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 "internal.h"
  16
  17 /*
  18  * detect wake up events generated by the unlocking of pages in which we're
  19  * interested
  20  * - we use this to detect read completion of backing pages
  21  * - the caller holds the waitqueue lock
  22  */
  23 static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode,
  24                                   int sync, void *_key)
  25 {
  26         struct cachefiles_one_read *monitor =
  27                 container_of(wait, struct cachefiles_one_read, monitor);
  28         struct cachefiles_object *object;
  29         struct wait_bit_key *key = _key;
  30         struct page *page = wait->private;
  31
  32         ASSERT(key);
  33
  34         _enter("{%lu},%u,%d,{%p,%u}",
  35                monitor->netfs_page->index, mode, sync,
  36                key->flags, key->bit_nr);
  37
  38         if (key->flags != &page->flags ||
  39             key->bit_nr != PG_locked)
  40                 return 0;
  41
  42         _debug("--- monitor %p %lx ---", page, page->flags);
  43
  44         if (!PageUptodate(page) && !PageError(page)) {
  45                 /* unlocked, not uptodate and not erronous? */
  46                 _debug("page probably truncated");
  47         }
  48
  49         /* remove from the waitqueue */
  50         list_del(&wait->task_list);
  51
  52         /* move onto the action list and queue for FS-Cache thread pool */
  53         ASSERT(monitor->op);
  54
  55         object = container_of(monitor->op->op.object,
  56                               struct cachefiles_object, fscache);
  57
  58         spin_lock(&object->work_lock);
  59         list_add_tail(&monitor->op_link, &monitor->op->to_do);
  60         spin_unlock(&object->work_lock);
  61
  62         fscache_enqueue_retrieval(monitor->op);
  63         return 0;
  64 }
  65
  66 /*
  67  * handle a probably truncated page
  68  * - check to see if the page is still relevant and reissue the read if
  69  *   possible
  70  * - return -EIO on error, -ENODATA if the page is gone, -EINPROGRESS if we
  71  *   must wait again and 0 if successful
  72  */
  73 static int cachefiles_read_reissue(struct cachefiles_object *object,
  74                                    struct cachefiles_one_read *monitor)
  75 {
  76         struct address_space *bmapping = object->backer->d_inode->i_mapping;
  77         struct page *backpage = monitor->back_page, *backpage2;
  78         int ret;
  79
  80         _enter("{ino=%lx},{%lx,%lx}",
  81                object->backer->d_inode->i_ino,
  82                backpage->index, backpage->flags);
  83
  84         /* skip if the page was truncated away completely */
  85         if (backpage->mapping != bmapping) {
  86                 _leave(" = -ENODATA [mapping]");
  87                 return -ENODATA;
  88         }
  89
  90         backpage2 = find_get_page(bmapping, backpage->index);
  91         if (!backpage2) {
  92                 _leave(" = -ENODATA [gone]");
  93                 return -ENODATA;
  94         }
  95
  96         if (backpage != backpage2) {
  97                 put_page(backpage2);
  98                 _leave(" = -ENODATA [different]");
  99                 return -ENODATA;
 100         }
 101
 102         /* the page is still there and we already have a ref on it, so we don't
 103          * need a second */
 104         put_page(backpage2);
 105
 106         INIT_LIST_HEAD(&monitor->op_link);
 107         add_page_wait_queue(backpage, &monitor->monitor);
 108
 109         if (trylock_page(backpage)) {
 110                 ret = -EIO;
 111                 if (PageError(backpage))
 112                         goto unlock_discard;
 113                 ret = 0;
 114                 if (PageUptodate(backpage))
 115                         goto unlock_discard;
 116
 117                 _debug("reissue read");
 118                 ret = bmapping->a_ops->readpage(NULL, backpage);
 119                 if (ret < 0)
 120                         goto unlock_discard;
 121         }
 122
 123         /* but the page may have been read before the monitor was installed, so
 124          * the monitor may miss the event - so we have to ensure that we do get
 125          * one in such a case */
 126         if (trylock_page(backpage)) {
 127                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 128                 unlock_page(backpage);
 129         }
 130
 131         /* it'll reappear on the todo list */
 132         _leave(" = -EINPROGRESS");
 133         return -EINPROGRESS;
 134
 135 unlock_discard:
 136         unlock_page(backpage);
 137         spin_lock_irq(&object->work_lock);
 138         list_del(&monitor->op_link);
 139         spin_unlock_irq(&object->work_lock);
 140         _leave(" = %d", ret);
