fs/cachefiles/rdwr.c

   1 /* Storage object read/write
   2  *
   3  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   4  * Written by David Howells (dhowells@redhat.com)
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public Licence
   8  * as published by the Free Software Foundation; either version
   9  * 2 of the Licence, or (at your option) any later version.
  10  */
  11
  12 #include <linux/mount.h>
  13 #include <linux/slab.h>
  14 #include <linux/file.h>
  15 #include <linux/swap.h>
  16 #include "internal.h"
  17
  18 /*
  19  * detect wake up events generated by the unlocking of pages in which we're
  20  * interested
  21  * - we use this to detect read completion of backing pages
  22  * - the caller holds the waitqueue lock
  23  */
  24 static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode,
  25                                   int sync, void *_key)
  26 {
  27         struct cachefiles_one_read *monitor =
  28                 container_of(wait, struct cachefiles_one_read, monitor);
  29         struct cachefiles_object *object;
  30         struct wait_bit_key *key = _key;
  31         struct page *page = wait->private;
  32
  33         ASSERT(key);
  34
  35         _enter("{%lu},%u,%d,{%p,%u}",
  36                monitor->netfs_page->index, mode, sync,
  37                key->flags, key->bit_nr);
  38
  39         if (key->flags != &page->flags ||
  40             key->bit_nr != PG_locked)
  41                 return 0;
  42
  43         _debug("--- monitor %p %lx ---", page, page->flags);
  44
  45         if (!PageUptodate(page) && !PageError(page)) {
  46                 /* unlocked, not uptodate and not erronous? */
  47                 _debug("page probably truncated");
  48         }
  49
  50         /* remove from the waitqueue */
  51         list_del(&wait->task_list);
  52
  53         /* move onto the action list and queue for FS-Cache thread pool */
  54         ASSERT(monitor->op);
  55
  56         object = container_of(monitor->op->op.object,
  57                               struct cachefiles_object, fscache);
  58
  59         spin_lock(&object->work_lock);
  60         list_add_tail(&monitor->op_link, &monitor->op->to_do);
  61         spin_unlock(&object->work_lock);
  62
  63         fscache_enqueue_retrieval(monitor->op);
  64         return 0;
  65 }
  66
  67 /*
  68  * handle a probably truncated page
  69  * - check to see if the page is still relevant and reissue the read if
  70  *   possible
  71  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  72  *   must wait again and 0 if successful
  73  */
  74 static int cachefiles_read_reissue(struct cachefiles_object *object,
  75                                    struct cachefiles_one_read *monitor)
  76 {
  77         struct address_space *bmapping = object->backer->d_inode->i_mapping;
  78         struct page *backpage = monitor->back_page, *backpage2;
  79         int ret;
  80
  81         _enter("{ino=%lx},{%lx,%lx}",
  82                object->backer->d_inode->i_ino,
  83                backpage->index, backpage->flags);
  84
  85         /* skip if the page was truncated away completely */
  86         if (backpage->mapping != bmapping) {
  87                 _leave(" = -ENODATA [mapping]");
  88                 return -ENODATA;
  89         }
  90
  91         backpage2 = find_get_page(bmapping, backpage->index);
  92         if (!backpage2) {
  93                 _leave(" = -ENODATA [gone]");
  94                 return -ENODATA;
  95         }
  96
  97         if (backpage != backpage2) {
  98                 put_page(backpage2);
  99                 _leave(" = -ENODATA [different]");
 100                 return -ENODATA;
 101         }
 102
 103         /* the page is still there and we already have a ref on it, so we don't
 104          * need a second */
 105         put_page(backpage2);
 106
 107         INIT_LIST_HEAD(&monitor->op_link);
 108         add_page_wait_queue(backpage, &monitor->monitor);
 109
 110         if (trylock_page(backpage)) {
 111                 ret = -EIO;
 112                 if (PageError(backpage))
 113                         goto unlock_discard;
 114                 ret = 0;
 115                 if (PageUptodate(backpage))
 116                         goto unlock_discard;
 117
 118                 _debug("reissue read");
 119                 ret = bmapping->a_ops->readpage(NULL, backpage);
 120                 if (ret < 0)
 121                         goto unlock_discard;
 122         }
 123
 124         /* but the page may have been read before the monitor was installed, so
 125          * the monitor may miss the event - so we have to ensure that we do get
 126          * one in such a case */
 127         if (trylock_page(backpage)) {
 128                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 129                 unlock_page(backpage);
 130         }
 131
 132         /* it'll reappear on the todo list */
 133         _leave(" = -EINPROGRESS");
 134         return -EINPROGRESS;
 135
 136 unlock_discard:
 137         unlock_page(backpage);
 138         spin_lock_irq(&object->work_lock);
 139         list_del(&monitor->op_link);
 140         spin_unlock_irq(&object->work_lock);
 141         _leave(" = %d", ret);
 142         return ret;
 143 }
 144
 145 /*
 146  * copy data from backing pages to netfs pages to complete a read operation
 147  * - driven by FS-Cache's thread pool
 148  */
 149 static void cachefiles_read_copier(struct fscache_operation *_op)
