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(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 add it
 277          * to the LRU list and start it reading */
 278 installed_new_backing_page:
 279         _debug("- new %p", newpage);
 280
 281         backpage = newpage;
 282         newpage = NULL;
 283
 284         lru_cache_add_file(backpage);
 285
 286 read_backing_page:
 287         ret = bmapping->a_ops->readpage(NULL, backpage);
 288         if (ret < 0)
 289                 goto read_error;
 290
 291         /* set the monitor to transfer the data across */
 292 monitor_backing_page:
 293         _debug("- monitor add");
 294
 295         /* install the monitor */
 296         page_cache_get(monitor->netfs_page);
 297         page_cache_get(backpage);
 298         monitor->back_page = backpage;
 299         monitor->monitor.private = backpage;
 300         add_page_wait_queue(backpage, &monitor->monitor);
 301         monitor = NULL;
 302
 303         /* but the page may have been read before the monitor was installed, so
 304          * the monitor may miss the event - so we have to ensure that we do get
 305          * one in such a case */
 306         if (trylock_page(backpage)) {
 307                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 308                 unlock_page(backpage);
 309         }
 310         goto success;
 311
 312         /* if the backing page is already present, it can be in one of
 313          * three states: read in progress, read failed or read okay */
 314 backing_page_already_present:
 315         _debug("- present");
 316
 317         if (newpage) {
 318                 page_cache_release(newpage);
 319                 newpage = NULL;
 320         }
 321
 322         if (PageError(backpage))
 323                 goto io_error;
 324
 325         if (PageUptodate(backpage))
 326                 goto backing_page_already_uptodate;
 327
 328         if (!trylock_page(backpage))
 329                 goto monitor_backing_page;
 330         _debug("read %p {%lx}", backpage, backpage->flags);
 331         goto read_backing_page;
 332
 333         /* the backing page is already up to date, attach the netfs
 334          * page to the pagecache and LRU and copy the data across */
 335 backing_page_already_uptodate:
 336         _debug("- uptodate");
 337
 338         fscache_mark_page_cached(op, netpage);
 339
 340         copy_highpage(netpage, backpage);
 341         fscache_end_io(op, netpage, 0);
 342         fscache_retrieval_complete(op, 1);
 343
 344 success:
 345         _debug("success");
 346         ret = 0;
 347
 348 out:
 349         if (backpage)
 350                 page_cache_release(backpage);
 351         if (monitor) {
 352                 fscache_put_retrieval(monitor->op);
 353                 kfree(monitor);
 354         }
 355         _leave(" = %d", ret);
 356         return ret;
 357
 358 read_error:
 359         _debug("read error %d", ret);
 360         if (ret == -ENOMEM) {
 361                 fscache_retrieval_complete(op, 1);
 362                 goto out;
 363         }
 364 io_error:
 365         cachefiles_io_error_obj(object, "Page read error on backing file");
 366         fscache_retrieval_complete(op, 1);
 367         ret = -ENOBUFS;
 368         goto out;
 369
 370 nomem_page:
 371         page_cache_release(newpage);
 372 nomem_monitor:
 373         fscache_put_retrieval(monitor->op);
 374         kfree(monitor);
 375 nomem:
 376         fscache_retrieval_complete(op, 1);
 377         _leave(" = -ENOMEM");
 378         return -ENOMEM;
 379 }
 380
 381 /*
 382  * read a page from the cache or allocate a block in which to store it
 383  * - cache withdrawal is prevented by the caller
 384  * - returns -EINTR if interrupted
 385  * - returns -ENOMEM if ran out of memory
 386  * - returns -ENOBUFS if no buffers can be made available
 387  * - returns -ENOBUFS if page is beyond EOF
 388  * - if the page is backed by a block in the cache:
 389  *   - a read will be started which will call the callback on completion
 390  *   - 0 will be returned
 391  * - else if the page is unbacked:
 392  *   - the metadata will be retained
 393  *   - -ENODATA will be returned
 394  */
 395 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 396                                   struct page *page,
 397                                   gfp_t gfp)
 398 {
 399         struct cachefiles_object *object;
 400         struct cachefiles_cache *cache;
 401         struct pagevec pagevec;
 402         struct inode *inode;
 403         sector_t block0, block;
 404         unsigned shift;