 141         return ret;
 142 }
 143
 144 /*
 145  * copy data from backing pages to netfs pages to complete a read operation
 146  * - driven by FS-Cache's thread pool
 147  */
 148 static void cachefiles_read_copier(struct fscache_operation *_op)
 149 {
 150         struct cachefiles_one_read *monitor;
 151         struct cachefiles_object *object;
 152         struct fscache_retrieval *op;
 153         struct pagevec pagevec;
 154         int error, max;
 155
 156         op = container_of(_op, struct fscache_retrieval, op);
 157         object = container_of(op->op.object,
 158                               struct cachefiles_object, fscache);
 159
 160         _enter("{ino=%lu}", object->backer->d_inode->i_ino);
 161
 162         pagevec_init(&pagevec, 0);
 163
 164         max = 8;
 165         spin_lock_irq(&object->work_lock);
 166
 167         while (!list_empty(&op->to_do)) {
 168                 monitor = list_entry(op->to_do.next,
 169                                      struct cachefiles_one_read, op_link);
 170                 list_del(&monitor->op_link);
 171
 172                 spin_unlock_irq(&object->work_lock);
 173
 174                 _debug("- copy {%lu}", monitor->back_page->index);
 175
 176         recheck:
 177                 if (test_bit(FSCACHE_COOKIE_INVALIDATING,
 178                              &object->fscache.cookie->flags)) {
 179                         error = -ESTALE;
 180                 } else if (PageUptodate(monitor->back_page)) {
 181                         copy_highpage(monitor->netfs_page, monitor->back_page);
 182                         fscache_mark_page_cached(monitor->op,
 183                                                  monitor->netfs_page);
 184                         error = 0;
 185                 } else if (!PageError(monitor->back_page)) {
 186                         /* the page has probably been truncated */
 187                         error = cachefiles_read_reissue(object, monitor);
 188                         if (error == -EINPROGRESS)
 189                                 goto next;
 190                         goto recheck;
 191                 } else {
 192                         cachefiles_io_error_obj(
 193                                 object,
 194                                 "Readpage failed on backing file %lx",
 195                                 (unsigned long) monitor->back_page->flags);
 196                         error = -EIO;
 197                 }
 198
 199                 page_cache_release(monitor->back_page);
 200
 201                 fscache_end_io(op, monitor->netfs_page, error);
 202                 page_cache_release(monitor->netfs_page);
 203                 fscache_retrieval_complete(op, 1);
 204                 fscache_put_retrieval(op);
 205                 kfree(monitor);
 206
 207         next:
 208                 /* let the thread pool have some air occasionally */
 209                 max--;
 210                 if (max < 0 || need_resched()) {
 211                         if (!list_empty(&op->to_do))
 212                                 fscache_enqueue_retrieval(op);
 213                         _leave(" [maxed out]");
 214                         return;
 215                 }
 216
 217                 spin_lock_irq(&object->work_lock);
 218         }
 219
 220         spin_unlock_irq(&object->work_lock);
 221         _leave("");
 222 }
 223
 224 /*
 225  * read the corresponding page to the given set from the backing file
 226  * - an uncertain page is simply discarded, to be tried again another time
 227  */
 228 static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
 229                                             struct fscache_retrieval *op,
 230                                             struct page *netpage,
 231                                             struct pagevec *pagevec)
 232 {
 233         struct cachefiles_one_read *monitor;
 234         struct address_space *bmapping;
 235         struct page *newpage, *backpage;
 236         int ret;
 237
 238         _enter("");
 239
 240         pagevec_reinit(pagevec);
 241
 242         _debug("read back %p{%lu,%d}",
 243                netpage, netpage->index, page_count(netpage));
 244
 245         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 246         if (!monitor)
 247                 goto nomem;
 248
 249         monitor->netfs_page = netpage;
 250         monitor->op = fscache_get_retrieval(op);
 251
 252         init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
 253
 254         /* attempt to get hold of the backing page */
 255         bmapping = object->backer->d_inode->i_mapping;
 256         newpage = NULL;
 257
 258         for (;;) {
 259                 backpage = find_get_page(bmapping, netpage->index);
 260                 if (backpage)
