread_backing_page:
 523                 ret = bmapping->a_ops->readpage(NULL, backpage);
 524                 if (ret < 0)
 525                         goto read_error;
 526
 527                 /* add the netfs page to the pagecache and LRU, and set the
 528                  * monitor to transfer the data across */
 529         monitor_backing_page:
 530                 _debug("- monitor add");
 531
 532                 ret = add_to_page_cache_lru(netpage, op->mapping,
 533                                             netpage->index, cachefiles_gfp);
 534                 if (ret < 0) {
 535                         if (ret == -EEXIST) {
 536                                 put_page(backpage);
 537                                 backpage = NULL;
 538                                 put_page(netpage);
 539                                 netpage = NULL;
 540                                 fscache_retrieval_complete(op, 1);
 541                                 continue;
 542                         }
 543                         goto nomem;
 544                 }
 545
 546                 /* install a monitor */
 547                 get_page(netpage);
 548                 monitor->netfs_page = netpage;
 549
 550                 get_page(backpage);
 551                 monitor->back_page = backpage;
 552                 monitor->monitor.private = backpage;
 553                 add_page_wait_queue(backpage, &monitor->monitor);
 554                 monitor = NULL;
 555
 556                 /* but the page may have been read before the monitor was
 557                  * installed, so the monitor may miss the event - so we have to
 558                  * ensure that we do get one in such a case */
 559                 if (trylock_page(backpage)) {
 560                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 561                         unlock_page(backpage);
 562                 }
 563
 564                 put_page(backpage);
 565                 backpage = NULL;
 566
 567                 put_page(netpage);
 568                 netpage = NULL;
 569                 continue;
 570
 571                 /* if the backing page is already present, it can be in one of
 572                  * three states: read in progress, read failed or read okay */
 573         backing_page_already_present:
 574                 _debug("- present %p", backpage);
 575
 576                 if (PageError(backpage))
 577                         goto io_error;
 578
 579                 if (PageUptodate(backpage))
 580                         goto backing_page_already_uptodate;
 581
 582                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 583
 584                 if (!trylock_page(backpage))
 585                         goto monitor_backing_page;
 586
 587                 if (PageError(backpage)) {
 588                         _debug("error %lx", backpage->flags);
 589                         unlock_page(backpage);
 590                         goto io_error;
 591                 }
 592
 593                 if (PageUptodate(backpage))
 594                         goto backing_page_already_uptodate_unlock;
 595
 596                 /* we've locked a page that's neither up to date nor erroneous,
 597                  * so we need to attempt to read it again */
 598                 goto reread_backing_page;
 599
 600                 /* the backing page is already up to date, attach the netfs
 601                  * page to the pagecache and LRU and copy the data across */
 602         backing_page_already_uptodate_unlock:
 603                 _debug("uptodate %lx", backpage->flags);
 604                 unlock_page(backpage);
 605         backing_page_already_uptodate:
 606                 _debug("- uptodate");
 607
 608                 ret = add_to_page_cache_lru(netpage, op->mapping,
 609                                             netpage->index, cachefiles_gfp);
 610                 if (ret < 0) {
 611                         if (ret == -EEXIST) {
 612                                 put_page(backpage);
 613                                 backpage = NULL;
 614                                 put_page(netpage);
 615                                 netpage = NULL;
 616                                 fscache_retrieval_complete(op, 1);
 617                                 continue;
 618                         }
 619                         goto nomem;
 620                 }
 621
 622                 copy_highpage(netpage, backpage);
 623
 624                 put_page(backpage);
 625                 backpage = NULL;
 626
 627                 fscache_mark_page_cached(op, netpage);
 628
 629                 /* the netpage is unlocked and marked up to date here */
 630                 fscache_end_io(op, netpage, 0);
 631                 put_page(netpage);
 632                 netpage = NULL;
 633                 fscache_retrieval_complete(op, 1);
 634                 continue;
 635         }
 636
 637         netpage = NULL;
 638
 639         _debug("out");
 640
 641 out:
 642         /* tidy up */
 643         if (newpage)
 644                 put_page(newpage);
 645         if (netpage)
 646                 put_page(netpage);
 647         if (backpage)
 648                 put_page(backpage);
 649         if (monitor) {
 650                 fscache_put_retrieval(op);
 651                 kfree(monitor);
 652         }
 653
 654         list_for_each_entry_safe(netpage, _n, list, lru) {
 655                 list_del(&netpage->lru);
 656                 put_page(netpage);
 657                 fscache_retrieval_complete(op, 1);
 658         }
 659
 660         _leave(" = %d", ret);
 661         return ret;
 662
