fs/cachefiles/rdwr.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Storage object read/write
   3  *
   4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  */
   7
   8 #include <linux/mount.h>
   9 #include <linux/slab.h>
  10 #include <linux/file.h>
  11 #include <linux/swap.h>
  12 #include "internal.h"
  13
  14 /*
  15  * detect wake up events generated by the unlocking of pages in which we're
  16  * interested
  17  * - we use this to detect read completion of backing pages
  18  * - the caller holds the waitqueue lock
  19  */
  20 static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
  21                                   int sync, void *_key)
  22 {
  23         struct cachefiles_one_read *monitor =
  24                 container_of(wait, struct cachefiles_one_read, monitor);
  25         struct cachefiles_object *object;
  26         struct fscache_retrieval *op = monitor->op;
  27         struct wait_bit_key *key = _key;
  28         struct page *page = wait->private;
  29
  30         ASSERT(key);
  31
  32         _enter("{%lu},%u,%d,{%p,%u}",
  33                monitor->netfs_page->index, mode, sync,
  34                key->flags, key->bit_nr);
  35
  36         if (key->flags != &page->flags ||
  37             key->bit_nr != PG_locked)
  38                 return 0;
  39
  40         _debug("--- monitor %p %lx ---", page, page->flags);
  41
  42         if (!PageUptodate(page) && !PageError(page)) {
  43                 /* unlocked, not uptodate and not erronous? */
  44                 _debug("page probably truncated");
  45         }
  46
  47         /* remove from the waitqueue */
  48         list_del(&wait->entry);
  49
  50         /* move onto the action list and queue for FS-Cache thread pool */
  51         ASSERT(op);
  52
  53         /* We need to temporarily bump the usage count as we don't own a ref
  54          * here otherwise cachefiles_read_copier() may free the op between the
  55          * monitor being enqueued on the op->to_do list and the op getting
  56          * enqueued on the work queue.
  57          */
  58         fscache_get_retrieval(op);
  59
  60         object = container_of(op->op.object, struct cachefiles_object, fscache);
  61         spin_lock(&object->work_lock);
  62         list_add_tail(&monitor->op_link, &op->to_do);
  63         spin_unlock(&object->work_lock);
  64
  65         fscache_enqueue_retrieval(op);
  66         fscache_put_retrieval(op);
  67         return 0;
  68 }
  69
  70 /*
  71  * handle a probably truncated page
  72  * - check to see if the page is still relevant and reissue the read if
  73  *   possible
  74  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  75  *   must wait again and 0 if successful
  76  */
  77 static int cachefiles_read_reissue(struct cachefiles_object *object,
  78                                    struct cachefiles_one_read *monitor)
  79 {
  80         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
  81         struct page *backpage = monitor->back_page, *backpage2;
  82         int ret;
  83
  84         _enter("{ino=%lx},{%lx,%lx}",
  85                d_backing_inode(object->backer)->i_ino,
  86                backpage->index, backpage->flags);
  87
  88         /* skip if the page was truncated away completely */
  89         if (backpage->mapping != bmapping) {
  90                 _leave(" = -ENODATA [mapping]");
  91                 return -ENODATA;
  92         }
  93
  94         backpage2 = find_get_page(bmapping, backpage->index);
  95         if (!backpage2) {
  96                 _leave(" = -ENODATA [gone]");
  97                 return -ENODATA;
  98         }
  99
 100         if (backpage != backpage2) {
 101                 put_page(backpage2);
 102                 _leave(" = -ENODATA [different]");
 103                 return -ENODATA;
 104         }
 105
 106         /* the page is still there and we already have a ref on it, so we don't
 107          * need a second */
 108         put_page(backpage2);
 109
 110         INIT_LIST_HEAD(&monitor->op_link);
 111         add_page_wait_queue(backpage, &monitor->monitor);
 112
 113         if (trylock_page(backpage)) {
 114                 ret = -EIO;
 115                 if (PageError(backpage))
 116                         goto unlock_discard;
 117                 ret = 0;
 118                 if (PageUptodate(backpage))
 119                         goto unlock_discard;
 120
 121                 _debug("reissue read");
 122                 ret = bmapping->a_ops->readpage(NULL, backpage);