 261                         goto backing_page_already_present;
 262
 263                 if (!newpage) {
 264                         newpage = __page_cache_alloc(cachefiles_gfp |
 265                                                      __GFP_COLD);
 266                         if (!newpage)
 267                                 goto nomem_monitor;
 268                 }
 269
 270                 ret = add_to_page_cache(newpage, bmapping,
 271                                         netpage->index, cachefiles_gfp);
 272                 if (ret == 0)
 273                         goto installed_new_backing_page;
 274                 if (ret != -EEXIST)
 275                         goto nomem_page;
 276         }
 277
 278         /* we've installed a new backing page, so now we need to add it
 279          * to the LRU list and start it reading */
 280 installed_new_backing_page:
 281         _debug("- new %p", newpage);
 282
 283         backpage = newpage;
 284         newpage = NULL;
 285
 286         page_cache_get(backpage);
 287         pagevec_add(pagevec, backpage);
 288         __pagevec_lru_add_file(pagevec);
 289
 290 read_backing_page:
 291         ret = bmapping->a_ops->readpage(NULL, backpage);
 292         if (ret < 0)
 293                 goto read_error;
 294
 295         /* set the monitor to transfer the data across */
 296 monitor_backing_page:
 297         _debug("- monitor add");
 298
 299         /* install the monitor */
 300         page_cache_get(monitor->netfs_page);
 301         page_cache_get(backpage);
 302         monitor->back_page = backpage;
 303         monitor->monitor.private = backpage;
 304         add_page_wait_queue(backpage, &monitor->monitor);
 305         monitor = NULL;
 306
 307         /* but the page may have been read before the monitor was installed, so
 308          * the monitor may miss the event - so we have to ensure that we do get
 309          * one in such a case */
 310         if (trylock_page(backpage)) {
 311                 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
 312                 unlock_page(backpage);
 313         }
 314         goto success;
 315
 316         /* if the backing page is already present, it can be in one of
 317          * three states: read in progress, read failed or read okay */
 318 backing_page_already_present:
 319         _debug("- present");
 320
 321         if (newpage) {
 322                 page_cache_release(newpage);
 323                 newpage = NULL;
 324         }
 325
 326         if (PageError(backpage))
 327                 goto io_error;
 328
 329         if (PageUptodate(backpage))
 330                 goto backing_page_already_uptodate;
 331
 332         if (!trylock_page(backpage))
 333                 goto monitor_backing_page;
 334         _debug("read %p {%lx}", backpage, backpage->flags);
 335         goto read_backing_page;
 336
 337         /* the backing page is already up to date, attach the netfs
 338          * page to the pagecache and LRU and copy the data across */
 339 backing_page_already_uptodate:
 340         _debug("- uptodate");
 341
 342         fscache_mark_page_cached(op, netpage);
 343
 344         copy_highpage(netpage, backpage);
 345         fscache_end_io(op, netpage, 0);
 346         fscache_retrieval_complete(op, 1);
 347
 348 success:
 349         _debug("success");
 350         ret = 0;
 351
 352 out:
 353         if (backpage)
 354                 page_cache_release(backpage);
 355         if (monitor) {
 356                 fscache_put_retrieval(monitor->op);
 357                 kfree(monitor);
 358         }
 359         _leave(" = %d", ret);
 360         return ret;
 361
 362 read_error:
 363         _debug("read error %d", ret);
 364         if (ret == -ENOMEM) {
 365                 fscache_retrieval_complete(op, 1);
 366                 goto out;
 367         }
 368 io_error:
 369         cachefiles_io_error_obj(object, "Page read error on backing file");
 370         fscache_retrieval_complete(op, 1);
 371         ret = -ENOBUFS;
 372         goto out;
 373
 374 nomem_page:
 375         page_cache_release(newpage);
 376 nomem_monitor:
 377         fscache_put_retrieval(monitor->op);
 378         kfree(monitor);
 379 nomem:
 380         fscache_retrieval_complete(op, 1);
 381         _leave(" = -ENOMEM");
 382         return -ENOMEM;
 383 }
 384
 385 /*
 386  * read a page from the cache or allocate a block in which to store it
 387  * - cache withdrawal is prevented by the caller
 388  * - returns -EINTR if interrupted
 389  * - returns -ENOMEM if ran out of memory
 390  * - returns -ENOBUFS if no buffers can be made available
 391  * - returns -ENOBUFS if page is beyond EOF
 392  * - if the page is backed by a block in the cache:
 393  *   - a read will be started which will call the callback on completion
 394  *   - 0 will be returned