 663 nomem:
 664         _debug("nomem");
 665         ret = -ENOMEM;
 666         goto record_page_complete;
 667
 668 read_error:
 669         _debug("read error %d", ret);
 670         if (ret == -ENOMEM)
 671                 goto record_page_complete;
 672 io_error:
 673         cachefiles_io_error_obj(object, "Page read error on backing file");
 674         ret = -ENOBUFS;
 675 record_page_complete:
 676         fscache_retrieval_complete(op, 1);
 677         goto out;
 678 }
 679
 680 /*
 681  * read a list of pages from the cache or allocate blocks in which to store
 682  * them
 683  */
 684 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 685                                    struct list_head *pages,
 686                                    unsigned *nr_pages,
 687                                    gfp_t gfp)
 688 {
 689         struct cachefiles_object *object;
 690         struct cachefiles_cache *cache;
 691         struct list_head backpages;
 692         struct pagevec pagevec;
 693         struct inode *inode;
 694         struct page *page, *_n;
 695         unsigned shift, nrbackpages;
 696         int ret, ret2, space;
 697
 698         object = container_of(op->op.object,
 699                               struct cachefiles_object, fscache);
 700         cache = container_of(object->fscache.cache,
 701                              struct cachefiles_cache, cache);
 702
 703         _enter("{OBJ%x,%d},,%d,,",
 704                object->fscache.debug_id, atomic_read(&op->op.usage),
 705                *nr_pages);
 706
 707         if (!object->backer)
 708                 goto all_enobufs;
 709
 710         space = 1;
 711         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 712                 space = 0;
 713
 714         inode = d_backing_inode(object->backer);
 715         ASSERT(S_ISREG(inode->i_mode));
 716         ASSERT(inode->i_mapping->a_ops->readpages);
 717
 718         /* calculate the shift required to use bmap */
 719         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 720
 721         pagevec_init(&pagevec);
 722
 723         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 724         op->op.flags |= FSCACHE_OP_ASYNC;
 725         op->op.processor = cachefiles_read_copier;
 726
 727         INIT_LIST_HEAD(&backpages);
 728         nrbackpages = 0;
 729
 730         ret = space ? -ENODATA : -ENOBUFS;
 731         list_for_each_entry_safe(page, _n, pages, lru) {
 732                 sector_t block;
 733
 734                 /* we assume the absence or presence of the first block is a
 735                  * good enough indication for the page as a whole
 736                  * - TODO: don't use bmap() for this as it is _not_ actually
 737                  *   good enough for this as it doesn't indicate errors, but
 738                  *   it's all we've got for the moment
 739                  */
 740                 block = page->index;
 741                 block <<= shift;
 742
 743                 ret2 = bmap(inode, &block);
 744                 ASSERT(ret2 == 0);
 745
 746                 _debug("%llx -> %llx",
 747                        (unsigned long long) (page->index << shift),
 748                        (unsigned long long) block);
 749
 750                 if (block) {
 751                         /* we have data - add it to the list to give to the
 752                          * backing fs */
 753                         list_move(&page->lru, &backpages);
 754                         (*nr_pages)--;
 755                         nrbackpages++;
 756                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 757                         fscache_mark_pages_cached(op, &pagevec);
 758                         fscache_retrieval_complete(op, 1);
 759                         ret = -ENODATA;
 760                 } else {
 761                         fscache_retrieval_complete(op, 1);
 762                 }
 763         }
 764
 765         if (pagevec_count(&pagevec) > 0)
 766                 fscache_mark_pages_cached(op, &pagevec);
 767
 768         if (list_empty(pages))
 769                 ret = 0;
 770
 771         /* submit the apparently valid pages to the backing fs to be read from
 772          * disk */
 773         if (nrbackpages > 0) {
 774                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 775                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 776                         ret = ret2;
 777         }
 778
 779         _leave(" = %d [nr=%u%s]",
 780                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 781         return ret;
 782
 783 all_enobufs:
 784         fscache_retrieval_complete(op, *nr_pages);
 785         return -ENOBUFS;
 786 }
 787
 788 /*
 789  * allocate a block in the cache in which to store a page
 790  * - cache withdrawal is prevented by the caller
 791  * - returns -EINTR if interrupted
 792  * - returns -ENOMEM if ran out of memory
 793  * - returns -ENOBUFS if no buffers can be made available
 794  * - returns -ENOBUFS if page is beyond EOF
 795  * - otherwise:
 796  *   - the metadata will be retained
 797  *   - 0 will be returned
 798  */
 799 int cachefiles_allocate_page(struct fscache_retrieval *op,
 800                              struct page *page,
 801                              gfp_t gfp)
 802 {
 803         struct cachefiles_object *object;
 804         struct cachefiles_cache *cache;
 805         int ret;
 806
 807         object = container_of(op->op.object,
 808                               struct cachefiles_object, fscache);
 809         cache = container_of(object->fscache.cache,
 810                              struct cachefiles_cache, cache);
