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 fscache_retrieval *op = monitor->op;
  31         struct wait_bit_key *key = _key;
  32         struct page *page = wait->private;
  33
  34         ASSERT(key);
  35
  36         _enter("{%lu},%u,%d,{%p,%u}",
  37                monitor->netfs_page->index, mode, sync,
  38                key->flags, key->bit_nr);
  39
  40         if (key->flags != &page->flags ||
  41             key->bit_nr != PG_locked)
  42                 return 0;
  43
  44         _debug("--- monitor %p %lx ---", page, page->flags);
  45
  46         if (!PageUptodate(page) && !PageError(page)) {
  47                 /* unlocked, not uptodate and not erronous? */
  48                 _debug("page probably truncated");
  49         }
  50
  51         /* remove from the waitqueue */
  52         list_del(&wait->task_list);
  53
  54         /* move onto the action list and queue for FS-Cache thread pool */
  55         ASSERT(op);
  56
  57         /* We need to temporarily bump the usage count as we don't own a ref
  58          * here otherwise cachefiles_read_copier() may free the op between the
  59          * monitor being enqueued on the op->to_do list and the op getting
  60          * enqueued on the work queue.
  61          */
  62         fscache_get_retrieval(op);
  63
  64         object = container_of(op->op.object, struct cachefiles_object, fscache);
  65         spin_lock(&object->work_lock);
  66         list_add_tail(&monitor->op_link, &op->to_do);
  67         fscache_enqueue_retrieval(op);
  68         spin_unlock(&object->work_lock);
  69
  70         fscache_put_retrieval(op);
  71         return 0;
  72 }
  73
  74 /*
  75  * handle a probably truncated page
  76  * - check to see if the page is still relevant and reissue the read if
  77  *   possible
  78  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  79  *   must wait again and 0 if successful
  80  */
  81 static int cachefiles_read_reissue(struct cachefiles_object *object,
  82                                    struct cachefiles_one_read *monitor)
  83 {
  84         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
  85         struct page *backpage = monitor->back_page, *backpage2;
  86         int ret;
  87
  88         _enter("{ino=%lx},{%lx,%lx}",
  89                d_backing_inode(object->backer)->i_ino,
  90                backpage->index, backpage->flags);
  91
  92         /* skip if the page was truncated away completely */
  93         if (backpage->mapping != bmapping) {
  94                 _leave(" = -ENODATA [mapping]");
  95                 return -ENODATA;
  96         }
  97
  98         backpage2 = find_get_page(bmapping, backpage->index);
  99         if (!backpage2) {
 100                 _leave(" = -ENODATA [gone]");
 101                 return -ENODATA;
 102         }
 103
 104         if (backpage != backpage2) {
 105                 put_page(backpage2);
 106                 _leave(" = -ENODATA [different]");
 107                 return -ENODATA;
 108         }
 109
 110         /* the page is still there and we already have a ref on it, so we don't
 111          * need a second */
 112         put_page(backpage2);
 113
 114         INIT_LIST_HEAD(&monitor->op_link);
 115         add_page_wait_queue(backpage, &monitor->monitor);
 116
 117         if (trylock_page(backpage)) {
 118                 ret = -EIO;
 119                 if (PageError(backpage))
 120                         goto unlock_discard;
 121                 ret = 0;
 122                 if (PageUptodate(backpage))
 123                         goto unlock_discard;
 124
 125                 _debug("reissue read");
 126                 ret = bmapping->a_ops->readpage(NULL, backpage);
 127                 if (ret < 0)
 128                         goto unlock_discard;
 129         }
 130
 131         /* but the page may have been read before the monitor was installed, so
 132          * the monitor may miss the event - so we have to ensure that we do get
 133          * one in such a case */
 134         if (trylock_page(backpage)) {
 135                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 136                 unlock_page(backpage);
 137         }
 138
 139         /* it'll reappear on the todo list */
 140         _leave(" = -EINPROGRESS");
 141         return -EINPROGRESS;
 142
 143 unlock_discard:
 144         unlock_page(backpage);
 145         spin_lock_irq(&object->work_lock);
 146         list_del(&monitor->op_link);
 147         spin_unlock_irq(&object->work_lock);
 148         _leave(" = %d", ret);
 149         return ret;
 150 }
 151
 152 /*
 153  * copy data from backing pages to netfs pages to complete a read operation
 154  * - driven by FS-Cache's thread pool
 155  */
 156 static void cachefiles_read_copier(struct fscache_operation *_op)
 157 {
 158         struct cachefiles_one_read *monitor;
