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_entry_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->entry);
  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 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 discard:
 146         spin_lock_irq(&object->work_lock);
 147         list_del(&monitor->op_link);
 148         spin_unlock_irq(&object->work_lock);
 149         _leave(" = %d", ret);
 150         return ret;
 151 }
 152
 153 /*
 154  * copy data from backing pages to netfs pages to complete a read operation
 155  * - driven by FS-Cache's thread pool
 156  */
 157 static void cachefiles_read_copier(struct fscache_operation *_op)
 158 {
 159         struct cachefiles_one_read *monitor;
 160         struct cachefiles_object *object;
 161         struct fscache_retrieval *op;
 162         int error, max;
 163
 164         op = container_of(_op, struct fscache_retrieval, op);
 165         object = container_of(op->op.object,
 166                               struct cachefiles_object, fscache);
 167
 168         _enter("{ino=%lu}", d_backing_inode(object->backer)->i_ino);
 169
 170         max = 8;
 171         spin_lock_irq(&object->work_lock);
 172
 173         while (!list_empty(&op->to_do)) {
 174                 monitor = list_entry(op->to_do.next,
 175                                      struct cachefiles_one_read, op_link);
 176                 list_del(&monitor->op_link);
 177
 178                 spin_unlock_irq(&object->work_lock);
 179
 180                 _debug("- copy {%lu}", monitor->back_page->index);
 181
 182         recheck:
 183                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 184                              &object->fscache.cookie->flags)) {
 185                         error = -ESTALE;
 186                 } else if (PageUptodate(monitor->back_page)) {
 187                         copy_highpage(monitor->netfs_page, monitor->back_page);
 188                         fscache_mark_page_cached(monitor->op,
 189                                                  monitor->netfs_page);
 190                         error = 0;
 191                 } else if (!PageError(monitor->back_page)) {
 192                         /* the page has probably been truncated */
 193                         error = cachefiles_read_reissue(object, monitor);
 194                         if (error == -EINPROGRESS)
 195                                 goto next;
 196                         goto recheck;
 197                 } else {
 198                         cachefiles_io_error_obj(
 199                                 object,
 200                                 "Readpage failed on backing file %lx",
 201                                 (unsigned long) monitor->back_page->flags);
 202                         error = -EIO;
 203                 }
 204
 205                 put_page(monitor->back_page);
 206
 207                 fscache_end_io(op, monitor->netfs_page, error);
 208                 put_page(monitor->netfs_page);
 209                 fscache_retrieval_complete(op, 1);
 210                 fscache_put_retrieval(op);
 211                 kfree(monitor);
 212
 213         next:
 214                 /* let the thread pool have some air occasionally */
 215                 max--;
 216                 if (max < 0 || need_resched()) {
 217                         if (!list_empty(&op->to_do))
 218                                 fscache_enqueue_retrieval(op);
 219                         _leave(" [maxed out]");
 220                         return;
 221                 }
 222
 223                 spin_lock_irq(&object->work_lock);
 224         }
 225
 226         spin_unlock_irq(&object->work_lock);
 227         _leave("");
 228 }
 229
 230 /*
 231  * read the corresponding page to the given set from the backing file
 232  * - an uncertain page is simply discarded, to be tried again another time
 233  */
 234 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 235                                             struct fscache_retrieval *op,
 236                                             struct page *netpage)
 237 {
 238         struct cachefiles_one_read *monitor;
 239         struct address_space *bmapping;
 240         struct page *newpage, *backpage;
 241         int ret;
 242
 243         _enter("");
 244
 245         _debug("read back %p{%lu,%d}",
 246                netpage, netpage->index, page_count(netpage));
 247
 248         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 249         if (!monitor)
 250                 goto nomem;
 251
 252         monitor->netfs_page = netpage;