 150 {
 151         struct cachefiles_one_read *monitor;
 152         struct cachefiles_object *object;
 153         struct fscache_retrieval *op;
 154         struct pagevec pagevec;
 155         int error, max;
 156
 157         op = container_of(_op, struct fscache_retrieval, op);
 158         object = container_of(op->op.object,
 159                               struct cachefiles_object, fscache);
 160
 161         _enter("{ino=%lu}", object->backer->d_inode->i_ino);
 162
 163         pagevec_init(&pagevec, 0);
 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                 page_cache_release(monitor->back_page);
 201
 202                 fscache_end_io(op, monitor->netfs_page, error);
 203                 page_cache_release(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  */
 229 static 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 = object->backer->d_inode->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                                                      __GFP_COLD);
 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         page_cache_get(monitor->netfs_page);
 295         page_cache_get(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                 page_cache_release(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                 page_cache_release(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         page_cache_release(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 pagevec pagevec;
 400         struct inode *inode;
 401         sector_t block0, block;
 402         unsigned shift;
 403         int ret;
 404
 405         object = container_of(op->op.object,
 406                               struct cachefiles_object, fscache);
 407         cache = container_of(object->fscache.cache,
 408                              struct cachefiles_cache, cache);
 409
 410         _enter("{%p},{%lx},,,", object, page->index);
 411
 412         if (!object->backer)
 413                 goto enobufs;
 414
 415         inode = object->backer->d_inode;
 416         ASSERT(S_ISREG(inode->i_mode));
 417         ASSERT(inode->i_mapping->a_ops->bmap);
 418         ASSERT(inode->i_mapping->a_ops->readpages);
 419
 420         /* calculate the shift required to use bmap */
 421         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 422                 goto enobufs;
 423
 424         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 425
 426         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 427         op->op.flags |= FSCACHE_OP_ASYNC;
 428         op->op.processor = cachefiles_read_copier;
 429
 430         pagevec_init(&pagevec, 0);
 431
 432         /* we assume the absence or presence of the first block is a good
 433          * enough indication for the page as a whole
 434          * - TODO: don't use bmap() for this as it is _not_ actually good
 435          *   enough for this as it doesn't indicate errors, but it's all we've
 436          *   got for the moment
 437          */
 438         block0 = page->index;
 439         block0 <<= shift;
 440
 441         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 442         _debug("%llx -> %llx",
 443                (unsigned long long) block0,
 444                (unsigned long long) block);
 445
 446         if (block) {
 447                 /* submit the apparently valid page to the backing fs to be
 448                  * read from disk */
 449                 ret = cachefiles_read_backing_file_one(object, op, page);
 450         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 451                 /* there's space in the cache we can use */
 452                 fscache_mark_page_cached(op, page);
 453                 fscache_retrieval_complete(op, 1);
 454                 ret = -ENODATA;
 455         } else {
 456                 goto enobufs;
 457         }
 458
 459         _leave(" = %d", ret);
 460         return ret;
 461
 462 enobufs:
 463         fscache_retrieval_complete(op, 1);
 464         _leave(" = -ENOBUFS");
 465         return -ENOBUFS;
 466 }
 467
 468 /*
 469  * read the corresponding pages to the given set from the backing file
 470  * - any uncertain pages are simply discarded, to be tried again another time
 471  */
 472 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 473                                         struct fscache_retrieval *op,
 474                                         struct list_head *list)
 475 {
 476         struct cachefiles_one_read *monitor = NULL;
 477         struct address_space *bmapping = object->backer->d_inode->i_mapping;
 478         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 479         int ret = 0;
 480
 481         _enter("");
 482
 483         list_for_each_entry_safe(netpage, _n, list, lru) {
 484                 list_del(&netpage->lru);
 485
 486                 _debug("read back %p{%lu,%d}",
 487                        netpage, netpage->index, page_count(netpage));
 488
 489                 if (!monitor) {
 490                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 491                         if (!monitor)
 492                                 goto nomem;
 493
 494                         monitor->op = fscache_get_retrieval(op);
 495                         init_waitqueue_func_entry(&monitor->monitor,
 496                                                   cachefiles_read_waiter);
 497                 }
 498
 499                 for (;;) {
 500                         backpage = find_get_page(bmapping, netpage->index);