 159         struct cachefiles_object *object;
 160         struct fscache_retrieval *op;
 161         int error, max;
 162
 163         op = container_of(_op, struct fscache_retrieval, op);
 164         object = container_of(op->op.object,
 165                               struct cachefiles_object, fscache);
 166
 167         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 168
 169         max = 8;
 170         spin_lock_irq(&object->work_lock);
 171
 172         while (!list_empty(&op->to_do)) {
 173                 monitor = list_entry(op->to_do.next,
 174                                      struct cachefiles_one_read, op_link);
 175                 list_del(&monitor->op_link);
 176
 177                 spin_unlock_irq(&object->work_lock);
 178
 179                 _debug("- copy {%lu}", monitor->back_page->index);
 180
 181         recheck:
 182                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 183                              &object->fscache.cookie->flags)) {
 184                         error = -ESTALE;
 185                 } else if (PageUptodate(monitor->back_page)) {
 186                         copy_highpage(monitor->netfs_page, monitor->back_page);
 187                         fscache_mark_page_cached(monitor->op,
 188                                                  monitor->netfs_page);
 189                         error = 0;
 190                 } else if (!PageError(monitor->back_page)) {
 191                         /* the page has probably been truncated */
 192                         error = cachefiles_read_reissue(object, monitor);
 193                         if (error == -EINPROGRESS)
 194                                 goto next;
 195                         goto recheck;
 196                 } else {
 197                         cachefiles_io_error_obj(
 198                                 object,
 199                                 "Readpage failed on backing file %lx",
 200                                 (unsigned long) monitor->back_page->flags);
 201                         error = -EIO;
 202                 }
 203
 204                 page_cache_release(monitor->back_page);
 205
 206                 fscache_end_io(op, monitor->netfs_page, error);
 207                 page_cache_release(monitor->netfs_page);
 208                 fscache_retrieval_complete(op, 1);
 209                 fscache_put_retrieval(op);
 210                 kfree(monitor);
 211
 212         next:
 213                 /* let the thread pool have some air occasionally */
 214                 max--;
 215                 if (max < 0 || need_resched()) {
 216                         if (!list_empty(&op->to_do))
 217                                 fscache_enqueue_retrieval(op);
 218                         _leave(" [maxed out]");
 219                         return;
 220                 }
 221
 222                 spin_lock_irq(&object->work_lock);
 223         }
 224
 225         spin_unlock_irq(&object->work_lock);
 226         _leave("");
 227 }
 228
 229 /*
 230  * read the corresponding page to the given set from the backing file
 231  * - an uncertain page is simply discarded, to be tried again another time
 232  */
 233 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 234                                             struct fscache_retrieval *op,
 235                                             struct page *netpage)
 236 {
 237         struct cachefiles_one_read *monitor;
 238         struct address_space *bmapping;
 239         struct page *newpage, *backpage;
 240         int ret;
 241
 242         _enter("");
 243
 244         _debug("read back %p{%lu,%d}",
 245                netpage, netpage->index, page_count(netpage));
 246
 247         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 248         if (!monitor)
 249                 goto nomem;
 250
 251         monitor->netfs_page = netpage;
 252         monitor->op = fscache_get_retrieval(op);
 253
 254         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 255
 256         /* attempt to get hold of the backing page */
 257         bmapping = d_backing_inode(object->backer)->i_mapping;
 258         newpage = NULL;
 259
 260         for (;;) {
 261                 backpage = find_get_page(bmapping, netpage->index);
 262                 if (backpage)
 263                         goto backing_page_already_present;
 264
 265                 if (!newpage) {
 266                         newpage = __page_cache_alloc(cachefiles_gfp |
 267                                                      __GFP_COLD);
 268                         if (!newpage)
 269                                 goto nomem_monitor;
 270                 }
 271
 272                 ret = add_to_page_cache_lru(newpage, bmapping,
 273                                             netpage->index, cachefiles_gfp);
 274                 if (ret == 0)
 275                         goto installed_new_backing_page;