 395  * - else if the page is unbacked:
 396  *   - the metadata will be retained
 397  *   - -ENODATA will be returned
 398  */
 399 int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
 400                                   struct page *page,
 401                                   gfp_t gfp)
 402 {
 403         struct cachefiles_object *object;
 404         struct cachefiles_cache *cache;
 405         struct pagevec pagevec;
 406         struct inode *inode;
 407         sector_t block0, block;
 408         unsigned shift;
 409         int ret;
 410
 411         object = container_of(op->op.object,
 412                               struct cachefiles_object, fscache);
 413         cache = container_of(object->fscache.cache,
 414                              struct cachefiles_cache, cache);
 415
 416         _enter("{%p},{%lx},,,", object, page->index);
 417
 418         if (!object->backer)
 419                 goto enobufs;
 420
 421         inode = object->backer->d_inode;
 422         ASSERT(S_ISREG(inode->i_mode));
 423         ASSERT(inode->i_mapping->a_ops->bmap);
 424         ASSERT(inode->i_mapping->a_ops->readpages);
 425
 426         /* calculate the shift required to use bmap */
 427         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 428                 goto enobufs;
 429
 430         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 431
 432         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 433         op->op.flags |= FSCACHE_OP_ASYNC;
 434         op->op.processor = cachefiles_read_copier;
 435
 436         pagevec_init(&pagevec, 0);
 437
 438         /* we assume the absence or presence of the first block is a good
 439          * enough indication for the page as a whole
 440          * - TODO: don't use bmap() for this as it is _not_ actually good
 441          *   enough for this as it doesn't indicate errors, but it's all we've
 442          *   got for the moment
 443          */
 444         block0 = page->index;
 445         block0 <<= shift;
 446
 447         block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
 448         _debug("%llx -> %llx",
 449                (unsigned long long) block0,
 450                (unsigned long long) block);
 451
 452         if (block) {
 453                 /* submit the apparently valid page to the backing fs to be
 454                  * read from disk */
 455                 ret = cachefiles_read_backing_file_one(object, op, page,
 456                                                        &pagevec);
 457         } else if (cachefiles_has_space(cache, 0, 1) == 0) {
 458                 /* there's space in the cache we can use */
 459                 fscache_mark_page_cached(op, page);
 460                 fscache_retrieval_complete(op, 1);
 461                 ret = -ENODATA;
 462         } else {
 463                 goto enobufs;
 464         }
 465
 466         _leave(" = %d", ret);
 467         return ret;
 468
 469 enobufs:
 470         fscache_retrieval_complete(op, 1);
 471         _leave(" = -ENOBUFS");
 472         return -ENOBUFS;
 473 }
 474
 475 /*
 476  * read the corresponding pages to the given set from the backing file
 477  * - any uncertain pages are simply discarded, to be tried again another time
 478  */
 479 static int cachefiles_read_backing_file(struct cachefiles_object *object,
 480                                         struct fscache_retrieval *op,
 481                                         struct list_head *list)
 482 {
 483         struct cachefiles_one_read *monitor = NULL;
 484         struct address_space *bmapping = object->backer->d_inode->i_mapping;
 485         struct pagevec lru_pvec;
 486         struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
 487         int ret = 0;
 488
 489         _enter("");
 490
 491         pagevec_init(&lru_pvec, 0);
 492
 493         list_for_each_entry_safe(netpage, _n, list, lru) {
 494                 list_del(&netpage->lru);
 495
 496                 _debug("read back %p{%lu,%d}",
 497                        netpage, netpage->index, page_count(netpage));
 498
 499                 if (!monitor) {
 500                         monitor = kzalloc(sizeof(*monitor), cachefiles_gfp);
 501                         if (!monitor)
 502                                 goto nomem;
 503
 504                         monitor->op = fscache_get_retrieval(op);
 505                         init_waitqueue_func_entry(&monitor->monitor,
 506                                                   cachefiles_read_waiter);
 507                 }
 508
 509                 for (;;) {
 510                         backpage = find_get_page(bmapping, netpage->index);
 511                         if (backpage)
 512                                 goto backing_page_already_present;
 513
 514                         if (!newpage) {
