kernel/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 = d_backing_inode(object->backer)->i_mapping;
  78         struct page *backpage = monitor->back_page, *backpage2;
  79         int ret;
  80
  81         _enter("{ino=%lx},{%lx,%lx}",
  82                d_backing_inode(object->backer)->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         int error, max;
 155
 156         op = container_of(_op, struct fscache_retrieval, op);
 157         object = container_of(op->op.object,
 158                               struct cachefiles_object, fscache);
 159
 160         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 161
 162         max = 8;
 163         spin_lock_irq(&object->work_lock);
 164
 165         while (!list_empty(&op->to_do)) {
 166                 monitor = list_entry(op->to_do.next,
 167                                      struct cachefiles_one_read, op_link);
 168                 list_del(&monitor->op_link);
 169
 170                 spin_unlock_irq(&object->work_lock);
 171
 172                 _debug("- copy {%lu}", monitor->back_page->index);
 173
 174         recheck:
 175                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 176                              &object->fscache.cookie->flags)) {
 177                         error = -ESTALE;
 178                 } else if (PageUptodate(monitor->back_page)) {
 179                         copy_highpage(monitor->netfs_page, monitor->back_page);
 180                         fscache_mark_page_cached(monitor->op,
 181                                                  monitor->netfs_page);
 182                         error = 0;
 183                 } else if (!PageError(monitor->back_page)) {
 184                         /* the page has probably been truncated */
 185                         error = cachefiles_read_reissue(object, monitor);
 186                         if (error == -EINPROGRESS)
 187                                 goto next;
 188                         goto recheck;
 189                 } else {
 190                         cachefiles_io_error_obj(
 191                                 object,
 192                                 "Readpage failed on backing file %lx",
 193                                 (unsigned long) monitor->back_page->flags);
 194                         error = -EIO;
 195                 }
 196
 197                 page_cache_release(monitor->back_page);
 198
 199                 fscache_end_io(op, monitor->netfs_page, error);
 200                 page_cache_release(monitor->netfs_page);
 201                 fscache_retrieval_complete(op, 1);
 202                 fscache_put_retrieval(op);
 203                 kfree(monitor);
 204
 205         next:
 206                 /* let the thread pool have some air occasionally */
 207                 max--;
 208                 if (max < 0 || need_resched()) {
 209                         if (!list_empty(&op->to_do))
 210                                 fscache_enqueue_retrieval(op);
 211                         _leave(" [maxed out]");
 212                         return;
 213                 }
 214
 215                 spin_lock_irq(&object->work_lock);
 216         }
 217
 218         spin_unlock_irq(&object->work_lock);
 219         _leave("");
 220 }
 221
 222 /*
 223  * read the corresponding page to the given set from the backing file
 224  * - an uncertain page is simply discarded, to be tried again another time
 225  */
 226 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 227                                             struct fscache_retrieval *op,
 228                                             struct page *netpage)
 229 {
 230         struct cachefiles_one_read *monitor;
 231         struct address_space *bmapping;
 232         struct page *newpage, *backpage;
 233         int ret;
 234
 235         _enter("");
 236
 237         _debug("read back %p{%lu,%d}",
 238                netpage, netpage->index, page_count(netpage));
 239
 240         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 241         if (!monitor)
 242                 goto nomem;
 243
 244         monitor->netfs_page = netpage;
 245         monitor->op = fscache_get_retrieval(op);
 246
 247         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 248
 249         /* attempt to get hold of the backing page */
 250         bmapping = d_backing_inode(object->backer)->i_mapping;
 251         newpage = NULL;
 252
 253         for (;;) {
 254                 backpage = find_get_page(bmapping, netpage->index);
 255                 if (backpage)
 256                         goto backing_page_already_present;
 257
 258                 if (!newpage) {
 259                         newpage = __page_cache_alloc(cachefiles_gfp |
 260                                                      __GFP_COLD);
 261                         if (!newpage)
 262                                 goto nomem_monitor;
 263                 }
 264
 265                 ret = add_to_page_cache_lru(newpage, bmapping,
 266                                             netpage->index, cachefiles_gfp);