 501                         if (backpage)
 502                                 goto backing_page_already_present;
 503
 504                         if (!newpage) {
 505                                 newpage = __page_cache_alloc(cachefiles_gfp |
 506                                                              __GFP_COLD);
 507                                 if (!newpage)
 508                                         goto nomem;
 509                         }
 510
 511                         ret = add_to_page_cache_lru(newpage, bmapping,
 512                                                     netpage->index,
 513                                                     cachefiles_gfp);
 514                         if (ret == 0)
 515                                 goto installed_new_backing_page;
 516                         if (ret != -EEXIST)
 517                                 goto nomem;
 518                 }
 519
 520                 /* we've installed a new backing page, so now we need
 521                  * to start it reading */
 522         installed_new_backing_page:
 523                 _debug("- new %p", newpage);
 524
 525                 backpage = newpage;
 526                 newpage = NULL;
 527
 528         reread_backing_page:
 529                 ret = bmapping->a_ops->readpage(NULL, backpage);
 530                 if (ret < 0)
 531                         goto read_error;
 532
 533                 /* add the netfs page to the pagecache and LRU, and set the
 534                  * monitor to transfer the data across */
 535         monitor_backing_page:
 536                 _debug("- monitor add");
 537
 538                 ret = add_to_page_cache_lru(netpage, op->mapping,
 539                                             netpage->index, cachefiles_gfp);
 540                 if (ret < 0) {
 541                         if (ret == -EEXIST) {
 542                                 page_cache_release(netpage);
 543                                 fscache_retrieval_complete(op, 1);
 544                                 continue;
 545                         }
 546                         goto nomem;
 547                 }
 548
 549                 /* install a monitor */
 550                 page_cache_get(netpage);
 551                 monitor->netfs_page = netpage;
 552
 553                 page_cache_get(backpage);
 554                 monitor->back_page = backpage;
 555                 monitor->monitor.private = backpage;
 556                 add_page_wait_queue(backpage, &monitor->monitor);
 557                 monitor = NULL;
 558
 559                 /* but the page may have been read before the monitor was
 560                  * installed, so the monitor may miss the event - so we have to
 561                  * ensure that we do get one in such a case */
 562                 if (trylock_page(backpage)) {
 563                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 564                         unlock_page(backpage);
 565                 }
 566
 567                 page_cache_release(backpage);
 568                 backpage = NULL;
 569
 570                 page_cache_release(netpage);
 571                 netpage = NULL;
 572                 continue;
 573
 574                 /* if the backing page is already present, it can be in one of
 575                  * three states: read in progress, read failed or read okay */
 576         backing_page_already_present:
 577                 _debug("- present %p", backpage);
 578
 579                 if (PageError(backpage))
 580                         goto io_error;
 581
 582                 if (PageUptodate(backpage))
 583                         goto backing_page_already_uptodate;
 584
 585                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 586
 587                 if (!trylock_page(backpage))
 588                         goto monitor_backing_page;
 589
 590                 if (PageError(backpage)) {
 591                         _debug("error %lx", backpage->flags);
 592                         unlock_page(backpage);
 593                         goto io_error;
 594                 }
 595
 596                 if (PageUptodate(backpage))
 597                         goto backing_page_already_uptodate_unlock;
 598
 599                 /* we've locked a page that's neither up to date nor erroneous,
 600                  * so we need to attempt to read it again */
 601                 goto reread_backing_page;
 602
 603                 /* the backing page is already up to date, attach the netfs
 604                  * page to the pagecache and LRU and copy the data across */
 605         backing_page_already_uptodate_unlock:
 606                 _debug("uptodate %lx", backpage->flags);
 607                 unlock_page(backpage);
 608         backing_page_already_uptodate:
 609                 _debug("- uptodate");
 610
 611                 ret = add_to_page_cache_lru(netpage, op->mapping,
 612                                             netpage->index, cachefiles_gfp);
 613                 if (ret < 0) {
 614                         if (ret == -EEXIST) {
 615                                 page_cache_release(netpage);
 616                                 fscache_retrieval_complete(op, 1);
 617                                 continue;
 618                         }
 619                         goto nomem;
 620                 }
 621
 622                 copy_highpage(netpage, backpage);
 623