 253         monitor->op = fscache_get_retrieval(op);
 254
 255         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 256
 257         /* attempt to get hold of the backing page */
 258         bmapping = d_backing_inode(object->backer)->i_mapping;
 259         newpage = NULL;
 260
 261         for (;;) {
 262                 backpage = find_get_page(bmapping, netpage->index);
 263                 if (backpage)
 264                         goto backing_page_already_present;
 265
 266                 if (!newpage) {
 267                         newpage = __page_cache_alloc(cachefiles_gfp |
 268                                                      __GFP_COLD);
 269                         if (!newpage)
 270                                 goto nomem_monitor;
 271                 }
 272
 273                 ret = add_to_page_cache_lru(newpage, bmapping,
 274                                             netpage->index, cachefiles_gfp);
 275                 if (ret == 0)
 276                         goto installed_new_backing_page;
 277                 if (ret != -EEXIST)
 278                         goto nomem_page;
 279         }
 280
 281         /* we've installed a new backing page, so now we need to start
 282          * it reading */
 283 installed_new_backing_page:
 284         _debug("- new %p", newpage);
 285
 286         backpage = newpage;
 287         newpage = NULL;
 288
 289 read_backing_page:
 290         ret = bmapping->a_ops->readpage(NULL, backpage);
 291         if (ret < 0)
 292                 goto read_error;
 293
 294         /* set the monitor to transfer the data across */
 295 monitor_backing_page:
 296         _debug("- monitor add");
 297
 298         /* install the monitor */
 299         get_page(monitor->netfs_page);
 300         get_page(backpage);
 301         monitor->back_page = backpage;
 302         monitor->monitor.private = backpage;
 303         add_page_wait_queue(backpage, &monitor->monitor);
 304         monitor = NULL;
 305
 306         /* but the page may have been read before the monitor was installed, so
 307          * the monitor may miss the event - so we have to ensure that we do get
 308          * one in such a case */
 309         if (trylock_page(backpage)) {
 310                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 311                 unlock_page(backpage);
 312         }
 313         goto success;
 314
 315         /* if the backing page is already present, it can be in one of
 316          * three states: read in progress, read failed or read okay */
 317 backing_page_already_present:
 318         _debug("- present");
 319
 320         if (newpage) {
 321                 put_page(newpage);
 322                 newpage = NULL;
 323         }
 324
 325         if (PageError(backpage))
 326                 goto io_error;
 327
 328         if (PageUptodate(backpage))
 329                 goto backing_page_already_uptodate;
 330
 331         if (!trylock_page(backpage))
 332                 goto monitor_backing_page;
 333         _debug("read %p {%lx}", backpage, backpage->flags);
 334         goto read_backing_page;
 335
 336         /* the backing page is already up to date, attach the netfs
 337          * page to the pagecache and LRU and copy the data across */
 338 backing_page_already_uptodate:
 339         _debug("- uptodate");
 340
 341         fscache_mark_page_cached(op, netpage);
 342
 343         copy_highpage(netpage, backpage);
 344         fscache_end_io(op, netpage, 0);
 345         fscache_retrieval_complete(op, 1);
 346
 347 success:
 348         _debug("success");
 349         ret = 0;
 350
 351 out:
 352         if (backpage)
 353                 put_page(backpage);
 354         if (monitor) {
 355                 fscache_put_retrieval(monitor->op);
 356                 kfree(monitor);
 357         }
 358         _leave(" = %d", ret);
 359         return ret;
 360
 361 read_error:
 362         _debug("read error %d", ret);
 363         if (ret == -ENOMEM) {
 364                 fscache_retrieval_complete(op, 1);
 365                 goto out;
 366         }
 367 io_error:
 368         cachefiles_io_error_obj(object, "Page read error on backing file");
 369         fscache_retrieval_complete(op, 1);
 370         ret = -ENOBUFS;
 371         goto out;
 372
 373 nomem_page:
 374         put_page(newpage);
 375 nomem_monitor:
 376         fscache_put_retrieval(monitor->op);
 377         kfree(monitor);
 378 nomem:
 379         fscache_retrieval_complete(op, 1);
 380         _leave(" = -ENOMEM");
 381         return -ENOMEM;
 382 }
 383
 384 /*
 385  * read a page from the cache or allocate a block in which to store it