 405         int ret;
 406
 407         object = container_of(op->op.object,
 408                               struct cachefiles_object, fscache);
 409         cache = container_of(object->fscache.cache,
 410                              struct cachefiles_cache, cache);
 411
 412         _enter("{%p},{%lx},,,", object, page->index);
 413
 414         if (!object->backer)
 415                 goto enobufs;
 416
 417         inode = object->backer->d_inode;
 418         ASSERT(S_ISREG(inode->i_mode));
 419         ASSERT(inode->i_mapping->a_ops->bmap);
 420         ASSERT(inode->i_mapping->a_ops->readpages);
 421
 422         /* calculate the shift required to use bmap */
 423         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 424                 goto enobufs;
 425
 426         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 427
 428         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 429         op->op.flags |= FSCACHE_OP_ASYNC;
 430         op->op.processor = cachefiles_read_copier;
 431
 432         pagevec_init(&pagevec, 0);
 433
 434         /* we assume the absence or presence of the first block is a good
 435          * enough indication for the page as a whole
 436          * - TODO: don't use bmap() for this as it is _not_ actually good
 437          *   enough for this as it doesn't indicate errors, but it's all we've
 438          *   got for the moment
 439          */
 440         block0 = page->index;
 441         block0 <<= shift;
 442
 443         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 444         _debug("%llx -> %llx",
 445                (unsigned long long) block0,
 446                (unsigned long long) block);
 447
 448         if (block) {
 449                 /* submit the apparently valid page to the backing fs to be
 450                  * read from disk */
 451                 ret = cachefiles_read_backing_file_one(object, op, page);
 452         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 453                 /* there's space in the cache we can use */
 454                 fscache_mark_page_cached(op, page);
 455                 fscache_retrieval_complete(op, 1);
 456                 ret = -ENODATA;
 457         } else {
 458                 goto enobufs;
 459         }
 460
 461         _leave(" = %d", ret);
 462         return ret;
 463
 464 enobufs:
 465         fscache_retrieval_complete(op, 1);
 466         _leave(" = -ENOBUFS");
 467         return -ENOBUFS;
 468 }
 469
 470 /*
 471  * read the corresponding pages to the given set from the backing file
 472  * - any uncertain pages are simply discarded, to be tried again another time
 473  */
 474 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 475                                         struct fscache_retrieval *op,
 476                                         struct list_head *list)
 477 {
 478         struct cachefiles_one_read *monitor = NULL;
 479         struct address_space *bmapping = object->backer->d_inode->i_mapping;
 480         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 481         int ret = 0;
 482
 483         _enter("");
 484
 485         list_for_each_entry_safe(netpage, _n, list, lru) {
 486                 list_del(&netpage->lru);
 487
 488                 _debug("read back %p{%lu,%d}",
 489                        netpage, netpage->index, page_count(netpage));
 490
 491                 if (!monitor) {
 492                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 493                         if (!monitor)
 494                                 goto nomem;
 495
 496                         monitor->op = fscache_get_retrieval(op);
 497                         init_waitqueue_func_entry(&monitor->monitor,
 498                                                   cachefiles_read_waiter);
 499                 }
 500
 501                 for (;;) {
 502                         backpage = find_get_page(bmapping, netpage->index);
 503                         if (backpage)
 504                                 goto backing_page_already_present;
 505
 506                         if (!newpage) {
 507                                 newpage = __page_cache_alloc(cachefiles_gfp |
 508                                                              __GFP_COLD);
 509                                 if (!newpage)
 510                                         goto nomem;
 511                         }
 512
 513                         ret = add_to_page_cache(newpage, bmapping,
 514                                                 netpage->index, cachefiles_gfp);
 515                         if (ret == 0)