 515                                 newpage = __page_cache_alloc(cachefiles_gfp |
 516                                                              __GFP_COLD);
 517                                 if (!newpage)
 518                                         goto nomem;
 519                         }
 520
 521                         ret = add_to_page_cache(newpage, bmapping,
 522                                                 netpage->index, cachefiles_gfp);
 523                         if (ret == 0)
 524                                 goto installed_new_backing_page;
 525                         if (ret != -EEXIST)
 526                                 goto nomem;
 527                 }
 528
 529                 /* we've installed a new backing page, so now we need to add it
 530                  * to the LRU list and start it reading */
 531         installed_new_backing_page:
 532                 _debug("- new %p", newpage);
 533
 534                 backpage = newpage;
 535                 newpage = NULL;
 536
 537                 page_cache_get(backpage);
 538                 if (!pagevec_add(&lru_pvec, backpage))
 539                         __pagevec_lru_add_file(&lru_pvec);
 540
 541         reread_backing_page:
 542                 ret = bmapping->a_ops->readpage(NULL, backpage);
 543                 if (ret < 0)
 544                         goto read_error;
 545
 546                 /* add the netfs page to the pagecache and LRU, and set the
 547                  * monitor to transfer the data across */
 548         monitor_backing_page:
 549                 _debug("- monitor add");
 550
 551                 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
 552                                         cachefiles_gfp);
 553                 if (ret < 0) {
 554                         if (ret == -EEXIST) {
 555                                 page_cache_release(netpage);
 556                                 fscache_retrieval_complete(op, 1);
 557                                 continue;
 558                         }
 559                         goto nomem;
 560                 }
 561
 562                 page_cache_get(netpage);
 563                 if (!pagevec_add(&lru_pvec, netpage))
 564                         __pagevec_lru_add_file(&lru_pvec);
 565
 566                 /* install a monitor */
 567                 page_cache_get(netpage);
 568                 monitor->netfs_page = netpage;
 569
 570                 page_cache_get(backpage);
 571                 monitor->back_page = backpage;
 572                 monitor->monitor.private = backpage;
 573                 add_page_wait_queue(backpage, &monitor->monitor);
 574                 monitor = NULL;
 575
 576                 /* but the page may have been read before the monitor was
 577                  * installed, so the monitor may miss the event - so we have to
 578                  * ensure that we do get one in such a case */
 579                 if (trylock_page(backpage)) {
 580                         _debug("2unlock %p {%lx}", backpage, backpage->flags);
 581                         unlock_page(backpage);
 582                 }
 583
 584                 page_cache_release(backpage);
 585                 backpage = NULL;
 586
 587                 page_cache_release(netpage);
 588                 netpage = NULL;
 589                 continue;
 590
 591                 /* if the backing page is already present, it can be in one of
 592                  * three states: read in progress, read failed or read okay */
 593         backing_page_already_present:
 594                 _debug("- present %p", backpage);
 595
 596                 if (PageError(backpage))
 597                         goto io_error;
 598
 599                 if (PageUptodate(backpage))
 600                         goto backing_page_already_uptodate;
 601
 602                 _debug("- not ready %p{%lx}", backpage, backpage->flags);
 603
 604                 if (!trylock_page(backpage))
 605                         goto monitor_backing_page;
 606
 607                 if (PageError(backpage)) {
 608                         _debug("error %lx", backpage->flags);
 609                         unlock_page(backpage);
 610                         goto io_error;
 611                 }
 612
 613                 if (PageUptodate(backpage))
 614                         goto backing_page_already_uptodate_unlock;
 615
 616                 /* we've locked a page that's neither up to date nor erroneous,
 617                  * so we need to attempt to read it again */
 618                 goto reread_backing_page;
 619
 620                 /* the backing page is already up to date, attach the netfs
 621                  * page to the pagecache and LRU and copy the data across */
 622         backing_page_already_uptodate_unlock:
