mm/truncate.c

   1 /*
   2  * mm/truncate.c - code for taking down pages from address_spaces
   3  *
   4  * Copyright (C) 2002, Linus Torvalds
   5  *
   6  * 10Sep2002    Andrew Morton
   7  *              Initial version.
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/backing-dev.h>
  12 #include <linux/dax.h>
  13 #include <linux/gfp.h>
  14 #include <linux/mm.h>
  15 #include <linux/swap.h>
  16 #include <linux/export.h>
  17 #include <linux/pagemap.h>
  18 #include <linux/highmem.h>
  19 #include <linux/pagevec.h>
  20 #include <linux/task_io_accounting_ops.h>
  21 #include <linux/buffer_head.h>  /* grr. try_to_release_page,
  22                                    do_invalidatepage */
  23 #include <linux/cleancache.h>
  24 #include <linux/rmap.h>
  25 #include "internal.h"
  26
  27 static void clear_exceptional_entry(struct address_space *mapping,
  28                                     pgoff_t index, void *entry)
  29 {
  30         struct radix_tree_node *node;
  31         void **slot;
  32
  33         /* Handled by shmem itself */
  34         if (shmem_mapping(mapping))
  35                 return;
  36
  37         if (dax_mapping(mapping)) {
  38                 dax_delete_mapping_entry(mapping, index);
  39                 return;
  40         }
  41         spin_lock_irq(&mapping->tree_lock);
  42         /*
  43          * Regular page slots are stabilized by the page lock even
  44          * without the tree itself locked.  These unlocked entries
  45          * need verification under the tree lock.
  46          */
  47         if (!__radix_tree_lookup(&mapping->page_tree, index, &node,
  48                                 &slot))
  49                 goto unlock;
  50         if (*slot != entry)
  51                 goto unlock;
  52         radix_tree_replace_slot(slot, NULL);
  53         mapping->nrexceptional--;
  54         if (!node)
  55                 goto unlock;
  56         workingset_node_shadows_dec(node);
  57         /*
  58          * Don't track node without shadow entries.
  59          *
  60          * Avoid acquiring the list_lru lock if already untracked.
  61          * The list_empty() test is safe as node->private_list is
  62          * protected by mapping->tree_lock.
  63          */
  64         if (!workingset_node_shadows(node) &&
  65             !list_empty(&node->private_list))
  66                 list_lru_del(&workingset_shadow_nodes,
  67                                 &node->private_list);
  68         __radix_tree_delete_node(&mapping->page_tree, node);
  69 unlock:
  70         spin_unlock_irq(&mapping->tree_lock);
  71 }
  72
  73 /**
  74  * do_invalidatepage - invalidate part or all of a page
  75  * @page: the page which is affected
  76  * @offset: start of the range to invalidate
  77  * @length: length of the range to invalidate
  78  *
  79  * do_invalidatepage() is called when all or part of the page has become
  80  * invalidated by a truncate operation.
  81  *
  82  * do_invalidatepage() does not have to release all buffers, but it must
  83  * ensure that no dirty buffer is left outside @offset and that no I/O
  84  * is underway against any of the blocks which are outside the truncation
  85  * point.  Because the caller is about to free (and possibly reuse) those
  86  * blocks on-disk.
  87  */
  88 void do_invalidatepage(struct page *page, unsigned int offset,
  89                        unsigned int length)
  90 {
  91         void (*invalidatepage)(struct page *, unsigned int, unsigned int);
  92
  93         invalidatepage = page->mapping->a_ops->invalidatepage;
  94 #ifdef CONFIG_BLOCK
  95         if (!invalidatepage)
  96                 invalidatepage = block_invalidatepage;
  97 #endif
  98         if (invalidatepage)
  99                 (*invalidatepage)(page, offset, length);
 100 }
 101
 102 /*
 103  * If truncate cannot remove the fs-private metadata from the page, the page
 104  * becomes orphaned.  It will be left on the LRU and may even be mapped into
 105  * user pagetables if we're racing with filemap_fault().