 276                 if (ret != -EEXIST)
 277                         goto nomem_page;
 278         }
 279
 280         /* we've installed a new backing page, so now we need to start
 281          * it reading */
 282 installed_new_backing_page:
 283         _debug("- new %p", newpage);
 284
 285         backpage = newpage;
 286         newpage = NULL;
 287
 288 read_backing_page:
 289         ret = bmapping->a_ops->readpage(NULL, backpage);
 290         if (ret < 0)
 291                 goto read_error;
 292
 293         /* set the monitor to transfer the data across */
 294 monitor_backing_page:
 295         _debug("- monitor add");
 296
 297         /* install the monitor */
 298         page_cache_get(monitor->netfs_page);
 299         page_cache_get(backpage);
 300         monitor->back_page = backpage;
 301         monitor->monitor.private = backpage;
 302         add_page_wait_queue(backpage, &monitor->monitor);
 303         monitor = NULL;
 304
 305         /* but the page may have been read before the monitor was installed, so
 306          * the monitor may miss the event - so we have to ensure that we do get
 307          * one in such a case */
 308         if (trylock_page(backpage)) {
 309                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 310                 unlock_page(backpage);
 311         }
 312         goto success;
 313
 314         /* if the backing page is already present, it can be in one of
 315          * three states: read in progress, read failed or read okay */
 316 backing_page_already_present:
 317         _debug("- present");
 318
 319         if (newpage) {
 320                 page_cache_release(newpage);
 321                 newpage = NULL;
 322         }
 323
 324         if (PageError(backpage))
 325                 goto io_error;
 326
 327         if (PageUptodate(backpage))
 328                 goto backing_page_already_uptodate;
 329
 330         if (!trylock_page(backpage))
 331                 goto monitor_backing_page;
 332         _debug("read %p {%lx}", backpage, backpage->flags);
 333         goto read_backing_page;
 334
 335         /* the backing page is already up to date, attach the netfs
 336          * page to the pagecache and LRU and copy the data across */
 337 backing_page_already_uptodate:
 338         _debug("- uptodate");
 339
 340         fscache_mark_page_cached(op, netpage);
 341
 342         copy_highpage(netpage, backpage);
 343         fscache_end_io(op, netpage, 0);
 344         fscache_retrieval_complete(op, 1);
 345
 346 success:
 347         _debug("success");
 348         ret = 0;
 349
 350 out:
 351         if (backpage)
 352                 page_cache_release(backpage);
 353         if (monitor) {
 354                 fscache_put_retrieval(monitor->op);
 355                 kfree(monitor);
 356         }
 357         _leave(" = %d", ret);
 358         return ret;
 359
 360 read_error:
 361         _debug("read error %d", ret);
 362         if (ret == -ENOMEM) {
 363                 fscache_retrieval_complete(op, 1);
 364                 goto out;
 365         }
 366 io_error:
 367         cachefiles_io_error_obj(object, "Page read error on backing file");
 368         fscache_retrieval_complete(op, 1);
 369         ret = -ENOBUFS;
 370         goto out;
 371
 372 nomem_page:
 373         page_cache_release(newpage);
 374 nomem_monitor:
 375         fscache_put_retrieval(monitor->op);
 376         kfree(monitor);
 377 nomem:
 378         fscache_retrieval_complete(op, 1);
 379         _leave(" = -ENOMEM");
 380         return -ENOMEM;
 381 }
 382
 383 /*
 384  * read a page from the cache or allocate a block in which to store it
 385  * - cache withdrawal is prevented by the caller
 386  * - returns -EINTR if interrupted
 387  * - returns -ENOMEM if ran out of memory
 388  * - returns -ENOBUFS if no buffers can be made available
 389  * - returns -ENOBUFS if page is beyond EOF
 390  * - if the page is backed by a block in the cache:
 391  *   - a read will be started which will call the callback on completion
 392  *   - 0 will be returned
 393  * - else if the page is unbacked:
 394  *   - the metadata will be retained
 395  *   - -ENODATA will be returned
 396  */
 397 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 398                                   struct page *page,
 399                                   gfp_t gfp)
 400 {
 401         struct cachefiles_object *object;
 402         struct cachefiles_cache *cache;
 403         struct inode *inode;
 404         sector_t block0, block;
 405         unsigned shift;
 406         int ret;
 407
 408         object = container_of(op->op.object,
 409                               struct cachefiles_object, fscache);
 410         cache = container_of(object->fscache.cache,
 411                              struct cachefiles_cache, cache);
 412
 413         _enter("{%p},{%lx},,,", object, page->index);