 386  * - cache withdrawal is prevented by the caller
 387  * - returns -EINTR if interrupted
 388  * - returns -ENOMEM if ran out of memory
 389  * - returns -ENOBUFS if no buffers can be made available
 390  * - returns -ENOBUFS if page is beyond EOF
 391  * - if the page is backed by a block in the cache:
 392  *   - a read will be started which will call the callback on completion
 393  *   - 0 will be returned
 394  * - else if the page is unbacked:
 395  *   - the metadata will be retained
 396  *   - -ENODATA will be returned
 397  */
 398 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 399                                   struct page *page,
 400                                   gfp_t gfp)
 401 {
 402         struct cachefiles_object *object;
 403         struct cachefiles_cache *cache;
 404         struct inode *inode;
 405         sector_t block0, block;
 406         unsigned shift;
 407         int ret;
 408
 409         object = container_of(op->op.object,
 410                               struct cachefiles_object, fscache);
 411         cache = container_of(object->fscache.cache,
 412                              struct cachefiles_cache, cache);
 413
 414         _enter("{%p},{%lx},,,", object, page->index);
 415
 416         if (!object->backer)
 417                 goto enobufs;
 418
 419         inode = d_backing_inode(object->backer);
 420         ASSERT(S_ISREG(inode->i_mode));
 421         ASSERT(inode->i_mapping->a_ops->bmap);
 422         ASSERT(inode->i_mapping->a_ops->readpages);
 423
 424         /* calculate the shift required to use bmap */
 425         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 426
 427         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 428         op->op.flags |= FSCACHE_OP_ASYNC;
 429         op->op.processor = cachefiles_read_copier;
 430
 431         /* we assume the absence or presence of the first block is a good
 432          * enough indication for the page as a whole
 433          * - TODO: don't use bmap() for this as it is _not_ actually good
 434          *   enough for this as it doesn't indicate errors, but it's all we've
 435          *   got for the moment
 436          */
 437         block0 = page->index;
 438         block0 <<= shift;
 439
 440         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 441         _debug("%llx -> %llx",
 442                (unsigned long long) block0,
 443                (unsigned long long) block);
 444
 445         if (block) {
 446                 /* submit the apparently valid page to the backing fs to be
 447                  * read from disk */
 448                 ret = cachefiles_read_backing_file_one(object, op, page);
 449         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 450                 /* there's space in the cache we can use */
 451                 fscache_mark_page_cached(op, page);
 452                 fscache_retrieval_complete(op, 1);
 453                 ret = -ENODATA;
 454         } else {
 455                 goto enobufs;
 456         }
 457
 458         _leave(" = %d", ret);
 459         return ret;
 460
 461 enobufs:
 462         fscache_retrieval_complete(op, 1);
 463         _leave(" = -ENOBUFS");
 464         return -ENOBUFS;
 465 }
 466
 467 /*
 468  * read the corresponding pages to the given set from the backing file
 469  * - any uncertain pages are simply discarded, to be tried again another time
 470  */
 471 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 472                                         struct fscache_retrieval *op,
 473                                         struct list_head *list)
 474 {
 475         struct cachefiles_one_read *monitor = NULL;
 476         struct address_space *bmapping = d_backing_inode(object->backer)->i_mapping;
 477         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 478         int ret = 0;
 479
 480         _enter("");
 481
 482         list_for_each_entry_safe(netpage, _n, list, lru) {
 483                 list_del(&netpage->lru);
 484
 485                 _debug("read back %p{%lu,%d}",
 486                        netpage, netpage->index, page_count(netpage));
 487
 488                 if (!monitor) {
 489                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 490                         if (!monitor)
 491                                 goto nomem;
 492
 493                         monitor->op = fscache_get_retrieval(op);
 494                         init_waitqueue_func_entry(&monitor->monitor,
 495                                                   cachefiles_read_waiter);