 123                 if (ret < 0)
 124                         goto unlock_discard;
 125         }
 126
 127         /* but the page may have been read before the monitor was installed, so
 128          * the monitor may miss the event - so we have to ensure that we do get
 129          * one in such a case */
 130         if (trylock_page(backpage)) {
 131                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 132                 unlock_page(backpage);
 133         }
 134
 135         /* it'll reappear on the todo list */
 136         _leave(" = -EINPROGRESS");
 137         return -EINPROGRESS;
 138
 139 unlock_discard:
 140         unlock_page(backpage);
 141         spin_lock_irq(&object->work_lock);
 142         list_del(&monitor->op_link);
 143         spin_unlock_irq(&object->work_lock);
 144         _leave(" = %d", ret);
 145         return ret;
 146 }
 147
 148 /*
 149  * copy data from backing pages to netfs pages to complete a read operation
 150  * - driven by FS-Cache's thread pool
 151  */
 152 static void cachefiles_read_copier(struct fscache_operation *_op)
 153 {
 154         struct cachefiles_one_read *monitor;
 155         struct cachefiles_object *object;
 156         struct fscache_retrieval *op;
 157         int error, max;
 158
 159         op = container_of(_op, struct fscache_retrieval, op);
 160         object = container_of(op->op.object,
 161                               struct cachefiles_object, fscache);
 162
 163         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 164
 165         max = 8;
 166         spin_lock_irq(&object->work_lock);
 167
 168         while (!list_empty(&op->to_do)) {
 169                 monitor = list_entry(op->to_do.next,
 170                                      struct cachefiles_one_read, op_link);
 171                 list_del(&monitor->op_link);
 172
 173                 spin_unlock_irq(&object->work_lock);
 174
 175                 _debug("- copy {%lu}", monitor->back_page->index);
 176
 177         recheck:
 178                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 179                              &object->fscache.cookie->flags)) {
 180                         error = -ESTALE;
 181                 } else if (PageUptodate(monitor->back_page)) {
 182                         copy_highpage(monitor->netfs_page, monitor->back_page);
 183                         fscache_mark_page_cached(monitor->op,
 184                                                  monitor->netfs_page);
 185                         error = 0;
 186                 } else if (!PageError(monitor->back_page)) {
 187                         /* the page has probably been truncated */
 188                         error = cachefiles_read_reissue(object, monitor);
 189                         if (error == -EINPROGRESS)
 190                                 goto next;
 191                         goto recheck;
 192                 } else {
 193                         cachefiles_io_error_obj(
 194                                 object,
 195                                 "Readpage failed on backing file %lx",
 196                                 (unsigned long) monitor->back_page->flags);
 197                         error = -EIO;
 198                 }
 199
 200                 put_page(monitor->back_page);
 201
 202                 fscache_end_io(op, monitor->netfs_page, error);
 203                 put_page(monitor->netfs_page);
 204                 fscache_retrieval_complete(op, 1);
 205                 fscache_put_retrieval(op);
 206                 kfree(monitor);
 207
 208         next:
 209                 /* let the thread pool have some air occasionally */
 210                 max--;
 211                 if (max < 0 || need_resched()) {
 212                         if (!list_empty(&op->to_do))
 213                                 fscache_enqueue_retrieval(op);
 214                         _leave(" [maxed out]");
 215                         return;
 216                 }
 217
 218                 spin_lock_irq(&object->work_lock);
 219         }
 220
 221         spin_unlock_irq(&object->work_lock);
 222         _leave("");
 223 }
 224
 225 /*
 226  * read the corresponding page to the given set from the backing file
 227  * - an uncertain page is simply discarded, to be tried again another time
 228  */
 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 = d_backing_inode(object->backer)->i_mapping;
 254         newpage = NULL;
 255
 256         for (;;) {
 257                 backpage = find_get_page(bmapping, netpage->index);