 414
 415         if (!object->backer)
 416                 goto enobufs;
 417
 418         inode = d_backing_inode(object->backer);
 419         ASSERT(S_ISREG(inode->i_mode));
 420         ASSERT(inode->i_mapping->a_ops->bmap);
 421         ASSERT(inode->i_mapping->a_ops->readpages);
 422
 423         /* calculate the shift required to use bmap */
 424         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 425
 426         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 427         op->op.flags |= FSCACHE_OP_ASYNC;
 428         op->op.processor = cachefiles_read_copier;
 429
 430         /* we assume the absence or presence of the first block is a good
 431          * enough indication for the page as a whole
 432          * - TODO: don't use bmap() for this as it is _not_ actually good
 433          *   enough for this as it doesn't indicate errors, but it's all we've
 434          *   got for the moment
 435          */
 436         block0 = page->index;
 437         block0 <<= shift;
 438
 439         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 440         _debug("%llx -> %llx",
 441                (unsigned long long) block0,
 442                (unsigned long long) block);
 443
 444         if (block) {
 445                 /* submit the apparently valid page to the backing fs to be
 446                  * read from disk */
 447                 ret = cachefiles_read_backing_file_one(object, op, page);
 448         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 449                 /* there's space in the cache we can use */
 450                 fscache_mark_page_cached(op, page);
 451                 fscache_retrieval_complete(op, 1);
 452                 ret = -ENODATA;
 453         } else {
 454                 goto enobufs;
 455         }
 456
 457         _leave(" = %d", ret);
 458         return ret;
 459
 460 enobufs:
 461         fscache_retrieval_complete(op, 1);
 462         _leave(" = -ENOBUFS");
 463         return -ENOBUFS;
 464 }
 465
 466 /*
 467  * read the corresponding pages to the given set from the backing file
 468  * - any uncertain pages are simply discarded, to be tried again another time
 469  */
 470 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 471                                         struct fscache_retrieval *op,
 472                                         struct list_head *list)
 473 {
 474         struct cachefiles_one_read *monitor = NULL;
 475         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 476         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 477         int ret = 0;
 478
 479         _enter("");
 480
 481         list_for_each_entry_safe(netpage, _n, list, lru) {
 482                 list_del(&netpage->lru);
 483
 484                 _debug("read back %p{%lu,%d}",
 485                        netpage, netpage->index, page_count(netpage));
 486
 487                 if (!monitor) {
 488                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 489                         if (!monitor)
 490                                 goto nomem;
 491
 492                         monitor->op = fscache_get_retrieval(op);
 493                         init_waitqueue_func_entry(&monitor->monitor,
 494                                                   cachefiles_read_waiter);
 495                 }
 496
 497                 for (;;) {
 498                         backpage = find_get_page(bmapping, netpage->index);
 499                         if (backpage)
 500                                 goto backing_page_already_present;
 501
 502                         if (!newpage) {
 503                                 newpage = __page_cache_alloc(cachefiles_gfp |
 504                                                              __GFP_COLD);
 505                                 if (!newpage)
 506                                         goto nomem;
 507                         }
 508
 509                         ret = add_to_page_cache_lru(newpage, bmapping,
 510                                                     netpage->index,
 511                                                     cachefiles_gfp);
 512                         if (ret == 0)
 513                                 goto installed_new_backing_page;
 514                         if (ret != -EEXIST)
 515                                 goto nomem;
 516                 }
 517
 518                 /* we've installed a new backing page, so now we need
 519                  * to start it reading */
 520         installed_new_backing_page:
 521                 _debug("- new %p", newpage);
 522
 523                 backpage = newpage;
 524                 newpage = NULL;
 525
 526         reread_backing_page:
 527                 ret = bmapping->a_ops->readpage(NULL, backpage);
 528                 if (ret < 0)
 529                         goto read_error;
 530