 267                 if (ret == 0)
 268                         goto installed_new_backing_page;
 269                 if (ret != -EEXIST)
 270                         goto nomem_page;
 271         }
 272
 273         /* we've installed a new backing page, so now we need to start
 274          * it reading */
 275 installed_new_backing_page:
 276         _debug("- new %p", newpage);
 277
 278         backpage = newpage;
 279         newpage = NULL;
 280
 281 read_backing_page:
 282         ret = bmapping->a_ops->readpage(NULL, backpage);
 283         if (ret < 0)
 284                 goto read_error;
 285
 286         /* set the monitor to transfer the data across */
 287 monitor_backing_page:
 288         _debug("- monitor add");
 289
 290         /* install the monitor */
 291         page_cache_get(monitor->netfs_page);
 292         page_cache_get(backpage);
 293         monitor->back_page = backpage;
 294         monitor->monitor.private = backpage;
 295         add_page_wait_queue(backpage, &monitor->monitor);
 296         monitor = NULL;
 297
 298         /* but the page may have been read before the monitor was installed, so
 299          * the monitor may miss the event - so we have to ensure that we do get
 300          * one in such a case */
 301         if (trylock_page(backpage)) {
 302                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 303                 unlock_page(backpage);
 304         }
 305         goto success;
 306
 307         /* if the backing page is already present, it can be in one of
 308          * three states: read in progress, read failed or read okay */
 309 backing_page_already_present:
 310         _debug("- present");
 311
 312         if (newpage) {
 313                 page_cache_release(newpage);
 314                 newpage = NULL;
 315         }
 316
 317         if (PageError(backpage))
 318                 goto io_error;
 319
 320         if (PageUptodate(backpage))
 321                 goto backing_page_already_uptodate;
 322
 323         if (!trylock_page(backpage))
 324                 goto monitor_backing_page;
 325         _debug("read %p {%lx}", backpage, backpage->flags);
 326         goto read_backing_page;
 327
 328         /* the backing page is already up to date, attach the netfs
 329          * page to the pagecache and LRU and copy the data across */
 330 backing_page_already_uptodate:
 331         _debug("- uptodate");
 332
 333         fscache_mark_page_cached(op, netpage);
 334
 335         copy_highpage(netpage, backpage);
 336         fscache_end_io(op, netpage, 0);
 337         fscache_retrieval_complete(op, 1);
 338
 339 success:
 340         _debug("success");
 341         ret = 0;
 342
 343 out:
 344         if (backpage)
 345                 page_cache_release(backpage);
 346         if (monitor) {
 347                 fscache_put_retrieval(monitor->op);
 348                 kfree(monitor);
 349         }
 350         _leave(" = %d", ret);
 351         return ret;
 352
 353 read_error:
 354         _debug("read error %d", ret);
 355         if (ret == -ENOMEM) {
 356                 fscache_retrieval_complete(op, 1);
 357                 goto out;
 358         }
 359 io_error:
 360         cachefiles_io_error_obj(object, "Page read error on backing file");
 361         fscache_retrieval_complete(op, 1);
 362         ret = -ENOBUFS;
 363         goto out;
 364
 365 nomem_page:
 366         page_cache_release(newpage);
 367 nomem_monitor:
 368         fscache_put_retrieval(monitor->op);
 369         kfree(monitor);
 370 nomem:
 371         fscache_retrieval_complete(op, 1);
 372         _leave(" = -ENOMEM");
 373         return -ENOMEM;
 374 }
 375
 376 /*
 377  * read a page from the cache or allocate a block in which to store it
 378  * - cache withdrawal is prevented by the caller
 379  * - returns -EINTR if interrupted
 380  * - returns -ENOMEM if ran out of memory
 381  * - returns -ENOBUFS if no buffers can be made available
 382  * - returns -ENOBUFS if page is beyond EOF
 383  * - if the page is backed by a block in the cache:
 384  *   - a read will be started which will call the callback on completion
 385  *   - 0 will be returned
 386  * - else if the page is unbacked:
 387  *   - the metadata will be retained
 388  *   - -ENODATA will be returned
 389  */
 390 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 391                                   struct page *page,
 392                                   gfp_t gfp)
 393 {
 394         struct cachefiles_object *object;
 395         struct cachefiles_cache *cache;
 396         struct inode *inode;
 397         sector_t block0, block;
 398         unsigned shift;
 399         int ret;
 400
 401         object = container_of(op->op.object,
 402                               struct cachefiles_object, fscache);
 403         cache = container_of(object->fscache.cache,
 404                              struct cachefiles_cache, cache);
 405
 406         _enter("{%p},{%lx},,,", object, page->index);
 407
 408         if (!object->backer)
 409                 goto enobufs;
 410