 106  *
 107  * We need to bale out if page->mapping is no longer equal to the original
 108  * mapping.  This happens a) when the VM reclaimed the page while we waited on
 109  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
 110  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
 111  */
 112 static int
 113 truncate_complete_page(struct address_space *mapping, struct page *page)
 114 {
 115         if (page->mapping != mapping)
 116                 return -EIO;
 117
 118         if (page_has_private(page))
 119                 do_invalidatepage(page, 0, PAGE_SIZE);
 120
 121         /*
 122          * Some filesystems seem to re-dirty the page even after
 123          * the VM has canceled the dirty bit (eg ext3 journaling).
 124          * Hence dirty accounting check is placed after invalidation.
 125          */
 126         cancel_dirty_page(page);
 127         ClearPageMappedToDisk(page);
 128         delete_from_page_cache(page);
 129         return 0;
 130 }
 131
 132 /*
 133  * This is for invalidate_mapping_pages().  That function can be called at
 134  * any time, and is not supposed to throw away dirty pages.  But pages can
 135  * be marked dirty at any time too, so use remove_mapping which safely
 136  * discards clean, unused pages.
 137  *
 138  * Returns non-zero if the page was successfully invalidated.
 139  */
 140 static int
 141 invalidate_complete_page(struct address_space *mapping, struct page *page)
 142 {
 143         int ret;
 144
 145         if (page->mapping != mapping)
 146                 return 0;
 147
 148         if (page_has_private(page) && !try_to_release_page(page, 0))
 149                 return 0;
 150
 151         ret = remove_mapping(mapping, page);
 152
 153         return ret;
 154 }
 155
 156 int truncate_inode_page(struct address_space *mapping, struct page *page)
 157 {
 158         if (page_mapped(page)) {
 159                 unmap_mapping_range(mapping,
 160                                    (loff_t)page->index << PAGE_SHIFT,
 161                                    PAGE_SIZE, 0);
 162         }
 163         return truncate_complete_page(mapping, page);
 164 }
 165
 166 /*
 167  * Used to get rid of pages on hardware memory corruption.
 168  */
 169 int generic_error_remove_page(struct address_space *mapping, struct page *page)
 170 {
 171         if (!mapping)
 172                 return -EINVAL;
 173         /*
 174          * Only punch for normal data pages for now.
 175          * Handling other types like directories would need more auditing.
 176          */
 177         if (!S_ISREG(mapping->host->i_mode))
 178                 return -EIO;
 179         return truncate_inode_page(mapping, page);
 180 }
 181 EXPORT_SYMBOL(generic_error_remove_page);
 182
 183 /*
 184  * Safely invalidate one page from its pagecache mapping.
 185  * It only drops clean, unused pages. The page must be locked.
 186  *
 187  * Returns 1 if the page is successfully invalidated, otherwise 0.
 188  */
 189 int invalidate_inode_page(struct page *page)
 190 {
 191         struct address_space *mapping = page_mapping(page);
 192         if (!mapping)
 193                 return 0;
 194         if (PageDirty(page) || PageWriteback(page))
 195                 return 0;
 196         if (page_mapped(page))
 197                 return 0;
 198         return invalidate_complete_page(mapping, page);
 199 }
 200
 201 /**
 202  * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
 203  * @mapping: mapping to truncate
 204  * @lstart: offset from which to truncate
 205  * @lend: offset to which to truncate (inclusive)
 206  *
 207  * Truncate the page cache, removing the pages that are between
 208  * specified offsets (and zeroing out partial pages
 209  * if lstart or lend + 1 is not page aligned).
 210  *
 211  * Truncate takes two passes - the first pass is nonblocking.  It will not
 212  * block on page locks and it will not block on writeback.  The second pass
 213  * will wait.  This is to prevent as much IO as possible in the affected region.
 214  * The first pass will remove most pages, so the search cost of the second pass
 215  * is low.
 216  *
 217  * We pass down the cache-hot hint to the page freeing code.  Even if the
 218  * mapping is large, it is probably the case that the final pages are the most
 219  * recently touched, and freeing happens in ascending file offset order.