 258                 if (backpage)
 259                         goto backing_page_already_present;
 260
 261                 if (!newpage) {
 262                         newpage = __page_cache_alloc(cachefiles_gfp);
 263                         if (!newpage)
 264                                 goto nomem_monitor;
 265                 }
 266
 267                 ret = add_to_page_cache_lru(newpage, bmapping,
 268                                             netpage->index, cachefiles_gfp);
 269                 if (ret == 0)
 270                         goto installed_new_backing_page;
 271                 if (ret != -EEXIST)
 272                         goto nomem_page;
 273         }
 274
 275         /* we've installed a new backing page, so now we need to start
 276          * it reading */
 277 installed_new_backing_page:
 278         _debug("- new %p", newpage);
 279
 280         backpage = newpage;
 281         newpage = NULL;
 282
 283 read_backing_page:
 284         ret = bmapping->a_ops->readpage(NULL, backpage);
 285         if (ret < 0)
 286                 goto read_error;
 287
 288         /* set the monitor to transfer the data across */
 289 monitor_backing_page:
 290         _debug("- monitor add");
 291
 292         /* install the monitor */
 293         get_page(monitor->netfs_page);
 294         get_page(backpage);
 295         monitor->back_page = backpage;
 296         monitor->monitor.private = backpage;
 297         add_page_wait_queue(backpage, &monitor->monitor);
 298         monitor = NULL;
 299
 300         /* but the page may have been read before the monitor was installed, so
 301          * the monitor may miss the event - so we have to ensure that we do get
 302          * one in such a case */
 303         if (trylock_page(backpage)) {
 304                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 305                 unlock_page(backpage);
 306         }
 307         goto success;
 308
 309         /* if the backing page is already present, it can be in one of
 310          * three states: read in progress, read failed or read okay */
 311 backing_page_already_present:
 312         _debug("- present");
 313
 314         if (newpage) {
 315                 put_page(newpage);
 316                 newpage = NULL;
 317         }
 318
 319         if (PageError(backpage))
 320                 goto io_error;
 321
 322         if (PageUptodate(backpage))
 323                 goto backing_page_already_uptodate;
 324
 325         if (!trylock_page(backpage))
 326                 goto monitor_backing_page;
 327         _debug("read %p {%lx}", backpage, backpage->flags);
 328         goto read_backing_page;
 329
 330         /* the backing page is already up to date, attach the netfs
 331          * page to the pagecache and LRU and copy the data across */
 332 backing_page_already_uptodate:
 333         _debug("- uptodate");
 334
 335         fscache_mark_page_cached(op, netpage);
 336
 337         copy_highpage(netpage, backpage);
 338         fscache_end_io(op, netpage, 0);
 339         fscache_retrieval_complete(op, 1);
 340
 341 success:
 342         _debug("success");
 343         ret = 0;
 344
 345 out:
 346         if (backpage)
 347                 put_page(backpage);
 348         if (monitor) {
 349                 fscache_put_retrieval(monitor->op);
 350                 kfree(monitor);
 351         }
 352         _leave(" = %d", ret);
 353         return ret;
 354
 355 read_error:
 356         _debug("read error %d", ret);
 357         if (ret == -ENOMEM) {
 358                 fscache_retrieval_complete(op, 1);
 359                 goto out;
 360         }
 361 io_error:
 362         cachefiles_io_error_obj(object, "Page read error on backing file");
 363         fscache_retrieval_complete(op, 1);
 364         ret = -ENOBUFS;
 365         goto out;
 366
 367 nomem_page:
 368         put_page(newpage);
 369 nomem_monitor:
 370         fscache_put_retrieval(monitor->op);
 371         kfree(monitor);
 372 nomem:
 373         fscache_retrieval_complete(op, 1);
 374         _leave(" = -ENOMEM");
 375         return -ENOMEM;
 376 }
 377
 378 /*
 379  * read a page from the cache or allocate a block in which to store it
 380  * - cache withdrawal is prevented by the caller
 381  * - returns -EINTR if interrupted
 382  * - returns -ENOMEM if ran out of memory
 383  * - returns -ENOBUFS if no buffers can be made available
 384  * - returns -ENOBUFS if page is beyond EOF
 385  * - if the page is backed by a block in the cache:
 386  *   - a read will be started which will call the callback on completion
 387  *   - 0 will be returned
 388  * - else if the page is unbacked:
 389  *   - the metadata will be retained
 390  *   - -ENODATA will be returned
 391  */
 392 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 393                                   struct page *page,
 394                                   gfp_t gfp)
 395 {
 396         struct cachefiles_object *object;
 397         struct cachefiles_cache *cache;
 398         struct inode *inode;
 399         sector_t block0, block;
 400         unsigned shift;
 401         int ret;
 402
 403         object = container_of(op->op.object,
 404                               struct cachefiles_object, fscache);
 405         cache = container_of(object->fscache.cache,
 406                              struct cachefiles_cache, cache);
 407
 408         _enter("{%p},{%lx},,,", object, page->index);
 409
 410         if (!object->backer)
 411                 goto enobufs;
 412
 413         inode = d_backing_inode(object->backer);
 414         ASSERT(S_ISREG(inode->i_mode));
 415         ASSERT(inode->i_mapping->a_ops->bmap);
 416         ASSERT(inode->i_mapping->a_ops->readpages);
 417
 418         /* calculate the shift required to use bmap */
 419         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 420
 421         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 422         op->op.flags |= FSCACHE_OP_ASYNC;
 423         op->op.processor = cachefiles_read_copier;
 424
 425         /* we assume the absence or presence of the first block is a good
 426          * enough indication for the page as a whole
 427          * - TODO: don't use bmap() for this as it is _not_ actually good
 428          *   enough for this as it doesn't indicate errors, but it's all we've
 429          *   got for the moment
 430          */
 431         block0 = page->index;
 432         block0 <<= shift;
 433
 434         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 435         _debug("%llx -> %llx",
 436                (unsigned long long) block0,
 437                (unsigned long long) block);
 438
 439         if (block) {
 440                 /* submit the apparently valid page to the backing fs to be
 441                  * read from disk */
 442                 ret = cachefiles_read_backing_file_one(object, op, page);
 443         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 444                 /* there's space in the cache we can use */
 445                 fscache_mark_page_cached(op, page);
 446                 fscache_retrieval_complete(op, 1);
 447                 ret = -ENODATA;
 448         } else {
 449                 goto enobufs;
 450         }
 451
 452         _leave(" = %d", ret);
 453         return ret;
 454
 455 enobufs:
 456         fscache_retrieval_complete(op, 1);
 457         _leave(" = -ENOBUFS");
 458         return -ENOBUFS;
 459 }
 460
 461 /*
 462  * read the corresponding pages to the given set from the backing file
 463  * - any uncertain pages are simply discarded, to be tried again another time
 464  */
 465 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 466                                         struct fscache_retrieval *op,
 467                                         struct list_head *list)
 468 {
 469         struct cachefiles_one_read *monitor = NULL;
 470         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 471         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 472         int ret = 0;
 473
 474         _enter("");
 475
 476         list_for_each_entry_safe(netpage, _n, list, lru) {
 477                 list_del(&netpage->lru);
 478
 479                 _debug("read back %p{%lu,%d}",
 480                        netpage, netpage->index, page_count(netpage));
 481
 482                 if (!monitor) {
 483                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 484                         if (!monitor)
 485                                 goto nomem;
 486
 487                         monitor->op = fscache_get_retrieval(op);
 488                         init_waitqueue_func_entry(&monitor->monitor,
 489                                                   cachefiles_read_waiter);
 490                 }
 491
 492                 for (;;) {
 493                         backpage = find_get_page(bmapping, netpage->index);
 494                         if (backpage)
 495                                 goto backing_page_already_present;
 496