 516                                 goto installed_new_backing_page;
 517                         if (ret != -EEXIST)
 518                                 goto nomem;
 519                 }
 520
 521                 /* we've installed a new backing page, so now we need to add it
 522                  * to the LRU list and start it reading */
 523         installed_new_backing_page:
 524                 _debug("- new %p", newpage);
 525
 526                 backpage = newpage;
 527                 newpage = NULL;
 528
 529                 lru_cache_add_file(backpage);
 530
 531         reread_backing_page:
 532                 ret = bmapping->a_ops->readpage(NULL, backpage);
 533                 if (ret < 0)
 534                         goto read_error;
 535
 536                 /* add the netfs page to the pagecache and LRU, and set the
 537                  * monitor to transfer the data across */
 538         monitor_backing_page:
 539                 _debug("- monitor add");
 540
 541                 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
 542                                         cachefiles_gfp);
 543                 if (ret < 0) {
 544                         if (ret == -EEXIST) {
 545                                 page_cache_release(netpage);
 546                                 fscache_retrieval_complete(op, 1);
 547                                 continue;
 548                         }
 549                         goto nomem;
 550                 }
 551
 552                 lru_cache_add_file(netpage);
 553
 554                 /* install a monitor */
 555                 page_cache_get(netpage);
 556                 monitor->netfs_page = netpage;
 557
 558                 page_cache_get(backpage);
 559                 monitor->back_page = backpage;
 560                 monitor->monitor.private = backpage;
 561                 add_page_wait_queue(backpage, &monitor->monitor);
 562                 monitor = NULL;
 563
 564                 /* but the page may have been read before the monitor was
 565                  * installed, so the monitor may miss the event - so we have to
 566                  * ensure that we do get one in such a case */
 567                 if (trylock_page(backpage)) {
 568                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 569                         unlock_page(backpage);
 570                 }
 571
 572                 page_cache_release(backpage);
 573                 backpage = NULL;
 574
 575                 page_cache_release(netpage);
 576                 netpage = NULL;
 577                 continue;
 578
 579                 /* if the backing page is already present, it can be in one of
 580                  * three states: read in progress, read failed or read okay */
 581         backing_page_already_present:
 582                 _debug("- present %p", backpage);
 583
 584                 if (PageError(backpage))
 585                         goto io_error;
 586
 587                 if (PageUptodate(backpage))
 588                         goto backing_page_already_uptodate;
 589
 590                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 591
 592                 if (!trylock_page(backpage))
 593                         goto monitor_backing_page;
 594
 595                 if (PageError(backpage)) {
 596                         _debug("error %lx", backpage->flags);
 597                         unlock_page(backpage);
 598                         goto io_error;
 599                 }
 600
 601                 if (PageUptodate(backpage))
 602                         goto backing_page_already_uptodate_unlock;
 603
 604                 /* we've locked a page that's neither up to date nor erroneous,
 605                  * so we need to attempt to read it again */
 606                 goto reread_backing_page;
 607
 608                 /* the backing page is already up to date, attach the netfs
 609                  * page to the pagecache and LRU and copy the data across */
 610         backing_page_already_uptodate_unlock:
 611                 _debug("uptodate %lx", backpage->flags);
 612                 unlock_page(backpage);
 613         backing_page_already_uptodate:
 614                 _debug("- uptodate");
 615
 616                 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
 617                                         cachefiles_gfp);
 618                 if (ret < 0) {
 619                         if (ret == -EEXIST) {
 620                                 page_cache_release(netpage);
 621                                 fscache_retrieval_complete(op, 1);