 624                 page_cache_release(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                 page_cache_release(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                 page_cache_release(newpage);
 645         if (netpage)
 646                 page_cache_release(netpage);
 647         if (backpage)
 648                 page_cache_release(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                 page_cache_release(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 = object->backer->d_inode;
 715         ASSERT(S_ISREG(inode->i_mode));
 716         ASSERT(inode->i_mapping->a_ops->bmap);
 717         ASSERT(inode->i_mapping->a_ops->readpages);
 718
 719         /* calculate the shift required to use bmap */
 720         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 721                 goto all_enobufs;
 722
 723         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 724
 725         pagevec_init(&pagevec, 0);
 726
 727         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 728         op->op.flags |= FSCACHE_OP_ASYNC;
 729         op->op.processor = cachefiles_read_copier;
 730
 731         INIT_LIST_HEAD(&backpages);
 732         nrbackpages = 0;
 733
 734         ret = space ? -ENODATA : -ENOBUFS;
 735         list_for_each_entry_safe(page, _n, pages, lru) {
 736                 sector_t block0, block;
 737
 738                 /* we assume the absence or presence of the first block is a
 739                  * good enough indication for the page as a whole
 740                  * - TODO: don't use bmap() for this as it is _not_ actually
 741                  *   good enough for this as it doesn't indicate errors, but
 742                  *   it's all we've got for the moment
 743                  */
 744                 block0 = page->index;
 745                 block0 <<= shift;
 746
 747                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 748                                                       block0);
 749                 _debug("%llx -> %llx",
 750                        (unsigned long long) block0,
 751                        (unsigned long long) block);
 752
 753                 if (block) {
 754                         /* we have data - add it to the list to give to the
 755                          * backing fs */
 756                         list_move(&page->lru, &backpages);
 757                         (*nr_pages)--;
 758                         nrbackpages++;
 759                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 760                         fscache_mark_pages_cached(op, &pagevec);
 761                         fscache_retrieval_complete(op, 1);
 762                         ret = -ENODATA;
 763                 } else {
 764                         fscache_retrieval_complete(op, 1);
 765                 }
 766         }
 767
 768         if (pagevec_count(&pagevec) > 0)
 769                 fscache_mark_pages_cached(op, &pagevec);
 770
 771         if (list_empty(pages))
 772                 ret = 0;
 773
 774         /* submit the apparently valid pages to the backing fs to be read from
 775          * disk */
 776         if (nrbackpages > 0) {
 777                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 778                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 779                         ret = ret2;
 780         }
 781
 782         _leave(" = %d [nr=%u%s]",
 783                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 784         return ret;
 785
 786 all_enobufs:
 787         fscache_retrieval_complete(op, *nr_pages);
 788         return -ENOBUFS;
 789 }
 790
 791 /*
 792  * allocate a block in the cache in which to store a page
 793  * - cache withdrawal is prevented by the caller
 794  * - returns -EINTR if interrupted
 795  * - returns -ENOMEM if ran out of memory
 796  * - returns -ENOBUFS if no buffers can be made available
 797  * - returns -ENOBUFS if page is beyond EOF
 798  * - otherwise:
 799  *   - the metadata will be retained
 800  *   - 0 will be returned
 801  */
 802 int cachefiles_allocate_page(struct fscache_retrieval *op,
 803                              struct page *page,
 804                              gfp_t gfp)
 805 {
 806         struct cachefiles_object *object;
 807         struct cachefiles_cache *cache;
 808         int ret;
 809
 810         object = container_of(op->op.object,
 811                               struct cachefiles_object, fscache);
 812         cache = container_of(object->fscache.cache,
 813                              struct cachefiles_cache, cache);
 814
 815         _enter("%p,{%lx},", object, page->index);
 816
 817         ret = cachefiles_has_space(cache, 0, 1);
 818         if (ret == 0)
 819                 fscache_mark_page_cached(op, page);
 820         else
 821                 ret = -ENOBUFS;
 822
 823         fscache_retrieval_complete(op, 1);
 824         _leave(" = %d", ret);
 825         return ret;
 826 }
 827
 828 /*
 829  * allocate blocks in the cache in which to store a set of pages
 830  * - cache withdrawal is prevented by the caller
 831  * - returns -EINTR if interrupted
 832  * - returns -ENOMEM if ran out of memory
 833  * - returns -ENOBUFS if some buffers couldn't be made available
 834  * - returns -ENOBUFS if some pages are beyond EOF
 835  * - otherwise:
 836  *   - -ENODATA will be returned
 837  * - metadata will be retained for any page marked
 838  */
 839 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 840                               struct list_head *pages,