 623                 _debug("uptodate %lx", backpage->flags);
 624                 unlock_page(backpage);
 625         backing_page_already_uptodate:
 626                 _debug("- uptodate");
 627
 628                 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
 629                                         cachefiles_gfp);
 630                 if (ret < 0) {
 631                         if (ret == -EEXIST) {
 632                                 page_cache_release(netpage);
 633                                 fscache_retrieval_complete(op, 1);
 634                                 continue;
 635                         }
 636                         goto nomem;
 637                 }
 638
 639                 copy_highpage(netpage, backpage);
 640
 641                 page_cache_release(backpage);
 642                 backpage = NULL;
 643
 644                 fscache_mark_page_cached(op, netpage);
 645
 646                 page_cache_get(netpage);
 647                 if (!pagevec_add(&lru_pvec, netpage))
 648                         __pagevec_lru_add_file(&lru_pvec);
 649
 650                 /* the netpage is unlocked and marked up to date here */
 651                 fscache_end_io(op, netpage, 0);
 652                 page_cache_release(netpage);
 653                 netpage = NULL;
 654                 fscache_retrieval_complete(op, 1);
 655                 continue;
 656         }
 657
 658         netpage = NULL;
 659
 660         _debug("out");
 661
 662 out:
 663         /* tidy up */
 664         pagevec_lru_add_file(&lru_pvec);
 665
 666         if (newpage)
 667                 page_cache_release(newpage);
 668         if (netpage)
 669                 page_cache_release(netpage);
 670         if (backpage)
 671                 page_cache_release(backpage);
 672         if (monitor) {
 673                 fscache_put_retrieval(op);
 674                 kfree(monitor);
 675         }
 676
 677         list_for_each_entry_safe(netpage, _n, list, lru) {
 678                 list_del(&netpage->lru);
 679                 page_cache_release(netpage);
 680                 fscache_retrieval_complete(op, 1);
 681         }
 682
 683         _leave(" = %d", ret);
 684         return ret;
 685
 686 nomem:
 687         _debug("nomem");
 688         ret = -ENOMEM;
 689         goto record_page_complete;
 690
 691 read_error:
 692         _debug("read error %d", ret);
 693         if (ret == -ENOMEM)
 694                 goto record_page_complete;
 695 io_error:
 696         cachefiles_io_error_obj(object, "Page read error on backing file");
 697         ret = -ENOBUFS;
 698 record_page_complete:
 699         fscache_retrieval_complete(op, 1);
 700         goto out;
 701 }
 702
 703 /*
 704  * read a list of pages from the cache or allocate blocks in which to store
 705  * them
 706  */
 707 int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
 708                                    struct list_head *pages,
 709                                    unsigned *nr_pages,
 710                                    gfp_t gfp)
 711 {
 712         struct cachefiles_object *object;
 713         struct cachefiles_cache *cache;
 714         struct list_head backpages;
 715         struct pagevec pagevec;
 716         struct inode *inode;
 717         struct page *page, *_n;
 718         unsigned shift, nrbackpages;
 719         int ret, ret2, space;
 720
 721         object = container_of(op->op.object,
 722                               struct cachefiles_object, fscache);
 723         cache = container_of(object->fscache.cache,
 724                              struct cachefiles_cache, cache);
 725
 726         _enter("{OBJ%x,%d},,%d,,",
 727                object->fscache.debug_id, atomic_read(&op->op.usage),
 728                *nr_pages);
 729
 730         if (!object->backer)
 731                 goto all_enobufs;
 732
 733         space = 1;
 734         if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
 735                 space = 0;
 736
 737         inode = object->backer->d_inode;
 738         ASSERT(S_ISREG(inode->i_mode));
 739         ASSERT(inode->i_mapping->a_ops->bmap);
 740         ASSERT(inode->i_mapping->a_ops->readpages);
 741
 742         /* calculate the shift required to use bmap */
 743         if (inode->i_sb->s_blocksize > PAGE_SIZE)
 744                 goto all_enobufs;
 745
 746         shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
 747
 748         pagevec_init(&pagevec, 0);
 749
 750         op->op.flags &= FSCACHE_OP_KEEP_FLAGS;
 751         op->op.flags |= FSCACHE_OP_ASYNC;
 752         op->op.processor = cachefiles_read_copier;
 753
 754         INIT_LIST_HEAD(&backpages);
 755         nrbackpages = 0;
 756
 757         ret = space ? -ENODATA : -ENOBUFS;
 758         list_for_each_entry_safe(page, _n, pages, lru) {