 622                                 continue;
 623                         }
 624                         goto nomem;
 625                 }
 626
 627                 copy_highpage(netpage, backpage);
 628
 629                 page_cache_release(backpage);
 630                 backpage = NULL;
 631
 632                 fscache_mark_page_cached(op, netpage);
 633
 634                 lru_cache_add_file(netpage);
 635
 636                 /* the netpage is unlocked and marked up to date here */
 637                 fscache_end_io(op, netpage, 0);
 638                 page_cache_release(netpage);
 639                 netpage = NULL;
 640                 fscache_retrieval_complete(op, 1);
 641                 continue;
 642         }
 643
 644         netpage = NULL;
 645
 646         _debug("out");
 647
 648 out:
 649         /* tidy up */
 650         if (newpage)
 651                 page_cache_release(newpage);
 652         if (netpage)
 653                 page_cache_release(netpage);
 654         if (backpage)
 655                 page_cache_release(backpage);
 656         if (monitor) {
 657                 fscache_put_retrieval(op);
 658                 kfree(monitor);
 659         }
 660
 661         list_for_each_entry_safe(netpage, _n, list, lru) {
 662                 list_del(&netpage->lru);
 663                 page_cache_release(netpage);
 664                 fscache_retrieval_complete(op, 1);
 665         }
 666
 667         _leave(" = %d", ret);
 668         return ret;
 669
 670 nomem:
 671         _debug("nomem");
 672         ret = -ENOMEM;
 673         goto record_page_complete;
 674
 675 read_error:
 676         _debug("read error %d", ret);
 677         if (ret == -ENOMEM)
 678                 goto record_page_complete;
 679 io_error:
 680         cachefiles_io_error_obj(object, "Page read error on backing file");
 681         ret = -ENOBUFS;
 682 record_page_complete:
 683         fscache_retrieval_complete(op, 1);
 684         goto out;
 685 }
 686
 687 /*
 688  * read a list of pages from the cache or allocate blocks in which to store
 689  * them
 690  */
 691 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 692                                    struct list_head *pages,
 693                                    unsigned *nr_pages,
 694                                    gfp_t gfp)
 695 {
 696         struct cachefiles_object *object;
 697         struct cachefiles_cache *cache;
 698         struct list_head backpages;
 699         struct pagevec pagevec;
 700         struct inode *inode;
 701         struct page *page, *_n;
 702         unsigned shift, nrbackpages;
 703         int ret, ret2, space;
 704
 705         object = container_of(op->op.object,
 706                               struct cachefiles_object, fscache);
 707         cache = container_of(object->fscache.cache,
 708                              struct cachefiles_cache, cache);
 709
 710         _enter("{OBJ%x,%d},,%d,,",
 711                object->fscache.debug_id, atomic_read(&op->op.usage),
 712                *nr_pages);
 713
 714         if (!object->backer)
 715                 goto all_enobufs;
 716
 717         space = 1;
 718         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 719                 space = 0;
 720
 721         inode = object->backer->d_inode;
 722         ASSERT(S_ISREG(inode->i_mode));
 723         ASSERT(inode->i_mapping->a_ops->bmap);
 724         ASSERT(inode->i_mapping->a_ops->readpages);
 725
 726         /* calculate the shift required to use bmap */
 727         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 728                 goto all_enobufs;
 729
 730         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 731
 732         pagevec_init(&pagevec, 0);
 733
 734         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 735         op->op.flags |= FSCACHE_OP_ASYNC;
 736         op->op.processor = cachefiles_read_copier;
 737
 738         INIT_LIST_HEAD(&backpages);
 739         nrbackpages = 0;
 740
 741         ret = space ? -ENODATA : -ENOBUFS;
 742         list_for_each_entry_safe(page, _n, pages, lru) {
 743                 sector_t block0, block;
 744
 745                 /* we assume the absence or presence of the first block is a
 746                  * good enough indication for the page as a whole
 747                  * - TODO: don't use bmap() for this as it is _not_ actually
 748                  *   good enough for this as it doesn't indicate errors, but
 749                  *   it's all we've got for the moment
 750                  */
 751                 block0 = page->index;