 496                 }
 497
 498                 for (;;) {
 499                         backpage = find_get_page(bmapping, netpage->index);
 500                         if (backpage)
 501                                 goto backing_page_already_present;
 502
 503                         if (!newpage) {
 504                                 newpage = __page_cache_alloc(cachefiles_gfp |
 505                                                              __GFP_COLD);
 506                                 if (!newpage)
 507                                         goto nomem;
 508                         }
 509
 510                         ret = add_to_page_cache_lru(newpage, bmapping,
 511                                                     netpage->index,
 512                                                     cachefiles_gfp);
 513                         if (ret == 0)
 514                                 goto installed_new_backing_page;
 515                         if (ret != -EEXIST)
 516                                 goto nomem;
 517                 }
 518
 519                 /* we've installed a new backing page, so now we need
 520                  * to start it reading */
 521         installed_new_backing_page:
 522                 _debug("- new %p", newpage);
 523
 524                 backpage = newpage;
 525                 newpage = NULL;
 526
 527         reread_backing_page:
 528                 ret = bmapping->a_ops->readpage(NULL, backpage);
 529                 if (ret < 0)
 530                         goto read_error;
 531
 532                 /* add the netfs page to the pagecache and LRU, and set the
 533                  * monitor to transfer the data across */
 534         monitor_backing_page:
 535                 _debug("- monitor add");
 536
 537                 ret = add_to_page_cache_lru(netpage, op->mapping,
 538                                             netpage->index, cachefiles_gfp);
 539                 if (ret < 0) {
 540                         if (ret == -EEXIST) {
 541                                 put_page(backpage);
 542                                 backpage = NULL;
 543                                 put_page(netpage);
 544                                 netpage = NULL;
 545                                 fscache_retrieval_complete(op, 1);
 546                                 continue;
 547                         }
 548                         goto nomem;
 549                 }
 550
 551                 /* install a monitor */
 552                 get_page(netpage);
 553                 monitor->netfs_page = netpage;
 554
 555                 get_page(backpage);
 556                 monitor->back_page = backpage;
 557                 monitor->monitor.private = backpage;
 558                 add_page_wait_queue(backpage, &monitor->monitor);
 559                 monitor = NULL;
 560
 561                 /* but the page may have been read before the monitor was
 562                  * installed, so the monitor may miss the event - so we have to
 563                  * ensure that we do get one in such a case */
 564                 if (trylock_page(backpage)) {
 565                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 566                         unlock_page(backpage);
 567                 }
 568
 569                 put_page(backpage);
 570                 backpage = NULL;
 571
 572                 put_page(netpage);
 573                 netpage = NULL;
 574                 continue;
 575
 576                 /* if the backing page is already present, it can be in one of
 577                  * three states: read in progress, read failed or read okay */
 578         backing_page_already_present:
 579                 _debug("- present %p", backpage);
 580
 581                 if (PageError(backpage))
 582                         goto io_error;
 583
 584                 if (PageUptodate(backpage))
 585                         goto backing_page_already_uptodate;
 586
 587                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 588
 589                 if (!trylock_page(backpage))
 590                         goto monitor_backing_page;
 591
 592                 if (PageError(backpage)) {
 593                         _debug("error %lx", backpage->flags);
 594                         unlock_page(backpage);
 595                         goto io_error;
 596                 }
 597
 598                 if (PageUptodate(backpage))
 599                         goto backing_page_already_uptodate_unlock;
 600
 601                 /* we've locked a page that's neither up to date nor erroneous,