 759                 sector_t block0, block;
 760
 761                 /* we assume the absence or presence of the first block is a
 762                  * good enough indication for the page as a whole
 763                  * - TODO: don't use bmap() for this as it is _not_ actually
 764                  *   good enough for this as it doesn't indicate errors, but
 765                  *   it's all we've got for the moment
 766                  */
 767                 block0 = page->index;
 768                 block0 <<= shift;
 769
 770                 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
 771                                                       block0);
 772                 _debug("%llx -> %llx",
 773                        (unsigned long long) block0,
 774                        (unsigned long long) block);
 775
 776                 if (block) {
 777                         /* we have data - add it to the list to give to the
 778                          * backing fs */
 779                         list_move(&page->lru, &backpages);
 780                         (*nr_pages)--;
 781                         nrbackpages++;
 782                 } else if (space && pagevec_add(&pagevec, page) == 0) {
 783                         fscache_mark_pages_cached(op, &pagevec);
 784                         fscache_retrieval_complete(op, 1);
 785                         ret = -ENODATA;
 786                 } else {
 787                         fscache_retrieval_complete(op, 1);
 788                 }
 789         }
 790
 791         if (pagevec_count(&pagevec) > 0)
 792                 fscache_mark_pages_cached(op, &pagevec);
 793
 794         if (list_empty(pages))
 795                 ret = 0;
 796
 797         /* submit the apparently valid pages to the backing fs to be read from
 798          * disk */
 799         if (nrbackpages > 0) {
 800                 ret2 = cachefiles_read_backing_file(object, op, &backpages);
 801                 if (ret2 == -ENOMEM || ret2 == -EINTR)
 802                         ret = ret2;
 803         }
 804
 805         _leave(" = %d [nr=%u%s]",
 806                ret, *nr_pages, list_empty(pages) ? " empty" : "");
 807         return ret;
 808
 809 all_enobufs:
 810         fscache_retrieval_complete(op, *nr_pages);
 811         return -ENOBUFS;
 812 }
 813
 814 /*
 815  * allocate a block in the cache in which to store a page
 816  * - cache withdrawal is prevented by the caller
 817  * - returns -EINTR if interrupted
 818  * - returns -ENOMEM if ran out of memory
 819  * - returns -ENOBUFS if no buffers can be made available
 820  * - returns -ENOBUFS if page is beyond EOF
 821  * - otherwise:
 822  *   - the metadata will be retained
 823  *   - 0 will be returned
 824  */
 825 int cachefiles_allocate_page(struct fscache_retrieval *op,
 826                              struct page *page,
 827                              gfp_t gfp)
 828 {
 829         struct cachefiles_object *object;
 830         struct cachefiles_cache *cache;
 831         int ret;
 832
 833         object = container_of(op->op.object,
 834                               struct cachefiles_object, fscache);
 835         cache = container_of(object->fscache.cache,
 836                              struct cachefiles_cache, cache);
 837
 838         _enter("%p,{%lx},", object, page->index);
 839
 840         ret = cachefiles_has_space(cache, 0, 1);
 841         if (ret == 0)
 842                 fscache_mark_page_cached(op, page);
 843         else
 844                 ret = -ENOBUFS;
 845
 846         fscache_retrieval_complete(op, 1);
 847         _leave(" = %d", ret);
 848         return ret;
 849 }
 850
 851 /*
 852  * allocate blocks in the cache in which to store a set of pages
 853  * - cache withdrawal is prevented by the caller
 854  * - returns -EINTR if interrupted
 855  * - returns -ENOMEM if ran out of memory
 856  * - returns -ENOBUFS if some buffers couldn't be made available
 857  * - returns -ENOBUFS if some pages are beyond EOF
 858  * - otherwise:
 859  *   - -ENODATA will be returned
 860  * - metadata will be retained for any page marked
 861  */
 862 int cachefiles_allocate_pages(struct fscache_retrieval *op,
 863                               struct list_head *pages,
 864                               unsigned *nr_pages,
 865                               gfp_t gfp)
 866 {
 867         struct cachefiles_object *object;
 868         struct cachefiles_cache *cache;
 869         struct pagevec pagevec;
 870         struct page *page;
 871         int ret;
 872
 873         object = container_of(op->op.object,
 874                               struct cachefiles_object, fscache);
 875         cache = container_of(object->fscache.cache,
 876                              struct cachefiles_cache, cache);
 877
 878         _enter("%p,,,%d,", object, *nr_pages);
