fs/cachefiles/rdwr.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /* Storage object read/write
   3  *
   4  * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
   5  * Written by David Howells (dhowells@redhat.com)
   6  */
   7
   8 #include <linux/mount.h>
   9 #include <linux/slab.h>
  10 #include <linux/file.h>
  11 #include <linux/swap.h>
  12 #include "internal.h"
  13
  14 /*
  15  * detect wake up events generated by the unlocking of pages in which we're
  16  * interested
  17  * - we use this to detect read completion of backing pages
  18  * - the caller holds the waitqueue lock
  19  */
  20 static int cachefiles_read_waiter(wait_queue_entry_t *wait, unsigned mode,
  21                                   int sync, void *_key)
  22 {
  23         struct cachefiles_one_read *monitor =
  24                 container_of(wait, struct cachefiles_one_read, monitor);
  25         struct cachefiles_object *object;
  26         struct fscache_retrieval *op = monitor->op;
  27         struct wait_bit_key *key = _key;
  28         struct page *page = wait->private;
  29
  30         ASSERT(key);
  31
  32         _enter("{%lu},%u,%d,{%p,%u}",
  33                monitor->netfs_page->index, mode, sync,
  34                key->flags, key->bit_nr);
  35
  36         if (key->flags != &page->flags ||
  37             key->bit_nr != PG_locked)
  38                 return 0;
  39
  40         _debug("--- monitor %p %lx ---", page, page->flags);
  41
  42         if (!PageUptodate(page) && !PageError(page)) {
  43                 /* unlocked, not uptodate and not erronous? */
  44                 _debug("page probably truncated");
  45         }
  46
  47         /* remove from the waitqueue */
  48         list_del(&wait->entry);
  49
  50         /* move onto the action list and queue for FS-Cache thread pool */
  51         ASSERT(op);
  52
  53         /* We need to temporarily bump the usage count as we don't own a ref
  54          * here otherwise cachefiles_read_copier() may free the op between the
  55          * monitor being enqueued on the op->to_do list and the op getting
  56          * enqueued on the work queue.
  57          */
  58         fscache_get_retrieval(op);
  59
  60         object = container_of(op->op.object, struct cachefiles_object, fscache);
  61         spin_lock(&object->work_lock);
  62         list_add_tail(&monitor->op_link, &op->to_do);
  63         fscache_enqueue_retrieval(op);
  64         spin_unlock(&object->work_lock);
  65
  66         fscache_put_retrieval(op);
  67         return 0;
  68 }
  69
  70 /*
  71  * handle a probably truncated page
  72  * - check to see if the page is still relevant and reissue the read if
  73  *   possible
  74  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  75  *   must wait again and 0 if successful
  76  */
  77 static int cachefiles_read_reissue(struct cachefiles_object *object,
  78                                    struct cachefiles_one_read *monitor)
  79 {
  80         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
  81         struct page *backpage = monitor->back_page, *backpage2;
  82         int ret;
  83
  84         _enter("{ino=%lx},{%lx,%lx}",
  85                d_backing_inode(object->backer)->i_ino,
  86                backpage->index, backpage->flags);
  87
  88         /* skip if the page was truncated away completely */
  89         if (backpage->mapping != bmapping) {
  90                 _leave(" = -ENODATA [mapping]");
  91                 return -ENODATA;
  92         }
  93
  94         backpage2 = find_get_page(bmapping, backpage->index);
  95         if (!backpage2) {
  96                 _leave(" = -ENODATA [gone]");
  97                 return -ENODATA;
  98         }
  99
 100         if (backpage != backpage2) {
 101                 put_page(backpage2);
 102                 _leave(" = -ENODATA [different]");
 103                 return -ENODATA;
 104         }
 105
 106         /* the page is still there and we already have a ref on it, so we don't
 107          * need a second */
 108         put_page(backpage2);
 109
 110         INIT_LIST_HEAD(&monitor->op_link);
 111         add_page_wait_queue(backpage, &monitor->monitor);
 112
 113         if (trylock_page(backpage)) {
 114                 ret = -EIO;
 115                 if (PageError(backpage))
 116                         goto unlock_discard;
 117                 ret = 0;
 118                 if (PageUptodate(backpage))
 119                         goto unlock_discard;
 120
 121                 _debug("reissue read");
 122                 ret = bmapping->a_ops->readpage(NULL, backpage);
 123                 if (ret < 0)
 124                         goto discard;
 125         }
 126
 127         /* but the page may have been read before the monitor was installed, so
 128          * the monitor may miss the event - so we have to ensure that we do get
 129          * one in such a case */
 130         if (trylock_page(backpage)) {
 131                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 132                 unlock_page(backpage);
 133         }
 134
 135         /* it'll reappear on the todo list */