 497                         if (!newpage) {
 498                                 newpage = __page_cache_alloc(cachefiles_gfp);
 499                                 if (!newpage)
 500                                         goto nomem;
 501                         }
 502
 503                         ret = add_to_page_cache_lru(newpage, bmapping,
 504                                                     netpage->index,
 505                                                     cachefiles_gfp);
 506                         if (ret == 0)
 507                                 goto installed_new_backing_page;
 508                         if (ret != -EEXIST)
 509                                 goto nomem;
 510                 }
 511
 512                 /* we've installed a new backing page, so now we need
 513                  * to start it reading */
 514         installed_new_backing_page:
 515                 _debug("- new %p", newpage);
 516
 517                 backpage = newpage;
 518                 newpage = NULL;
 519
 520         reread_backing_page:
 521                 ret = bmapping->a_ops->readpage(NULL, backpage);
 522                 if (ret < 0)
 523                         goto read_error;
 524
 525                 /* add the netfs page to the pagecache and LRU, and set the
 526                  * monitor to transfer the data across */
 527         monitor_backing_page:
 528                 _debug("- monitor add");
 529
 530                 ret = add_to_page_cache_lru(netpage, op->mapping,
 531                                             netpage->index, cachefiles_gfp);
 532                 if (ret < 0) {
 533                         if (ret == -EEXIST) {
 534                                 put_page(backpage);
 535                                 backpage = NULL;
 536                                 put_page(netpage);
 537                                 netpage = NULL;
 538                                 fscache_retrieval_complete(op, 1);
 539                                 continue;
 540                         }
 541                         goto nomem;
 542                 }
 543
 544                 /* install a monitor */
 545                 get_page(netpage);
 546                 monitor->netfs_page = netpage;
 547
 548                 get_page(backpage);
 549                 monitor->back_page = backpage;
 550                 monitor->monitor.private = backpage;
 551                 add_page_wait_queue(backpage, &monitor->monitor);
 552                 monitor = NULL;
 553
 554                 /* but the page may have been read before the monitor was
 555                  * installed, so the monitor may miss the event - so we have to
 556                  * ensure that we do get one in such a case */
 557                 if (trylock_page(backpage)) {
 558                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 559                         unlock_page(backpage);
 560                 }
 561
 562                 put_page(backpage);
 563                 backpage = NULL;
 564
 565                 put_page(netpage);
 566                 netpage = NULL;
 567                 continue;
 568
 569                 /* if the backing page is already present, it can be in one of
 570                  * three states: read in progress, read failed or read okay */
 571         backing_page_already_present:
 572                 _debug("- present %p", backpage);
 573
 574                 if (PageError(backpage))
 575                         goto io_error;
 576
 577                 if (PageUptodate(backpage))
 578                         goto backing_page_already_uptodate;
 579
 580                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 581
 582                 if (!trylock_page(backpage))
 583                         goto monitor_backing_page;
 584
 585                 if (PageError(backpage)) {
 586                         _debug("error %lx", backpage->flags);
 587                         unlock_page(backpage);
 588                         goto io_error;
 589                 }
 590
 591                 if (PageUptodate(backpage))
 592                         goto backing_page_already_uptodate_unlock;
 593
 594                 /* we've locked a page that's neither up to date nor erroneous,
 595                  * so we need to attempt to read it again */
 596                 goto reread_backing_page;
 597
 598                 /* the backing page is already up to date, attach the netfs
 599                  * page to the pagecache and LRU and copy the data across */