 602                  * so we need to attempt to read it again */
 603                 goto reread_backing_page;
 604
 605                 /* the backing page is already up to date, attach the netfs
 606                  * page to the pagecache and LRU and copy the data across */
 607         backing_page_already_uptodate_unlock:
 608                 _debug("uptodate %lx", backpage->flags);
 609                 unlock_page(backpage);
 610         backing_page_already_uptodate:
 611                 _debug("- uptodate");
 612
 613                 ret = add_to_page_cache_lru(netpage, op->mapping,
 614                                             netpage->index, cachefiles_gfp);
 615                 if (ret < 0) {
 616                         if (ret == -EEXIST) {
 617                                 put_page(backpage);
 618                                 backpage = NULL;
 619                                 put_page(netpage);
 620                                 netpage = NULL;
 621                                 fscache_retrieval_complete(op, 1);
 622                                 continue;
 623                         }
 624                         goto nomem;
 625                 }
 626
 627                 copy_highpage(netpage, backpage);
 628
 629                 put_page(backpage);
 630                 backpage = NULL;
 631
 632                 fscache_mark_page_cached(op, netpage);
 633
 634                 /* the netpage is unlocked and marked up to date here */
 635                 fscache_end_io(op, netpage, 0);
 636                 put_page(netpage);
 637                 netpage = NULL;
 638                 fscache_retrieval_complete(op, 1);
 639                 continue;
 640         }
 641
 642         netpage = NULL;
 643
 644         _debug("out");
 645
 646 out:
 647         /* tidy up */
 648         if (newpage)
 649                 put_page(newpage);
 650         if (netpage)
 651                 put_page(netpage);
 652         if (backpage)
 653                 put_page(backpage);
 654         if (monitor) {
 655                 fscache_put_retrieval(op);
 656                 kfree(monitor);
 657         }
 658
 659         list_for_each_entry_safe(netpage, _n, list, lru) {
 660                 list_del(&netpage->lru);
 661                 put_page(netpage);
 662                 fscache_retrieval_complete(op, 1);
 663         }
 664
 665         _leave(" = %d", ret);
 666         return ret;
 667
 668 nomem:
 669         _debug("nomem");
 670         ret = -ENOMEM;
 671         goto record_page_complete;
 672
 673 read_error:
 674         _debug("read error %d", ret);
 675         if (ret == -ENOMEM)
 676                 goto record_page_complete;
 677 io_error:
 678         cachefiles_io_error_obj(object, "Page read error on backing file");
 679         ret = -ENOBUFS;
 680 record_page_complete:
 681         fscache_retrieval_complete(op, 1);
 682         goto out;
 683 }
 684
 685 /*
 686  * read a list of pages from the cache or allocate blocks in which to store
 687  * them
 688  */
 689 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 690                                    struct list_head *pages,
 691                                    unsigned *nr_pages,
 692                                    gfp_t gfp)
 693 {
 694         struct cachefiles_object *object;
 695         struct cachefiles_cache *cache;
 696         struct list_head backpages;
 697         struct pagevec pagevec;
 698         struct inode *inode;
 699         struct page *page, *_n;
 700         unsigned shift, nrbackpages;
 701         int ret, ret2, space;
 702
 703         object = container_of(op->op.object,
 704                               struct cachefiles_object, fscache);
 705         cache = container_of(object->fscache.cache,
 706                              struct cachefiles_cache, cache);
 707
 708         _enter("{OBJ%x,%d},,%d,,",
 709                object->fscache.debug_id, atomic_read(&op->op.usage),
 710                *nr_pages);
 711
 712         if (!object->backer)
 713                 goto all_enobufs;
 714
 715         space = 1;
 716         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 717                 space = 0;
 718
 719         inode = d_backing_inode(object->backer);
 720         ASSERT(S_ISREG(inode->i_mode));
 721         ASSERT(inode->i_mapping->a_ops->bmap);
 722         ASSERT(inode->i_mapping->a_ops->readpages);
 723
 724         /* calculate the shift required to use bmap */
 725         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 726
 727         pagevec_init(&pagevec, 0);
 728