 411         inode = d_backing_inode(object->backer);
 412         ASSERT(S_ISREG(inode->i_mode));
 413         ASSERT(inode->i_mapping->a_ops->bmap);
 414         ASSERT(inode->i_mapping->a_ops->readpages);
 415
 416         /* calculate the shift required to use bmap */
 417         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 418                 goto enobufs;
 419
 420         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 421
 422         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 423         op->op.flags |= FSCACHE_OP_ASYNC;
 424         op->op.processor = cachefiles_read_copier;
 425
 426         /* we assume the absence or presence of the first block is a good
 427          * enough indication for the page as a whole
 428          * - TODO: don't use bmap() for this as it is _not_ actually good
 429          *   enough for this as it doesn't indicate errors, but it's all we've
 430          *   got for the moment
 431          */
 432         block0 = page->index;
 433         block0 <<= shift;
 434
 435         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 436         _debug("%llx -> %llx",
 437                (unsigned long long) block0,
 438                (unsigned long long) block);
 439
 440         if (block) {
 441                 /* submit the apparently valid page to the backing fs to be
 442                  * read from disk */
 443                 ret = cachefiles_read_backing_file_one(object, op, page);
 444         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 445                 /* there's space in the cache we can use */
 446                 fscache_mark_page_cached(op, page);
 447                 fscache_retrieval_complete(op, 1);
 448                 ret = -ENODATA;
 449         } else {
 450                 goto enobufs;
 451         }
 452
 453         _leave(" = %d", ret);
 454         return ret;
 455
 456 enobufs:
 457         fscache_retrieval_complete(op, 1);
 458         _leave(" = -ENOBUFS");
 459         return -ENOBUFS;
 460 }
 461
 462 /*
 463  * read the corresponding pages to the given set from the backing file
 464  * - any uncertain pages are simply discarded, to be tried again another time
 465  */
 466 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 467                                         struct fscache_retrieval *op,
 468                                         struct list_head *list)
 469 {
 470         struct cachefiles_one_read *monitor = NULL;
 471         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 472         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 473         int ret = 0;
 474
 475         _enter("");
 476
 477         list_for_each_entry_safe(netpage, _n, list, lru) {
 478                 list_del(&netpage->lru);
 479
 480                 _debug("read back %p{%lu,%d}",
 481                        netpage, netpage->index, page_count(netpage));
 482
 483                 if (!monitor) {
 484                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 485                         if (!monitor)
 486                                 goto nomem;
 487
 488                         monitor->op = fscache_get_retrieval(op);
 489                         init_waitqueue_func_entry(&monitor->monitor,
 490                                                   cachefiles_read_waiter);
 491                 }
 492
 493                 for (;;) {
 494                         backpage = find_get_page(bmapping, netpage->index);
 495                         if (backpage)
 496                                 goto backing_page_already_present;
 497
 498                         if (!newpage) {
 499                                 newpage = __page_cache_alloc(cachefiles_gfp |
 500                                                              __GFP_COLD);
 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                                 page_cache_release(netpage);
 537                                 fscache_retrieval_complete(op, 1);
 538                                 continue;
 539                         }
 540                         goto nomem;
 541                 }
 542
 543                 /* install a monitor */
 544                 page_cache_get(netpage);
 545                 monitor->netfs_page = netpage;
 546
 547                 page_cache_get(backpage);
 548                 monitor->back_page = backpage;
 549                 monitor->monitor.private = backpage;
 550                 add_page_wait_queue(backpage, &monitor->monitor);
 551                 monitor = NULL;
 552
 553                 /* but the page may have been read before the monitor was
 554                  * installed, so the monitor may miss the event - so we have to
 555                  * ensure that we do get one in such a case */
 556                 if (trylock_page(backpage)) {
 557                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 558                         unlock_page(backpage);
 559                 }
 560
 561                 page_cache_release(backpage);
 562                 backpage = NULL;
 563
 564                 page_cache_release(netpage);