 136         _leave(" = -EINPROGRESS");
 137         return -EINPROGRESS;
 138
 139 unlock_discard:
 140         unlock_page(backpage);
 141 discard:
 142         spin_lock_irq(&object->work_lock);
 143         list_del(&monitor->op_link);
 144         spin_unlock_irq(&object->work_lock);
 145         _leave(" = %d", ret);
 146         return ret;
 147 }
 148
 149 /*
 150  * copy data from backing pages to netfs pages to complete a read operation
 151  * - driven by FS-Cache's thread pool
 152  */
 153 static void cachefiles_read_copier(struct fscache_operation *_op)
 154 {
 155         struct cachefiles_one_read *monitor;
 156         struct cachefiles_object *object;
 157         struct fscache_retrieval *op;
 158         int error, max;
 159
 160         op = container_of(_op, struct fscache_retrieval, op);
 161         object = container_of(op->op.object,
 162                               struct cachefiles_object, fscache);
 163
 164         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 165
 166         max = 8;
 167         spin_lock_irq(&object->work_lock);
 168
 169         while (!list_empty(&op->to_do)) {
 170                 monitor = list_entry(op->to_do.next,
 171                                      struct cachefiles_one_read, op_link);
 172                 list_del(&monitor->op_link);
 173
 174                 spin_unlock_irq(&object->work_lock);
 175
 176                 _debug("- copy {%lu}", monitor->back_page->index);
 177
 178         recheck:
 179                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 180                              &object->fscache.cookie->flags)) {
 181                         error = -ESTALE;
 182                 } else if (PageUptodate(monitor->back_page)) {
 183                         copy_highpage(monitor->netfs_page, monitor->back_page);
 184                         fscache_mark_page_cached(monitor->op,
 185                                                  monitor->netfs_page);
 186                         error = 0;
 187                 } else if (!PageError(monitor->back_page)) {
 188                         /* the page has probably been truncated */
 189                         error = cachefiles_read_reissue(object, monitor);
 190                         if (error == -EINPROGRESS)
 191                                 goto next;
 192                         goto recheck;
 193                 } else {
 194                         cachefiles_io_error_obj(
 195                                 object,
 196                                 "Readpage failed on backing file %lx",
 197                                 (unsigned long) monitor->back_page->flags);
 198                         error = -EIO;
 199                 }
 200
 201                 put_page(monitor->back_page);
 202
 203                 fscache_end_io(op, monitor->netfs_page, error);
 204                 put_page(monitor->netfs_page);
 205                 fscache_retrieval_complete(op, 1);
 206                 fscache_put_retrieval(op);
 207                 kfree(monitor);
 208
 209         next:
 210                 /* let the thread pool have some air occasionally */
 211                 max--;
 212                 if (max < 0 || need_resched()) {
 213                         if (!list_empty(&op->to_do))
 214                                 fscache_enqueue_retrieval(op);
 215                         _leave(" [maxed out]");
 216                         return;
 217                 }
 218
 219                 spin_lock_irq(&object->work_lock);
 220         }
 221
 222         spin_unlock_irq(&object->work_lock);
 223         _leave("");
 224 }
 225
 226 /*
 227  * read the corresponding page to the given set from the backing file
 228  * - an uncertain page is simply discarded, to be tried again another time
 229  */
 230 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 231                                             struct fscache_retrieval *op,
 232                                             struct page *netpage)
 233 {
 234         struct cachefiles_one_read *monitor;
 235         struct address_space *bmapping;
 236         struct page *newpage, *backpage;
 237         int ret;
 238
 239         _enter("");
 240
 241         _debug("read back %p{%lu,%d}",
 242                netpage, netpage->index, page_count(netpage));
 243
 244         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 245         if (!monitor)
 246                 goto nomem;
 247
 248         monitor->netfs_page = netpage;
 249         monitor->op = fscache_get_retrieval(op);
 250
 251         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 252
 253         /* attempt to get hold of the backing page */
 254         bmapping = d_backing_inode(object->backer)->i_mapping;
 255         newpage = NULL;
 256
 257         for (;;) {
 258                 backpage = find_get_page(bmapping, netpage->index);
 259                 if (backpage)
 260                         goto backing_page_already_present;
 261
 262                 if (!newpage) {
 263                         newpage = __page_cache_alloc(cachefiles_gfp);