 752                 block0 <<= shift;
 753
 754                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 755                                                       block0);
 756                 _debug("%llx -> %llx",
 757                        (unsigned long long) block0,
 758                        (unsigned long long) block);
 759
 760                 if (block) {
 761                         /* we have data - add it to the list to give to the
 762                          * backing fs */
 763                         list_move(&page->lru, &backpages);
 764                         (*nr_pages)--;
 765                         nrbackpages++;
 766                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 767                         fscache_mark_pages_cached(op, &pagevec);
 768                         fscache_retrieval_complete(op, 1);
 769                         ret = -ENODATA;
 770                 } else {
 771                         fscache_retrieval_complete(op, 1);
 772                 }
 773         }
 774
 775         if (pagevec_count(&pagevec) > 0)
 776                 fscache_mark_pages_cached(op, &pagevec);
 777
 778         if (list_empty(pages))
 779                 ret = 0;
 780
 781         /* submit the apparently valid pages to the backing fs to be read from
 782          * disk */
 783         if (nrbackpages > 0) {
 784                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 785                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 786                         ret = ret2;
 787         }
 788
 789         _leave(" = %d [nr=%u%s]",
 790                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 791         return ret;
 792
 793 all_enobufs:
 794         fscache_retrieval_complete(op, *nr_pages);
 795         return -ENOBUFS;
 796 }
 797
 798 /*
 799  * allocate a block in the cache in which to store a page
 800  * - cache withdrawal is prevented by the caller
 801  * - returns -EINTR if interrupted
 802  * - returns -ENOMEM if ran out of memory
 803  * - returns -ENOBUFS if no buffers can be made available
 804  * - returns -ENOBUFS if page is beyond EOF
 805  * - otherwise:
 806  *   - the metadata will be retained
 807  *   - 0 will be returned
 808  */
 809 int cachefiles_allocate_page(struct fscache_retrieval *op,
 810                              struct page *page,
 811                              gfp_t gfp)
 812 {
 813         struct cachefiles_object *object;
 814         struct cachefiles_cache *cache;
 815         int ret;
 816
 817         object = container_of(op->op.object,
 818                               struct cachefiles_object, fscache);
 819         cache = container_of(object->fscache.cache,
 820                              struct cachefiles_cache, cache);
 821
 822         _enter("%p,{%lx},", object, page->index);
 823
 824         ret = cachefiles_has_space(cache, 0, 1);
 825         if (ret == 0)
 826                 fscache_mark_page_cached(op, page);
 827         else
 828                 ret = -ENOBUFS;
 829
 830         fscache_retrieval_complete(op, 1);
 831         _leave(" = %d", ret);
 832         return ret;
 833 }
 834
 835 /*
 836  * allocate blocks in the cache in which to store a set of pages
 837  * - cache withdrawal is prevented by the caller
 838  * - returns -EINTR if interrupted
 839  * - returns -ENOMEM if ran out of memory
 840  * - returns -ENOBUFS if some buffers couldn't be made available
 841  * - returns -ENOBUFS if some pages are beyond EOF
 842  * - otherwise:
 843  *   - -ENODATA will be returned
 844  * - metadata will be retained for any page marked
 845  */
 846 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 847                               struct list_head *pages,
 848                               unsigned *nr_pages,
 849                               gfp_t gfp)
 850 {
 851         struct cachefiles_object *object;
 852         struct cachefiles_cache *cache;
 853         struct pagevec pagevec;
 854         struct page *page;
 855         int ret;
 856
 857         object = container_of(op->op.object,
 858                               struct cachefiles_object, fscache);
 859         cache = container_of(object->fscache.cache,
 860                              struct cachefiles_cache, cache);
 861
 862         _enter("%p,,,%d,", object, *nr_pages);
 863
 864         ret = cachefiles_has_space(cache, 0, *nr_pages);
 865         if (ret == 0) {
 866                 pagevec_init(&pagevec, 0);
 867
 868                 list_for_each_entry(page, pages, lru) {
 869                         if (pagevec_add(&pagevec, page) == 0)