 729         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 730         op->op.flags |= FSCACHE_OP_ASYNC;
 731         op->op.processor = cachefiles_read_copier;
 732
 733         INIT_LIST_HEAD(&backpages);
 734         nrbackpages = 0;
 735
 736         ret = space ? -ENODATA : -ENOBUFS;
 737         list_for_each_entry_safe(page, _n, pages, lru) {
 738                 sector_t block0, block;
 739
 740                 /* we assume the absence or presence of the first block is a
 741                  * good enough indication for the page as a whole
 742                  * - TODO: don't use bmap() for this as it is _not_ actually
 743                  *   good enough for this as it doesn't indicate errors, but
 744                  *   it's all we've got for the moment
 745                  */
 746                 block0 = page->index;
 747                 block0 <<= shift;
 748
 749                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 750                                                       block0);
 751                 _debug("%llx -> %llx",
 752                        (unsigned long long) block0,
 753                        (unsigned long long) block);
 754
 755                 if (block) {
 756                         /* we have data - add it to the list to give to the
 757                          * backing fs */
 758                         list_move(&page->lru, &backpages);
 759                         (*nr_pages)--;
 760                         nrbackpages++;
 761                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 762                         fscache_mark_pages_cached(op, &pagevec);
 763                         fscache_retrieval_complete(op, 1);
 764                         ret = -ENODATA;
 765                 } else {
 766                         fscache_retrieval_complete(op, 1);
 767                 }
 768         }
 769
 770         if (pagevec_count(&pagevec) > 0)
 771                 fscache_mark_pages_cached(op, &pagevec);
 772
 773         if (list_empty(pages))
 774                 ret = 0;
 775
 776         /* submit the apparently valid pages to the backing fs to be read from
 777          * disk */
 778         if (nrbackpages > 0) {
 779                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 780                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 781                         ret = ret2;
 782         }
 783
 784         _leave(" = %d [nr=%u%s]",
 785                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 786         return ret;
 787
 788 all_enobufs:
 789         fscache_retrieval_complete(op, *nr_pages);
 790         return -ENOBUFS;
 791 }
 792
 793 /*
 794  * allocate a block in the cache in which to store a page
 795  * - cache withdrawal is prevented by the caller
 796  * - returns -EINTR if interrupted
 797  * - returns -ENOMEM if ran out of memory
 798  * - returns -ENOBUFS if no buffers can be made available
 799  * - returns -ENOBUFS if page is beyond EOF
 800  * - otherwise:
 801  *   - the metadata will be retained
 802  *   - 0 will be returned
 803  */
 804 int cachefiles_allocate_page(struct fscache_retrieval *op,
 805                              struct page *page,
 806                              gfp_t gfp)
 807 {
 808         struct cachefiles_object *object;
 809         struct cachefiles_cache *cache;
 810         int ret;
 811
 812         object = container_of(op->op.object,
 813                               struct cachefiles_object, fscache);
 814         cache = container_of(object->fscache.cache,
 815                              struct cachefiles_cache, cache);
 816
 817         _enter("%p,{%lx},", object, page->index);
 818
 819         ret = cachefiles_has_space(cache, 0, 1);
 820         if (ret == 0)
 821                 fscache_mark_page_cached(op, page);
 822         else
 823                 ret = -ENOBUFS;
 824
 825         fscache_retrieval_complete(op, 1);
 826         _leave(" = %d", ret);
 827         return ret;
 828 }
 829
 830 /*
 831  * allocate blocks in the cache in which to store a set of pages
 832  * - cache withdrawal is prevented by the caller
 833  * - returns -EINTR if interrupted
 834  * - returns -ENOMEM if ran out of memory
 835  * - returns -ENOBUFS if some buffers couldn't be made available
 836  * - returns -ENOBUFS if some pages are beyond EOF
 837  * - otherwise:
 838  *   - -ENODATA will be returned
 839  * - metadata will be retained for any page marked
 840  */
 841 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 842                               struct list_head *pages,
 843                               unsigned *nr_pages,