 565                 netpage = NULL;
 566                 continue;
 567
 568                 /* if the backing page is already present, it can be in one of
 569                  * three states: read in progress, read failed or read okay */
 570         backing_page_already_present:
 571                 _debug("- present %p", backpage);
 572
 573                 if (PageError(backpage))
 574                         goto io_error;
 575
 576                 if (PageUptodate(backpage))
 577                         goto backing_page_already_uptodate;
 578
 579                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 580
 581                 if (!trylock_page(backpage))
 582                         goto monitor_backing_page;
 583
 584                 if (PageError(backpage)) {
 585                         _debug("error %lx", backpage->flags);
 586                         unlock_page(backpage);
 587                         goto io_error;
 588                 }
 589
 590                 if (PageUptodate(backpage))
 591                         goto backing_page_already_uptodate_unlock;
 592
 593                 /* we've locked a page that's neither up to date nor erroneous,
 594                  * so we need to attempt to read it again */
 595                 goto reread_backing_page;
 596
 597                 /* the backing page is already up to date, attach the netfs
 598                  * page to the pagecache and LRU and copy the data across */
 599         backing_page_already_uptodate_unlock:
 600                 _debug("uptodate %lx", backpage->flags);
 601                 unlock_page(backpage);
 602         backing_page_already_uptodate:
 603                 _debug("- uptodate");
 604
 605                 ret = add_to_page_cache_lru(netpage, op->mapping,
 606                                             netpage->index, cachefiles_gfp);
 607                 if (ret < 0) {
 608                         if (ret == -EEXIST) {
 609                                 page_cache_release(netpage);
 610                                 fscache_retrieval_complete(op, 1);
 611                                 continue;
 612                         }
 613                         goto nomem;
 614                 }
 615
 616                 copy_highpage(netpage, backpage);
 617
 618                 page_cache_release(backpage);
 619                 backpage = NULL;
 620
 621                 fscache_mark_page_cached(op, netpage);
 622
 623                 /* the netpage is unlocked and marked up to date here */
 624                 fscache_end_io(op, netpage, 0);
 625                 page_cache_release(netpage);
 626                 netpage = NULL;
 627                 fscache_retrieval_complete(op, 1);
 628                 continue;
 629         }
 630
 631         netpage = NULL;
 632
 633         _debug("out");
 634
 635 out:
 636         /* tidy up */
 637         if (newpage)
 638                 page_cache_release(newpage);
 639         if (netpage)
 640                 page_cache_release(netpage);
 641         if (backpage)
 642                 page_cache_release(backpage);
 643         if (monitor) {
 644                 fscache_put_retrieval(op);
 645                 kfree(monitor);
 646         }
 647
 648         list_for_each_entry_safe(netpage, _n, list, lru) {
 649                 list_del(&netpage->lru);
 650                 page_cache_release(netpage);
 651                 fscache_retrieval_complete(op, 1);
 652         }
 653
 654         _leave(" = %d", ret);
 655         return ret;
 656
 657 nomem:
 658         _debug("nomem");
 659         ret = -ENOMEM;
 660         goto record_page_complete;
 661
 662 read_error:
 663         _debug("read error %d", ret);
 664         if (ret == -ENOMEM)
 665                 goto record_page_complete;
 666 io_error:
 667         cachefiles_io_error_obj(object, "Page read error on backing file");
 668         ret = -ENOBUFS;
 669 record_page_complete:
 670         fscache_retrieval_complete(op, 1);
 671         goto out;
 672 }
 673
 674 /*
 675  * read a list of pages from the cache or allocate blocks in which to store
 676  * them
 677  */
 678 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 679                                    struct list_head *pages,
 680                                    unsigned *nr_pages,
 681                                    gfp_t gfp)
 682 {
 683         struct cachefiles_object *object;
 684         struct cachefiles_cache *cache;
 685         struct list_head backpages;
 686         struct pagevec pagevec;
 687         struct inode *inode;
 688         struct page *page, *_n;
 689         unsigned shift, nrbackpages;
 690         int ret, ret2, space;
 691
 692         object = container_of(op->op.object,
 693                               struct cachefiles_object, fscache);
 694         cache = container_of(object->fscache.cache,
 695                              struct cachefiles_cache, cache);
 696
 697         _enter("{OBJ%x,%d},,%d,,",
 698                object->fscache.debug_id, atomic_read(&op->op.usage),