 600         backing_page_already_uptodate_unlock:
 601                 _debug("uptodate %lx", backpage->flags);
 602                 unlock_page(backpage);
 603         backing_page_already_uptodate:
 604                 _debug("- uptodate");
 605
 606                 ret = add_to_page_cache_lru(netpage, op->mapping,
 607                                             netpage->index, cachefiles_gfp);
 608                 if (ret < 0) {
 609                         if (ret == -EEXIST) {
 610                                 put_page(backpage);
 611                                 backpage = NULL;
 612                                 put_page(netpage);
 613                                 netpage = NULL;
 614                                 fscache_retrieval_complete(op, 1);
 615                                 continue;
 616                         }
 617                         goto nomem;
 618                 }
 619
 620                 copy_highpage(netpage, backpage);
 621
 622                 put_page(backpage);
 623                 backpage = NULL;
 624
 625                 fscache_mark_page_cached(op, netpage);
 626
 627                 /* the netpage is unlocked and marked up to date here */
 628                 fscache_end_io(op, netpage, 0);
 629                 put_page(netpage);
 630                 netpage = NULL;
 631                 fscache_retrieval_complete(op, 1);
 632                 continue;
 633         }
 634
 635         netpage = NULL;
 636
 637         _debug("out");
 638
 639 out:
 640         /* tidy up */
 641         if (newpage)
 642                 put_page(newpage);
 643         if (netpage)
 644                 put_page(netpage);
 645         if (backpage)
 646                 put_page(backpage);
 647         if (monitor) {
 648                 fscache_put_retrieval(op);
 649                 kfree(monitor);
 650         }
 651
 652         list_for_each_entry_safe(netpage, _n, list, lru) {
 653                 list_del(&netpage->lru);
 654                 put_page(netpage);
 655                 fscache_retrieval_complete(op, 1);
 656         }
 657
 658         _leave(" = %d", ret);
 659         return ret;
 660
 661 nomem:
 662         _debug("nomem");
 663         ret = -ENOMEM;
 664         goto record_page_complete;
 665
 666 read_error:
 667         _debug("read error %d", ret);
 668         if (ret == -ENOMEM)
 669                 goto record_page_complete;
 670 io_error:
 671         cachefiles_io_error_obj(object, "Page read error on backing file");
 672         ret = -ENOBUFS;
 673 record_page_complete:
 674         fscache_retrieval_complete(op, 1);
 675         goto out;
 676 }
 677
 678 /*
 679  * read a list of pages from the cache or allocate blocks in which to store
 680  * them
 681  */
 682 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 683                                    struct list_head *pages,
 684                                    unsigned *nr_pages,
 685                                    gfp_t gfp)
 686 {
 687         struct cachefiles_object *object;
 688         struct cachefiles_cache *cache;
 689         struct list_head backpages;
 690         struct pagevec pagevec;
 691         struct inode *inode;
 692         struct page *page, *_n;
 693         unsigned shift, nrbackpages;
 694         int ret, ret2, space;
 695
 696         object = container_of(op->op.object,
 697                               struct cachefiles_object, fscache);
 698         cache = container_of(object->fscache.cache,
 699                              struct cachefiles_cache, cache);
 700
 701         _enter("{OBJ%x,%d},,%d,,",
 702                object->fscache.debug_id, atomic_read(&op->op.usage),
 703                *nr_pages);
 704
 705         if (!object->backer)
 706                 goto all_enobufs;
 707
 708         space = 1;
 709         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 710                 space = 0;
 711
 712         inode = d_backing_inode(object->backer);
 713         ASSERT(S_ISREG(inode->i_mode));
 714         ASSERT(inode->i_mapping->a_ops->bmap);
 715         ASSERT(inode->i_mapping->a_ops->readpages);
 716
 717         /* calculate the shift required to use bmap */
 718         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 719
 720         pagevec_init(&pagevec);
 721
 722         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 723         op->op.flags |= FSCACHE_OP_ASYNC;
 724         op->op.processor = cachefiles_read_copier;
 725
 726         INIT_LIST_HEAD(&backpages);