 264                         if (!newpage)
 265                                 goto nomem_monitor;
 266                 }
 267
 268                 ret = add_to_page_cache_lru(newpage, bmapping,
 269                                             netpage->index, cachefiles_gfp);
 270                 if (ret == 0)
 271                         goto installed_new_backing_page;
 272                 if (ret != -EEXIST)
 273                         goto nomem_page;
 274         }
 275
 276         /* we've installed a new backing page, so now we need to start
 277          * it reading */
 278 installed_new_backing_page:
 279         _debug("- new %p", newpage);
 280
 281         backpage = newpage;
 282         newpage = NULL;
 283
 284 read_backing_page:
 285         ret = bmapping->a_ops->readpage(NULL, backpage);
 286         if (ret < 0)
 287                 goto read_error;
 288
 289         /* set the monitor to transfer the data across */
 290 monitor_backing_page:
 291         _debug("- monitor add");
 292
 293         /* install the monitor */
 294         get_page(monitor->netfs_page);
 295         get_page(backpage);
 296         monitor->back_page = backpage;
 297         monitor->monitor.private = backpage;
 298         add_page_wait_queue(backpage, &monitor->monitor);
 299         monitor = NULL;
 300
 301         /* but the page may have been read before the monitor was installed, so
 302          * the monitor may miss the event - so we have to ensure that we do get
 303          * one in such a case */
 304         if (trylock_page(backpage)) {
 305                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 306                 unlock_page(backpage);
 307         }
 308         goto success;
 309
 310         /* if the backing page is already present, it can be in one of
 311          * three states: read in progress, read failed or read okay */
 312 backing_page_already_present:
 313         _debug("- present");
 314
 315         if (newpage) {
 316                 put_page(newpage);
 317                 newpage = NULL;
 318         }
 319
 320         if (PageError(backpage))
 321                 goto io_error;
 322
 323         if (PageUptodate(backpage))
 324                 goto backing_page_already_uptodate;
 325
 326         if (!trylock_page(backpage))
 327                 goto monitor_backing_page;
 328         _debug("read %p {%lx}", backpage, backpage->flags);
 329         goto read_backing_page;
 330
 331         /* the backing page is already up to date, attach the netfs
 332          * page to the pagecache and LRU and copy the data across */
 333 backing_page_already_uptodate:
 334         _debug("- uptodate");
 335
 336         fscache_mark_page_cached(op, netpage);
 337
 338         copy_highpage(netpage, backpage);
 339         fscache_end_io(op, netpage, 0);
 340         fscache_retrieval_complete(op, 1);
 341
 342 success:
 343         _debug("success");
 344         ret = 0;
 345
 346 out:
 347         if (backpage)
 348                 put_page(backpage);
 349         if (monitor) {
 350                 fscache_put_retrieval(monitor->op);
 351                 kfree(monitor);
 352         }
 353         _leave(" = %d", ret);
 354         return ret;
 355
 356 read_error:
 357         _debug("read error %d", ret);
 358         if (ret == -ENOMEM) {
 359                 fscache_retrieval_complete(op, 1);
 360                 goto out;
 361         }
 362 io_error:
 363         cachefiles_io_error_obj(object, "Page read error on backing file");
 364         fscache_retrieval_complete(op, 1);
 365         ret = -ENOBUFS;
 366         goto out;
 367
 368 nomem_page:
 369         put_page(newpage);
 370 nomem_monitor:
 371         fscache_put_retrieval(monitor->op);
 372         kfree(monitor);
 373 nomem:
 374         fscache_retrieval_complete(op, 1);
 375         _leave(" = -ENOMEM");
 376         return -ENOMEM;
 377 }
 378
 379 /*
 380  * read a page from the cache or allocate a block in which to store it
 381  * - cache withdrawal is prevented by the caller
 382  * - returns -EINTR if interrupted
 383  * - returns -ENOMEM if ran out of memory
 384  * - returns -ENOBUFS if no buffers can be made available
 385  * - returns -ENOBUFS if page is beyond EOF
 386  * - if the page is backed by a block in the cache:
 387  *   - a read will be started which will call the callback on completion
 388  *   - 0 will be returned
 389  * - else if the page is unbacked:
 390  *   - the metadata will be retained
 391  *   - -ENODATA will be returned
 392  */
 393 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 394                                   struct page *page,
 395                                   gfp_t gfp)
 396 {
 397         struct cachefiles_object *object;
 398         struct cachefiles_cache *cache;
 399         struct inode *inode;
 400         sector_t block;
 401         unsigned shift;
 402         int ret, ret2;
 403
 404         object = container_of(op->op.object,
 405                               struct cachefiles_object, fscache);
 406         cache = container_of(object->fscache.cache,
 407                              struct cachefiles_cache, cache);