 699                *nr_pages);
 700
 701         if (!object->backer)
 702                 goto all_enobufs;
 703
 704         space = 1;
 705         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 706                 space = 0;
 707
 708         inode = d_backing_inode(object->backer);
 709         ASSERT(S_ISREG(inode->i_mode));
 710         ASSERT(inode->i_mapping->a_ops->bmap);
 711         ASSERT(inode->i_mapping->a_ops->readpages);
 712
 713         /* calculate the shift required to use bmap */
 714         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 715                 goto all_enobufs;
 716
 717         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 718
 719         pagevec_init(&pagevec, 0);
 720
 721         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 722         op->op.flags |= FSCACHE_OP_ASYNC;
 723         op->op.processor = cachefiles_read_copier;
 724
 725         INIT_LIST_HEAD(&backpages);
 726         nrbackpages = 0;
 727
 728         ret = space ? -ENODATA : -ENOBUFS;
 729         list_for_each_entry_safe(page, _n, pages, lru) {
 730                 sector_t block0, block;
 731
 732                 /* we assume the absence or presence of the first block is a
 733                  * good enough indication for the page as a whole
 734                  * - TODO: don't use bmap() for this as it is _not_ actually
 735                  *   good enough for this as it doesn't indicate errors, but
 736                  *   it's all we've got for the moment
 737                  */
 738                 block0 = page->index;
 739                 block0 <<= shift;
 740
 741                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 742                                                       block0);
 743                 _debug("%llx -> %llx",
 744                        (unsigned long long) block0,
 745                        (unsigned long long) block);
 746
 747                 if (block) {
 748                         /* we have data - add it to the list to give to the
 749                          * backing fs */
 750                         list_move(&page->lru, &backpages);
 751                         (*nr_pages)--;
 752                         nrbackpages++;
 753                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 754                         fscache_mark_pages_cached(op, &pagevec);
 755                         fscache_retrieval_complete(op, 1);
 756                         ret = -ENODATA;
 757                 } else {
 758                         fscache_retrieval_complete(op, 1);
 759                 }
 760         }
 761
 762         if (pagevec_count(&pagevec) > 0)
 763                 fscache_mark_pages_cached(op, &pagevec);
 764
 765         if (list_empty(pages))
 766                 ret = 0;
 767
 768         /* submit the apparently valid pages to the backing fs to be read from
 769          * disk */
 770         if (nrbackpages > 0) {
 771                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 772                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 773                         ret = ret2;
 774         }
 775
 776         _leave(" = %d [nr=%u%s]",
 777                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 778         return ret;
 779
 780 all_enobufs:
 781         fscache_retrieval_complete(op, *nr_pages);
 782         return -ENOBUFS;
 783 }
 784
 785 /*
 786  * allocate a block in the cache in which to store a page
 787  * - cache withdrawal is prevented by the caller
 788  * - returns -EINTR if interrupted
 789  * - returns -ENOMEM if ran out of memory
 790  * - returns -ENOBUFS if no buffers can be made available
 791  * - returns -ENOBUFS if page is beyond EOF
 792  * - otherwise:
 793  *   - the metadata will be retained
 794  *   - 0 will be returned
 795  */
 796 int cachefiles_allocate_page(struct fscache_retrieval *op,
 797                              struct page *page,
 798                              gfp_t gfp)
 799 {
 800         struct cachefiles_object *object;
 801         struct cachefiles_cache *cache;
 802         int ret;
 803
 804         object = container_of(op->op.object,
 805                               struct cachefiles_object, fscache);
 806         cache = container_of(object->fscache.cache,
 807                              struct cachefiles_cache, cache);
 808
 809         _enter("%p,{%lx},", object, page->index);
 810
 811         ret = cachefiles_has_space(cache, 0, 1);
 812         if (ret == 0)
 813                 fscache_mark_page_cached(op, page);
 814         else
 815                 ret = -ENOBUFS;
 816
 817         fscache_retrieval_complete(op, 1);
 818         _leave(" = %d", ret);
 819         return ret;
 820 }
 821
 822 /*
 823  * allocate blocks in the cache in which to store a set of pages
 824  * - cache withdrawal is prevented by the caller
 825  * - returns -EINTR if interrupted
 826  * - returns -ENOMEM if ran out of memory
 827  * - returns -ENOBUFS if some buffers couldn't be made available
 828  * - returns -ENOBUFS if some pages are beyond EOF