 727         nrbackpages = 0;
 728
 729         ret = space ? -ENODATA : -ENOBUFS;
 730         list_for_each_entry_safe(page, _n, pages, lru) {
 731                 sector_t block0, block;
 732
 733                 /* we assume the absence or presence of the first block is a
 734                  * good enough indication for the page as a whole
 735                  * - TODO: don't use bmap() for this as it is _not_ actually
 736                  *   good enough for this as it doesn't indicate errors, but
 737                  *   it's all we've got for the moment
 738                  */
 739                 block0 = page->index;
 740                 block0 <<= shift;
 741
 742                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 743                                                       block0);
 744                 _debug("%llx -> %llx",
 745                        (unsigned long long) block0,
 746                        (unsigned long long) block);
 747
 748                 if (block) {
 749                         /* we have data - add it to the list to give to the
 750                          * backing fs */
 751                         list_move(&page->lru, &backpages);
 752                         (*nr_pages)--;
 753                         nrbackpages++;
 754                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 755                         fscache_mark_pages_cached(op, &pagevec);
 756                         fscache_retrieval_complete(op, 1);
 757                         ret = -ENODATA;
 758                 } else {
 759                         fscache_retrieval_complete(op, 1);
 760                 }
 761         }
 762
 763         if (pagevec_count(&pagevec) > 0)
 764                 fscache_mark_pages_cached(op, &pagevec);
 765
 766         if (list_empty(pages))
 767                 ret = 0;
 768
 769         /* submit the apparently valid pages to the backing fs to be read from
 770          * disk */
 771         if (nrbackpages > 0) {
 772                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 773                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 774                         ret = ret2;
 775         }
 776
 777         _leave(" = %d [nr=%u%s]",
 778                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 779         return ret;
 780
 781 all_enobufs:
 782         fscache_retrieval_complete(op, *nr_pages);
 783         return -ENOBUFS;
 784 }
 785
 786 /*
 787  * allocate a block in the cache in which to store a page
 788  * - cache withdrawal is prevented by the caller
 789  * - returns -EINTR if interrupted
 790  * - returns -ENOMEM if ran out of memory
 791  * - returns -ENOBUFS if no buffers can be made available
 792  * - returns -ENOBUFS if page is beyond EOF
 793  * - otherwise:
 794  *   - the metadata will be retained
 795  *   - 0 will be returned
 796  */
 797 int cachefiles_allocate_page(struct fscache_retrieval *op,
 798                              struct page *page,
 799                              gfp_t gfp)
 800 {
 801         struct cachefiles_object *object;
 802         struct cachefiles_cache *cache;
 803         int ret;
 804
 805         object = container_of(op->op.object,
 806                               struct cachefiles_object, fscache);
 807         cache = container_of(object->fscache.cache,
 808                              struct cachefiles_cache, cache);
 809
 810         _enter("%p,{%lx},", object, page->index);
 811
 812         ret = cachefiles_has_space(cache, 0, 1);
 813         if (ret == 0)
 814                 fscache_mark_page_cached(op, page);
 815         else
 816                 ret = -ENOBUFS;
 817
 818         fscache_retrieval_complete(op, 1);
 819         _leave(" = %d", ret);
 820         return ret;
 821 }
 822
 823 /*
 824  * allocate blocks in the cache in which to store a set of pages
 825  * - cache withdrawal is prevented by the caller
 826  * - returns -EINTR if interrupted
 827  * - returns -ENOMEM if ran out of memory
 828  * - returns -ENOBUFS if some buffers couldn't be made available
 829  * - returns -ENOBUFS if some pages are beyond EOF
 830  * - otherwise:
 831  *   - -ENODATA will be returned
 832  * - metadata will be retained for any page marked
 833  */
 834 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 835                               struct list_head *pages,
 836                               unsigned *nr_pages,
 837                               gfp_t gfp)
 838 {
 839         struct cachefiles_object *object;