 844                               gfp_t gfp)
 845 {
 846         struct cachefiles_object *object;
 847         struct cachefiles_cache *cache;
 848         struct pagevec pagevec;
 849         struct page *page;
 850         int ret;
 851
 852         object = container_of(op->op.object,
 853                               struct cachefiles_object, fscache);
 854         cache = container_of(object->fscache.cache,
 855                              struct cachefiles_cache, cache);
 856
 857         _enter("%p,,,%d,", object, *nr_pages);
 858
 859         ret = cachefiles_has_space(cache, 0, *nr_pages);
 860         if (ret == 0) {
 861                 pagevec_init(&pagevec, 0);
 862
 863                 list_for_each_entry(page, pages, lru) {
 864                         if (pagevec_add(&pagevec, page) == 0)
 865                                 fscache_mark_pages_cached(op, &pagevec);
 866                 }
 867
 868                 if (pagevec_count(&pagevec) > 0)
 869                         fscache_mark_pages_cached(op, &pagevec);
 870                 ret = -ENODATA;
 871         } else {
 872                 ret = -ENOBUFS;
 873         }
 874
 875         fscache_retrieval_complete(op, *nr_pages);
 876         _leave(" = %d", ret);
 877         return ret;
 878 }
 879
 880 /*
 881  * request a page be stored in the cache
 882  * - cache withdrawal is prevented by the caller
 883  * - this request may be ignored if there's no cache block available, in which
 884  *   case -ENOBUFS will be returned
 885  * - if the op is in progress, 0 will be returned
 886  */
 887 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 888 {
 889         struct cachefiles_object *object;
 890         struct cachefiles_cache *cache;
 891         struct file *file;
 892         struct path path;
 893         loff_t pos, eof;
 894         size_t len;
 895         void *data;
 896         int ret = -ENOBUFS;
 897
 898         ASSERT(op != NULL);
 899         ASSERT(page != NULL);
 900
 901         object = container_of(op->op.object,
 902                               struct cachefiles_object, fscache);
 903
 904         _enter("%p,%p{%lx},,,", object, page, page->index);
 905
 906         if (!object->backer) {
 907                 _leave(" = -ENOBUFS");
 908                 return -ENOBUFS;
 909         }
 910
 911         ASSERT(d_is_reg(object->backer));
 912
 913         cache = container_of(object->fscache.cache,
 914                              struct cachefiles_cache, cache);
 915
 916         pos = (loff_t)page->index << PAGE_SHIFT;
 917
 918         /* We mustn't write more data than we have, so we have to beware of a
 919          * partial page at EOF.
 920          */
 921         eof = object->fscache.store_limit_l;
 922         if (pos >= eof)
 923                 goto error;
 924
 925         /* write the page to the backing filesystem and let it store it in its
 926          * own time */
 927         path.mnt = cache->mnt;
 928         path.dentry = object->backer;
 929         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 930         if (IS_ERR(file)) {
 931                 ret = PTR_ERR(file);
 932                 goto error_2;
 933         }
 934
 935         len = PAGE_SIZE;
 936         if (eof & ~PAGE_MASK) {
 937                 if (eof - pos < PAGE_SIZE) {
 938                         _debug("cut short %llx to %llx",
 939                                pos, eof);
 940                         len = eof - pos;
 941                         ASSERTCMP(pos + len, ==, eof);
 942                 }
 943         }
 944
 945         data = kmap(page);
 946         ret = __kernel_write(file, data, len, &pos);
 947         kunmap(page);
 948         fput(file);
 949         if (ret != len)
 950                 goto error_eio;
 951
 952         _leave(" = 0");
 953         return 0;
 954
 955 error_eio:
 956         ret = -EIO;
 957 error_2:
 958         if (ret == -EIO)
 959                 cachefiles_io_error_obj(object,
 960                                         "Write page to backing file failed");
 961 error:
 962         _leave(" = -ENOBUFS [%d]", ret);
 963         return -ENOBUFS;
 964 }
 965
 966 /*
 967  * detach a backing block from a page
 968  * - cache withdrawal is prevented by the caller
 969  */
 970 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 971 {
 972         struct cachefiles_object *object;
 973
 974         object = container_of(_object, struct cachefiles_object, fscache);
 975
 976         _enter("%p,{%lu}", object, page->index);
 977
 978         spin_unlock(&object->fscache.cookie->lock);
 979 }