 841                               unsigned *nr_pages,
 842                               gfp_t gfp)
 843 {
 844         struct cachefiles_object *object;
 845         struct cachefiles_cache *cache;
 846         struct pagevec pagevec;
 847         struct page *page;
 848         int ret;
 849
 850         object = container_of(op->op.object,
 851                               struct cachefiles_object, fscache);
 852         cache = container_of(object->fscache.cache,
 853                              struct cachefiles_cache, cache);
 854
 855         _enter("%p,,,%d,", object, *nr_pages);
 856
 857         ret = cachefiles_has_space(cache, 0, *nr_pages);
 858         if (ret == 0) {
 859                 pagevec_init(&pagevec, 0);
 860
 861                 list_for_each_entry(page, pages, lru) {
 862                         if (pagevec_add(&pagevec, page) == 0)
 863                                 fscache_mark_pages_cached(op, &pagevec);
 864                 }
 865
 866                 if (pagevec_count(&pagevec) > 0)
 867                         fscache_mark_pages_cached(op, &pagevec);
 868                 ret = -ENODATA;
 869         } else {
 870                 ret = -ENOBUFS;
 871         }
 872
 873         fscache_retrieval_complete(op, *nr_pages);
 874         _leave(" = %d", ret);
 875         return ret;
 876 }
 877
 878 /*
 879  * request a page be stored in the cache
 880  * - cache withdrawal is prevented by the caller
 881  * - this request may be ignored if there's no cache block available, in which
 882  *   case -ENOBUFS will be returned
 883  * - if the op is in progress, 0 will be returned
 884  */
 885 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 886 {
 887         struct cachefiles_object *object;
 888         struct cachefiles_cache *cache;
 889         mm_segment_t old_fs;
 890         struct file *file;
 891         struct path path;
 892         loff_t pos, eof;
 893         size_t len;
 894         void *data;
 895         int ret;
 896
 897         ASSERT(op != NULL);
 898         ASSERT(page != NULL);
 899
 900         object = container_of(op->op.object,
 901                               struct cachefiles_object, fscache);
 902
 903         _enter("%p,%p{%lx},,,", object, page, page->index);
 904
 905         if (!object->backer) {
 906                 _leave(" = -ENOBUFS");
 907                 return -ENOBUFS;
 908         }
 909
 910         ASSERT(S_ISREG(object->backer->d_inode->i_mode));
 911
 912         cache = container_of(object->fscache.cache,
 913                              struct cachefiles_cache, cache);
 914
 915         /* write the page to the backing filesystem and let it store it in its
 916          * own time */
 917         path.mnt = cache->mnt;
 918         path.dentry = object->backer;
 919         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 920         if (IS_ERR(file)) {
 921                 ret = PTR_ERR(file);
 922         } else {
 923                 ret = -EIO;
 924                 if (file->f_op->write) {
 925                         pos = (loff_t) page->index << PAGE_SHIFT;
 926
 927                         /* we mustn't write more data than we have, so we have
 928                          * to beware of a partial page at EOF */
 929                         eof = object->fscache.store_limit_l;
 930                         len = PAGE_SIZE;
 931                         if (eof & ~PAGE_MASK) {
 932                                 ASSERTCMP(pos, <, eof);
 933                                 if (eof - pos < PAGE_SIZE) {
 934                                         _debug("cut short %llx to %llx",
 935                                                pos, eof);
 936                                         len = eof - pos;
 937                                         ASSERTCMP(pos + len, ==, eof);
 938                                 }
 939                         }
 940
 941                         data = kmap(page);
 942                         file_start_write(file);
 943                         old_fs = get_fs();
 944                         set_fs(KERNEL_DS);
 945                         ret = file->f_op->write(
 946                                 file, (const void __user *) data, len, &pos);
 947                         set_fs(old_fs);
 948                         kunmap(page);
 949                         file_end_write(file);
 950                         if (ret != len)
 951                                 ret = -EIO;
 952                 }
 953                 fput(file);
 954         }
 955
 956         if (ret < 0) {
 957                 if (ret == -EIO)
 958                         cachefiles_io_error_obj(
 959                                 object, "Write page to backing file failed");
 960                 ret = -ENOBUFS;
 961         }
 962
 963         _leave(" = %d", ret);
 964         return ret;
 965 }
 966
 967 /*
 968  * detach a backing block from a page
 969  * - cache withdrawal is prevented by the caller
 970  */
 971 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 972 {
 973         struct cachefiles_object *object;
 974         struct cachefiles_cache *cache;
 975
 976         object = container_of(_object, struct cachefiles_object, fscache);
 977         cache = container_of(object->fscache.cache,
 978                              struct cachefiles_cache, cache);
 979
 980         _enter("%p,{%lu}", object, page->index);
 981
 982         spin_unlock(&object->fscache.cookie->lock);
 983 }