 840         struct cachefiles_cache *cache;
 841         struct pagevec pagevec;
 842         struct page *page;
 843         int ret;
 844
 845         object = container_of(op->op.object,
 846                               struct cachefiles_object, fscache);
 847         cache = container_of(object->fscache.cache,
 848                              struct cachefiles_cache, cache);
 849
 850         _enter("%p,,,%d,", object, *nr_pages);
 851
 852         ret = cachefiles_has_space(cache, 0, *nr_pages);
 853         if (ret == 0) {
 854                 pagevec_init(&pagevec);
 855
 856                 list_for_each_entry(page, pages, lru) {
 857                         if (pagevec_add(&pagevec, page) == 0)
 858                                 fscache_mark_pages_cached(op, &pagevec);
 859                 }
 860
 861                 if (pagevec_count(&pagevec) > 0)
 862                         fscache_mark_pages_cached(op, &pagevec);
 863                 ret = -ENODATA;
 864         } else {
 865                 ret = -ENOBUFS;
 866         }
 867
 868         fscache_retrieval_complete(op, *nr_pages);
 869         _leave(" = %d", ret);
 870         return ret;
 871 }
 872
 873 /*
 874  * request a page be stored in the cache
 875  * - cache withdrawal is prevented by the caller
 876  * - this request may be ignored if there's no cache block available, in which
 877  *   case -ENOBUFS will be returned
 878  * - if the op is in progress, 0 will be returned
 879  */
 880 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 881 {
 882         struct cachefiles_object *object;
 883         struct cachefiles_cache *cache;
 884         struct file *file;
 885         struct path path;
 886         loff_t pos, eof;
 887         size_t len;
 888         void *data;
 889         int ret = -ENOBUFS;
 890
 891         ASSERT(op != NULL);
 892         ASSERT(page != NULL);
 893
 894         object = container_of(op->op.object,
 895                               struct cachefiles_object, fscache);
 896
 897         _enter("%p,%p{%lx},,,", object, page, page->index);
 898
 899         if (!object->backer) {
 900                 _leave(" = -ENOBUFS");
 901                 return -ENOBUFS;
 902         }
 903
 904         ASSERT(d_is_reg(object->backer));
 905
 906         cache = container_of(object->fscache.cache,
 907                              struct cachefiles_cache, cache);
 908
 909         pos = (loff_t)page->index << PAGE_SHIFT;
 910
 911         /* We mustn't write more data than we have, so we have to beware of a
 912          * partial page at EOF.
 913          */
 914         eof = object->fscache.store_limit_l;
 915         if (pos >= eof)
 916                 goto error;
 917
 918         /* write the page to the backing filesystem and let it store it in its
 919          * own time */
 920         path.mnt = cache->mnt;
 921         path.dentry = object->backer;
 922         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 923         if (IS_ERR(file)) {
 924                 ret = PTR_ERR(file);
 925                 goto error_2;
 926         }
 927
 928         len = PAGE_SIZE;
 929         if (eof & ~PAGE_MASK) {
 930                 if (eof - pos < PAGE_SIZE) {
 931                         _debug("cut short %llx to %llx",
 932                                pos, eof);
 933                         len = eof - pos;
 934                         ASSERTCMP(pos + len, ==, eof);
 935                 }
 936         }
 937
 938         data = kmap(page);
 939         ret = __kernel_write(file, data, len, &pos);
 940         kunmap(page);
 941         fput(file);
 942         if (ret != len)
 943                 goto error_eio;
 944
 945         _leave(" = 0");
 946         return 0;
 947
 948 error_eio:
 949         ret = -EIO;
 950 error_2:
 951         if (ret == -EIO)
 952                 cachefiles_io_error_obj(object,
 953                                         "Write page to backing file failed");
 954 error:
 955         _leave(" = -ENOBUFS [%d]", ret);
 956         return -ENOBUFS;
 957 }
 958
 959 /*
 960  * detach a backing block from a page
 961  * - cache withdrawal is prevented by the caller
 962  */
 963 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 964         __releases(&object->fscache.cookie->lock)
 965 {
 966         struct cachefiles_object *object;
 967
 968         object = container_of(_object, struct cachefiles_object, fscache);
 969
 970         _enter("%p,{%lu}", object, page->index);
 971
 972         spin_unlock(&object->fscache.cookie->lock);
 973 }