 408
 409         _enter("{%p},{%lx},,,", object, page->index);
 410
 411         if (!object->backer)
 412                 goto enobufs;
 413
 414         inode = d_backing_inode(object->backer);
 415         ASSERT(S_ISREG(inode->i_mode));
 416
 417         /* calculate the shift required to use bmap */
 418         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 419
 420         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 421         op->op.flags |= FSCACHE_OP_ASYNC;
 422         op->op.processor = cachefiles_read_copier;
 423
 424         /* we assume the absence or presence of the first block is a good
 425          * enough indication for the page as a whole
 426          * - TODO: don't use bmap() for this as it is _not_ actually good
 427          *   enough for this as it doesn't indicate errors, but it's all we've
 428          *   got for the moment
 429          */
 430         block = page->index;
 431         block <<= shift;
 432
 433         ret2 = bmap(inode, &block);
 434         ASSERT(ret2 == 0);
 435
 436         _debug("%llx -> %llx",
 437                (unsigned long long) (page->index << shift),
 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                                 if (!newpage)
 501                                         goto nomem;
 502                         }
 503
 504                         ret = add_to_page_cache_lru(newpage, bmapping,
 505                                                     netpage->index,
 506                                                     cachefiles_gfp);
 507                         if (ret == 0)
 508                                 goto installed_new_backing_page;
 509                         if (ret != -EEXIST)
 510                                 goto nomem;
 511                 }
 512
 513                 /* we've installed a new backing page, so now we need
 514                  * to start it reading */
 515         installed_new_backing_page:
 516                 _debug("- new %p", newpage);
 517
 518                 backpage = newpage;
 519                 newpage = NULL;
 520
 521         reread_backing_page:
 522                 ret = bmapping->a_ops->readpage(NULL, backpage);
 523                 if (ret < 0)
 524                         goto read_error;
 525
 526                 /* add the netfs page to the pagecache and LRU, and set the
 527                  * monitor to transfer the data across */
 528         monitor_backing_page:
 529                 _debug("- monitor add");
 530
 531                 ret = add_to_page_cache_lru(netpage, op->mapping,
 532                                             netpage->index, cachefiles_gfp);
 533                 if (ret < 0) {
 534                         if (ret == -EEXIST) {
 535                                 put_page(backpage);
 536                                 backpage = NULL;
 537                                 put_page(netpage);
 538                                 netpage = NULL;
 539                                 fscache_retrieval_complete(op, 1);
 540                                 continue;
 541                         }
 542                         goto nomem;
 543                 }
 544
 545                 /* install a monitor */
 546                 get_page(netpage);
 547                 monitor->netfs_page = netpage;
 548
 549                 get_page(backpage);
 550                 monitor->back_page = backpage;
 551                 monitor->monitor.private = backpage;
 552                 add_page_wait_queue(backpage, &monitor->monitor);
 553                 monitor = NULL;
 554
 555                 /* but the page may have been read before the monitor was
 556                  * installed, so the monitor may miss the event - so we have to
 557                  * ensure that we do get one in such a case */
 558                 if (trylock_page(backpage)) {
 559                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 560                         unlock_page(backpage);
 561                 }
 562
 563                 put_page(backpage);
 564                 backpage = NULL;
 565
 566                 put_page(netpage);
 567                 netpage = NULL;
 568                 continue;
 569
 570                 /* if the backing page is already present, it can be in one of
 571                  * three states: read in progress, read failed or read okay */
 572         backing_page_already_present:
 573                 _debug("- present %p", backpage);
 574
 575                 if (PageError(backpage))
 576                         goto io_error;
 577
 578                 if (PageUptodate(backpage))
 579                         goto backing_page_already_uptodate;
 580
 581                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 582
 583                 if (!trylock_page(backpage))
 584                         goto monitor_backing_page;
 585
 586                 if (PageError(backpage)) {
 587                         _debug("error %lx", backpage->flags);
 588                         unlock_page(backpage);
 589                         goto io_error;
 590                 }
 591