 870                                 fscache_mark_pages_cached(op, &pagevec);
 871                 }
 872
 873                 if (pagevec_count(&pagevec) > 0)
 874                         fscache_mark_pages_cached(op, &pagevec);
 875                 ret = -ENODATA;
 876         } else {
 877                 ret = -ENOBUFS;
 878         }
 879
 880         fscache_retrieval_complete(op, *nr_pages);
 881         _leave(" = %d", ret);
 882         return ret;
 883 }
 884
 885 /*
 886  * request a page be stored in the cache
 887  * - cache withdrawal is prevented by the caller
 888  * - this request may be ignored if there's no cache block available, in which
 889  *   case -ENOBUFS will be returned
 890  * - if the op is in progress, 0 will be returned
 891  */
 892 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 893 {
 894         struct cachefiles_object *object;
 895         struct cachefiles_cache *cache;
 896         mm_segment_t old_fs;
 897         struct file *file;
 898         struct path path;
 899         loff_t pos, eof;
 900         size_t len;
 901         void *data;
 902         int ret;
 903
 904         ASSERT(op != NULL);
 905         ASSERT(page != NULL);
 906
 907         object = container_of(op->op.object,
 908                               struct cachefiles_object, fscache);
 909
 910         _enter("%p,%p{%lx},,,", object, page, page->index);
 911
 912         if (!object->backer) {
 913                 _leave(" = -ENOBUFS");
 914                 return -ENOBUFS;
 915         }
 916
 917         ASSERT(S_ISREG(object->backer->d_inode->i_mode));
 918
 919         cache = container_of(object->fscache.cache,
 920                              struct cachefiles_cache, cache);
 921
 922         /* write the page to the backing filesystem and let it store it in its
 923          * own time */
 924         path.mnt = cache->mnt;
 925         path.dentry = object->backer;
 926         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 927         if (IS_ERR(file)) {
 928                 ret = PTR_ERR(file);
 929         } else {
 930                 ret = -EIO;
 931                 if (file->f_op->write) {
 932                         pos = (loff_t) page->index << PAGE_SHIFT;
 933
 934                         /* we mustn't write more data than we have, so we have
 935                          * to beware of a partial page at EOF */
 936                         eof = object->fscache.store_limit_l;
 937                         len = PAGE_SIZE;
 938                         if (eof & ~PAGE_MASK) {
 939                                 ASSERTCMP(pos, <, eof);
 940                                 if (eof - pos < PAGE_SIZE) {
 941                                         _debug("cut short %llx to %llx",
 942                                                pos, eof);
 943                                         len = eof - pos;
 944                                         ASSERTCMP(pos + len, ==, eof);
 945                                 }
 946                         }
 947
 948                         data = kmap(page);
 949                         file_start_write(file);
 950                         old_fs = get_fs();
 951                         set_fs(KERNEL_DS);
 952                         ret = file->f_op->write(
 953                                 file, (const void __user *) data, len, &pos);
 954                         set_fs(old_fs);
 955                         kunmap(page);
 956                         file_end_write(file);
 957                         if (ret != len)
 958                                 ret = -EIO;
 959                 }
 960                 fput(file);
 961         }
 962
 963         if (ret < 0) {
 964                 if (ret == -EIO)
 965                         cachefiles_io_error_obj(
 966                                 object, "Write page to backing file failed");
 967                 ret = -ENOBUFS;
 968         }
 969
 970         _leave(" = %d", ret);
 971         return ret;
 972 }
 973
 974 /*
 975  * detach a backing block from a page
 976  * - cache withdrawal is prevented by the caller
 977  */
 978 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 979 {
 980         struct cachefiles_object *object;
 981         struct cachefiles_cache *cache;
 982
 983         object = container_of(_object, struct cachefiles_object, fscache);
 984         cache = container_of(object->fscache.cache,
 985                              struct cachefiles_cache, cache);
 986
 987         _enter("%p,{%lu}", object, page->index);
 988
 989         spin_unlock(&object->fscache.cookie->lock);
 990 }