 879
 880         ret = cachefiles_has_space(cache, 0, *nr_pages);
 881         if (ret == 0) {
 882                 pagevec_init(&pagevec, 0);
 883
 884                 list_for_each_entry(page, pages, lru) {
 885                         if (pagevec_add(&pagevec, page) == 0)
 886                                 fscache_mark_pages_cached(op, &pagevec);
 887                 }
 888
 889                 if (pagevec_count(&pagevec) > 0)
 890                         fscache_mark_pages_cached(op, &pagevec);
 891                 ret = -ENODATA;
 892         } else {
 893                 ret = -ENOBUFS;
 894         }
 895
 896         fscache_retrieval_complete(op, *nr_pages);
 897         _leave(" = %d", ret);
 898         return ret;
 899 }
 900
 901 /*
 902  * request a page be stored in the cache
 903  * - cache withdrawal is prevented by the caller
 904  * - this request may be ignored if there's no cache block available, in which
 905  *   case -ENOBUFS will be returned
 906  * - if the op is in progress, 0 will be returned
 907  */
 908 int cachefiles_write_page(struct fscache_storage *op, struct page *page)
 909 {
 910         struct cachefiles_object *object;
 911         struct cachefiles_cache *cache;
 912         mm_segment_t old_fs;
 913         struct file *file;
 914         struct path path;
 915         loff_t pos, eof;
 916         size_t len;
 917         void *data;
 918         int ret;
 919
 920         ASSERT(op != NULL);
 921         ASSERT(page != NULL);
 922
 923         object = container_of(op->op.object,
 924                               struct cachefiles_object, fscache);
 925
 926         _enter("%p,%p{%lx},,,", object, page, page->index);
 927
 928         if (!object->backer) {
 929                 _leave(" = -ENOBUFS");
 930                 return -ENOBUFS;
 931         }
 932
 933         ASSERT(S_ISREG(object->backer->d_inode->i_mode));
 934
 935         cache = container_of(object->fscache.cache,
 936                              struct cachefiles_cache, cache);
 937
 938         /* write the page to the backing filesystem and let it store it in its
 939          * own time */
 940         path.mnt = cache->mnt;
 941         path.dentry = object->backer;
 942         file = dentry_open(&path, O_RDWR | O_LARGEFILE, cache->cache_cred);
 943         if (IS_ERR(file)) {
 944                 ret = PTR_ERR(file);
 945         } else {
 946                 ret = -EIO;
 947                 if (file->f_op->write) {
 948                         pos = (loff_t) page->index << PAGE_SHIFT;
 949
 950                         /* we mustn't write more data than we have, so we have
 951                          * to beware of a partial page at EOF */
 952                         eof = object->fscache.store_limit_l;
 953                         len = PAGE_SIZE;
 954                         if (eof & ~PAGE_MASK) {
 955                                 ASSERTCMP(pos, <, eof);
 956                                 if (eof - pos < PAGE_SIZE) {
 957                                         _debug("cut short %llx to %llx",
 958                                                pos, eof);
 959                                         len = eof - pos;
 960                                         ASSERTCMP(pos + len, ==, eof);
 961                                 }
 962                         }
 963
 964                         data = kmap(page);
 965                         old_fs = get_fs();
 966                         set_fs(KERNEL_DS);
 967                         ret = file->f_op->write(
 968                                 file, (const void __user *) data, len, &pos);
 969                         set_fs(old_fs);
 970                         kunmap(page);
 971                         if (ret != len)
 972                                 ret = -EIO;
 973                 }
 974                 fput(file);
 975         }
 976
 977         if (ret < 0) {
 978                 if (ret == -EIO)
 979                         cachefiles_io_error_obj(
 980                                 object, "Write page to backing file failed");
 981                 ret = -ENOBUFS;
 982         }
 983
 984         _leave(" = %d", ret);
 985         return ret;
 986 }
 987
 988 /*
 989  * detach a backing block from a page
 990  * - cache withdrawal is prevented by the caller
 991  */
 992 void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
 993 {
 994         struct cachefiles_object *object;
 995         struct cachefiles_cache *cache;
 996
 997         object = container_of(_object, struct cachefiles_object, fscache);
 998         cache = container_of(object->fscache.cache,
 999                              struct cachefiles_cache, cache);
1000
1001         _enter("%p,{%lu}", object, page->index);
1002
1003         spin_unlock(&object->fscache.cookie->lock);
1004 }