 829  * - otherwise:
 830  *   - -ENODATA will be returned
 831  * - metadata will be retained for any page marked
 832  */
 833 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 834                               struct list_head *pages,
 835                               unsigned *nr_pages,
 836                               gfp_t gfp)
 837 {
 838         struct cachefiles_object *object;
 839         struct cachefiles_cache *cache;
 840         struct pagevec pagevec;
 841         struct page *page;
 842         int ret;
 843
 844         object = container_of(op->op.object,
 845                               struct cachefiles_object, fscache);
 846         cache = container_of(object->fscache.cache,
 847                              struct cachefiles_cache, cache);
 848
 849         _enter("%p,,,%d,", object, *nr_pages);
 850
 851         ret = cachefiles_has_space(cache, 0, *nr_pages);
 852         if (ret == 0) {
 853                 pagevec_init(&pagevec, 0);
 854
 855                 list_for_each_entry(page, pages, lru) {
 856                         if (pagevec_add(&pagevec, page) == 0)
 857                                 fscache_mark_pages_cached(op, &pagevec);
 858                 }
 859
 860                 if (pagevec_count(&pagevec) > 0)
 861                         fscache_mark_pages_cached(op, &pagevec);
 862                 ret = -ENODATA;
 863         } else {
 864                 ret = -ENOBUFS;
 865         }
 866
 867         fscache_retrieval_complete(op, *nr_pages);
 868         _leave(" = %d", ret);
 869         return ret;
 870 }
 871
 872 /*
 873  * request a page be stored in the cache
 874  * - cache withdrawal is prevented by the caller
 875  * - this request may be ignored if there's no cache block available, in which
 876  *   case -ENOBUFS will be returned
 877  * - if the op is in progress, 0 will be returned
 878  */
 879 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 880 {
 881         struct cachefiles_object *object;
 882         struct cachefiles_cache *cache;
 883         struct file *file;
 884         struct path path;
 885         loff_t pos, eof;
 886         size_t len;
 887         void *data;
 888         int ret;
 889
 890         ASSERT(op != NULL);
 891         ASSERT(page != NULL);
 892
 893         object = container_of(op->op.object,
 894                               struct cachefiles_object, fscache);
 895
 896         _enter("%p,%p{%lx},,,", object, page, page->index);
 897
 898         if (!object->backer) {
 899                 _leave(" = -ENOBUFS");
 900                 return -ENOBUFS;
 901         }
 902
 903         ASSERT(d_is_reg(object->backer));
 904
 905         cache = container_of(object->fscache.cache,
 906                              struct cachefiles_cache, cache);
 907
 908         /* write the page to the backing filesystem and let it store it in its
 909          * own time */
 910         path.mnt = cache->mnt;
 911         path.dentry = object->backer;
 912         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 913         if (IS_ERR(file)) {
 914                 ret = PTR_ERR(file);
 915         } else {
 916                 pos = (loff_t) page->index << PAGE_SHIFT;
 917
 918                 /* we mustn't write more data than we have, so we have
 919                  * to beware of a partial page at EOF */
 920                 eof = object->fscache.store_limit_l;
 921                 len = PAGE_SIZE;
 922                 if (eof & ~PAGE_MASK) {
 923                         ASSERTCMP(pos, <, eof);
 924                         if (eof - pos < PAGE_SIZE) {
 925                                 _debug("cut short %llx to %llx",
 926                                        pos, eof);
 927                                 len = eof - pos;
 928                                 ASSERTCMP(pos + len, ==, eof);
 929                         }
 930                 }
 931
 932                 data = kmap(page);
 933                 ret = __kernel_write(file, data, len, &pos);
 934                 kunmap(page);
 935                 if (ret != len)
 936                         ret = -EIO;
 937                 fput(file);
 938         }
 939
 940         if (ret < 0) {
 941                 if (ret == -EIO)
 942                         cachefiles_io_error_obj(
 943                                 object, "Write page to backing file failed");
 944                 ret = -ENOBUFS;
 945         }
 946
 947         _leave(" = %d", ret);
 948         return ret;
 949 }
 950
 951 /*
 952  * detach a backing block from a page
 953  * - cache withdrawal is prevented by the caller
 954  */
 955 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 956 {
 957         struct cachefiles_object *object;
 958         struct cachefiles_cache *cache;
 959
 960         object = container_of(_object, struct cachefiles_object, fscache);
 961         cache = container_of(object->fscache.cache,
 962                              struct cachefiles_cache, cache);
 963
 964         _enter("%p,{%lu}", object, page->index);
 965
 966         spin_unlock(&object->fscache.cookie->lock);
 967 }