 592                 if (PageUptodate(backpage))
 593                         goto backing_page_already_uptodate_unlock;
 594
 595                 /* we've locked a page that's neither up to date nor erroneous,
 596                  * so we need to attempt to read it again */
 597                 goto reread_backing_page;
 598
 599                 /* the backing page is already up to date, attach the netfs
 600                  * page to the pagecache and LRU and copy the data across */
 601         backing_page_already_uptodate_unlock:
 602                 _debug("uptodate %lx", backpage->flags);
 603                 unlock_page(backpage);
 604         backing_page_already_uptodate:
 605                 _debug("- uptodate");
 606
 607                 ret = add_to_page_cache_lru(netpage, op->mapping,
 608                                             netpage->index, cachefiles_gfp);
 609                 if (ret < 0) {
 610                         if (ret == -EEXIST) {
 611                                 put_page(backpage);
 612                                 backpage = NULL;
 613                                 put_page(netpage);
 614                                 netpage = NULL;
 615                                 fscache_retrieval_complete(op, 1);
 616                                 continue;
 617                         }
 618                         goto nomem;
 619                 }
 620
 621                 copy_highpage(netpage, backpage);
 622
 623                 put_page(backpage);
 624                 backpage = NULL;
 625
 626                 fscache_mark_page_cached(op, netpage);
 627
 628                 /* the netpage is unlocked and marked up to date here */
 629                 fscache_end_io(op, netpage, 0);
 630                 put_page(netpage);
 631                 netpage = NULL;
 632                 fscache_retrieval_complete(op, 1);
 633                 continue;
 634         }
 635
 636         netpage = NULL;
 637
 638         _debug("out");
 639
 640 out:
 641         /* tidy up */
 642         if (newpage)
 643                 put_page(newpage);
 644         if (netpage)
 645                 put_page(netpage);
 646         if (backpage)
 647                 put_page(backpage);
 648         if (monitor) {
 649                 fscache_put_retrieval(op);
 650                 kfree(monitor);
 651         }
 652
 653         list_for_each_entry_safe(netpage, _n, list, lru) {
 654                 list_del(&netpage->lru);
 655                 put_page(netpage);
 656                 fscache_retrieval_complete(op, 1);
 657         }
 658
 659         _leave(" = %d", ret);
 660         return ret;
 661
 662 nomem:
 663         _debug("nomem");
 664         ret = -ENOMEM;
 665         goto record_page_complete;
 666
 667 read_error:
 668         _debug("read error %d", ret);
 669         if (ret == -ENOMEM)
 670                 goto record_page_complete;
 671 io_error:
 672         cachefiles_io_error_obj(object, "Page read error on backing file");
 673         ret = -ENOBUFS;
 674 record_page_complete:
 675         fscache_retrieval_complete(op, 1);
 676         goto out;
 677 }
 678
 679 /*
 680  * read a list of pages from the cache or allocate blocks in which to store
 681  * them
 682  */
 683 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 684                                    struct list_head *pages,
 685                                    unsigned *nr_pages,
 686                                    gfp_t gfp)
 687 {
 688         struct cachefiles_object *object;
 689         struct cachefiles_cache *cache;
 690         struct list_head backpages;
 691         struct pagevec pagevec;
 692         struct inode *inode;
 693         struct page *page, *_n;
 694         unsigned shift, nrbackpages;
 695         int ret, ret2, space;
 696
 697         object = container_of(op->op.object,
 698                               struct cachefiles_object, fscache);
 699         cache = container_of(object->fscache.cache,
 700                              struct cachefiles_cache, cache);
 701
 702         _enter("{OBJ%x,%d},,%d,,",
 703                object->fscache.debug_id, atomic_read(&op->op.usage),
 704                *nr_pages);
 705
 706         if (!object->backer)
 707                 goto all_enobufs;
 708
 709         space = 1;
 710         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 711                 space = 0;
 712
 713         inode = d_backing_inode(object->backer);
 714         ASSERT(S_ISREG(inode->i_mode));
 715
 716         /* calculate the shift required to use bmap */
 717         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 718
 719         pagevec_init(&pagevec);
 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 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                 block = page->index;
 739                 block <<= shift;
 740
 741                 ret2 = bmap(inode, &block);
 742                 ASSERT(ret2 == 0);
 743
 744                 _debug("%llx -> %llx",
 745                        (unsigned long long) (page->index << shift),
 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 }