 811
 812         _enter("%p,{%lx},", object, page->index);
 813
 814         ret = cachefiles_has_space(cache, 0, 1);
 815         if (ret == 0)
 816                 fscache_mark_page_cached(op, page);
 817         else
 818                 ret = -ENOBUFS;
 819
 820         fscache_retrieval_complete(op, 1);
 821         _leave(" = %d", ret);
 822         return ret;
 823 }
 824
 825 /*
 826  * allocate blocks in the cache in which to store a set of pages
 827  * - cache withdrawal is prevented by the caller
 828  * - returns -EINTR if interrupted
 829  * - returns -ENOMEM if ran out of memory
 830  * - returns -ENOBUFS if some buffers couldn't be made available
 831  * - returns -ENOBUFS if some pages are beyond EOF
 832  * - otherwise:
 833  *   - -ENODATA will be returned
 834  * - metadata will be retained for any page marked
 835  */
 836 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 837                               struct list_head *pages,
 838                               unsigned *nr_pages,
 839                               gfp_t gfp)
 840 {
 841         struct cachefiles_object *object;
 842         struct cachefiles_cache *cache;
 843         struct pagevec pagevec;
 844         struct page *page;
 845         int ret;
 846
 847         object = container_of(op->op.object,
 848                               struct cachefiles_object, fscache);
 849         cache = container_of(object->fscache.cache,
 850                              struct cachefiles_cache, cache);
 851
 852         _enter("%p,,,%d,", object, *nr_pages);
 853
 854         ret = cachefiles_has_space(cache, 0, *nr_pages);
 855         if (ret == 0) {
 856                 pagevec_init(&pagevec);
 857
 858                 list_for_each_entry(page, pages, lru) {
 859                         if (pagevec_add(&pagevec, page) == 0)
 860                                 fscache_mark_pages_cached(op, &pagevec);
 861                 }
 862
 863                 if (pagevec_count(&pagevec) > 0)
 864                         fscache_mark_pages_cached(op, &pagevec);
 865                 ret = -ENODATA;
 866         } else {
 867                 ret = -ENOBUFS;
 868         }
 869
 870         fscache_retrieval_complete(op, *nr_pages);
 871         _leave(" = %d", ret);
 872         return ret;
 873 }
 874
 875 /*
 876  * request a page be stored in the cache
 877  * - cache withdrawal is prevented by the caller
 878  * - this request may be ignored if there's no cache block available, in which
 879  *   case -ENOBUFS will be returned
 880  * - if the op is in progress, 0 will be returned
 881  */
 882 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 883 {
 884         struct cachefiles_object *object;
 885         struct cachefiles_cache *cache;
 886         struct file *file;
 887         struct path path;
 888         loff_t pos, eof;
 889         size_t len;
 890         void *data;
 891         int ret = -ENOBUFS;
 892
 893         ASSERT(op != NULL);
 894         ASSERT(page != NULL);
 895
 896         object = container_of(op->op.object,
 897                               struct cachefiles_object, fscache);
 898
 899         _enter("%p,%p{%lx},,,", object, page, page->index);
 900
 901         if (!object->backer) {
 902                 _leave(" = -ENOBUFS");
 903                 return -ENOBUFS;
 904         }
 905
 906         ASSERT(d_is_reg(object->backer));
 907
 908         cache = container_of(object->fscache.cache,
 909                              struct cachefiles_cache, cache);
 910
 911         pos = (loff_t)page->index << PAGE_SHIFT;
 912
 913         /* We mustn't write more data than we have, so we have to beware of a
 914          * partial page at EOF.
 915          */
 916         eof = object->fscache.store_limit_l;
 917         if (pos >= eof)
 918                 goto error;
 919
 920         /* write the page to the backing filesystem and let it store it in its
 921          * own time */
 922         path.mnt = cache->mnt;
 923         path.dentry = object->backer;
 924         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 925         if (IS_ERR(file)) {
 926                 ret = PTR_ERR(file);
 927                 goto error_2;
 928         }
 929
 930         len = PAGE_SIZE;
 931         if (eof & ~PAGE_MASK) {
 932                 if (eof - pos < PAGE_SIZE) {
 933                         _debug("cut short %llx to %llx",
 934                                pos, eof);
 935                         len = eof - pos;
 936                         ASSERTCMP(pos + len, ==, eof);
 937                 }
 938         }
 939
 940         data = kmap(page);
 941         ret = kernel_write(file, data, len, &pos);
 942         kunmap(page);
 943         fput(file);
 944         if (ret != len)
 945                 goto error_eio;
 946
 947         _leave(" = 0");
 948         return 0;
 949
 950 error_eio:
 951         ret = -EIO;
 952 error_2:
 953         if (ret == -EIO)
 954                 cachefiles_io_error_obj(object,
 955                                         "Write page to backing file failed");
 956 error:
 957         _leave(" = -ENOBUFS [%d]", ret);
 958         return -ENOBUFS;
 959 }
 960
 961 /*
 962  * detach a backing block from a page
 963  * - cache withdrawal is prevented by the caller
 964  */
 965 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 966         __releases(&object->fscache.cookie->lock)
 967 {
 968         struct cachefiles_object *object;
 969
 970         object = container_of(_object, struct cachefiles_object, fscache);
 971
 972         _enter("%p,{%lu}", object, page->index);
 973
 974         spin_unlock(&object->fscache.cookie->lock);
 975 }