 531                 /* add the netfs page to the pagecache and LRU, and set the
 532                  * monitor to transfer the data across */
 533         monitor_backing_page:
 534                 _debug("- monitor add");
 535
 536                 ret = add_to_page_cache_lru(netpage, op->mapping,
 537                                             netpage->index, cachefiles_gfp);
 538                 if (ret < 0) {
 539                         if (ret == -EEXIST) {
 540                                 page_cache_release(netpage);
 541                                 fscache_retrieval_complete(op, 1);
 542                                 continue;
 543                         }
 544                         goto nomem;
 545                 }
 546
 547                 /* install a monitor */
 548                 page_cache_get(netpage);
 549                 monitor->netfs_page = netpage;
 550
 551                 page_cache_get(backpage);
 552                 monitor->back_page = backpage;
 553                 monitor->monitor.private = backpage;
 554                 add_page_wait_queue(backpage, &monitor->monitor);
 555                 monitor = NULL;
 556
 557                 /* but the page may have been read before the monitor was
 558                  * installed, so the monitor may miss the event - so we have to
 559                  * ensure that we do get one in such a case */
 560                 if (trylock_page(backpage)) {
 561                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 562                         unlock_page(backpage);
 563                 }
 564
 565                 page_cache_release(backpage);
 566                 backpage = NULL;
 567
 568                 page_cache_release(netpage);
 569                 netpage = NULL;
 570                 continue;
 571
 572                 /* if the backing page is already present, it can be in one of
 573                  * three states: read in progress, read failed or read okay */
 574         backing_page_already_present:
 575                 _debug("- present %p", backpage);
 576
 577                 if (PageError(backpage))
 578                         goto io_error;
 579
 580                 if (PageUptodate(backpage))
 581                         goto backing_page_already_uptodate;
 582
 583                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 584
 585                 if (!trylock_page(backpage))
 586                         goto monitor_backing_page;
 587
 588                 if (PageError(backpage)) {
 589                         _debug("error %lx", backpage->flags);
 590                         unlock_page(backpage);
 591                         goto io_error;
 592                 }
 593
 594                 if (PageUptodate(backpage))
 595                         goto backing_page_already_uptodate_unlock;
 596
 597                 /* we've locked a page that's neither up to date nor erroneous,
 598                  * so we need to attempt to read it again */
 599                 goto reread_backing_page;
 600
 601                 /* the backing page is already up to date, attach the netfs
 602                  * page to the pagecache and LRU and copy the data across */
 603         backing_page_already_uptodate_unlock:
 604                 _debug("uptodate %lx", backpage->flags);
 605                 unlock_page(backpage);
 606         backing_page_already_uptodate:
 607                 _debug("- uptodate");
 608
 609                 ret = add_to_page_cache_lru(netpage, op->mapping,
 610                                             netpage->index, cachefiles_gfp);
 611                 if (ret < 0) {
 612                         if (ret == -EEXIST) {
 613                                 page_cache_release(netpage);
 614                                 fscache_retrieval_complete(op, 1);
 615                                 continue;
 616                         }
 617                         goto nomem;
 618                 }
 619
 620                 copy_highpage(netpage, backpage);
 621
 622                 page_cache_release(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                 page_cache_release(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                 page_cache_release(newpage);
 643         if (netpage)
 644                 page_cache_release(netpage);
 645         if (backpage)
 646                 page_cache_release(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                 page_cache_release(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, 0);
 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, 0);
 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 {
 965         struct cachefiles_object *object;
 966
 967         object = container_of(_object, struct cachefiles_object, fscache);
 968
 969         _enter("%p,{%lu}", object, page->index);
 970
 971         spin_unlock(&object->fscache.cookie->lock);
 972 }