 220  *
 221  * Note that since ->invalidatepage() accepts range to invalidate
 222  * truncate_inode_pages_range is able to handle cases where lend + 1 is not
 223  * page aligned properly.
 224  */
 225 void truncate_inode_pages_range(struct address_space *mapping,
 226                                 loff_t lstart, loff_t lend)
 227 {
 228         pgoff_t         start;          /* inclusive */
 229         pgoff_t         end;            /* exclusive */
 230         unsigned int    partial_start;  /* inclusive */
 231         unsigned int    partial_end;    /* exclusive */
 232         struct pagevec  pvec;
 233         pgoff_t         indices[PAGEVEC_SIZE];
 234         pgoff_t         index;
 235         int             i;
 236
 237         cleancache_invalidate_inode(mapping);
 238         if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
 239                 return;
 240
 241         /* Offsets within partial pages */
 242         partial_start = lstart & (PAGE_SIZE - 1);
 243         partial_end = (lend + 1) & (PAGE_SIZE - 1);
 244
 245         /*
 246          * 'start' and 'end' always covers the range of pages to be fully
 247          * truncated. Partial pages are covered with 'partial_start' at the
 248          * start of the range and 'partial_end' at the end of the range.
 249          * Note that 'end' is exclusive while 'lend' is inclusive.
 250          */
 251         start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
 252         if (lend == -1)
 253                 /*
 254                  * lend == -1 indicates end-of-file so we have to set 'end'
 255                  * to the highest possible pgoff_t and since the type is
 256                  * unsigned we're using -1.
 257                  */
 258                 end = -1;
 259         else
 260                 end = (lend + 1) >> PAGE_SHIFT;
 261
 262         pagevec_init(&pvec, 0);
 263         index = start;
 264         while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
 265                         min(end - index, (pgoff_t)PAGEVEC_SIZE),
 266                         indices)) {
 267                 for (i = 0; i < pagevec_count(&pvec); i++) {
 268                         struct page *page = pvec.pages[i];
 269
 270                         /* We rely upon deletion not changing page->index */
 271                         index = indices[i];
 272                         if (index >= end)
 273                                 break;
 274
 275                         if (radix_tree_exceptional_entry(page)) {
 276                                 clear_exceptional_entry(mapping, index, page);
 277                                 continue;
 278                         }
 279
 280                         if (!trylock_page(page))
 281                                 continue;
 282                         WARN_ON(page->index != index);
 283                         if (PageWriteback(page)) {
 284                                 unlock_page(page);
 285                                 continue;
 286                         }
 287                         truncate_inode_page(mapping, page);
 288                         unlock_page(page);
 289                 }
 290                 pagevec_remove_exceptionals(&pvec);
 291                 pagevec_release(&pvec);
 292                 cond_resched();
 293                 index++;
 294         }
 295
 296         if (partial_start) {
 297                 struct page *page = find_lock_page(mapping, start - 1);
 298                 if (page) {
 299                         unsigned int top = PAGE_SIZE;
 300                         if (start > end) {
 301                                 /* Truncation within a single page */
 302                                 top = partial_end;
 303                                 partial_end = 0;
 304                         }
 305                         wait_on_page_writeback(page);
 306                         zero_user_segment(page, partial_start, top);
 307                         cleancache_invalidate_page(mapping, page);
 308                         if (page_has_private(page))
 309                                 do_invalidatepage(page, partial_start,
 310                                                   top - partial_start);
 311                         unlock_page(page);
 312                         put_page(page);
 313                 }
 314         }
 315         if (partial_end) {
 316                 struct page *page = find_lock_page(mapping, end);
 317                 if (page) {
 318                         wait_on_page_writeback(page);
 319                         zero_user_segment(page, 0, partial_end);
 320                         cleancache_invalidate_page(mapping, page);
 321                         if (page_has_private(page))
 322                                 do_invalidatepage(page, 0,
 323                                                   partial_end);
 324                         unlock_page(page);
 325                         put_page(page);
 326                 }
 327         }
 328         /*
 329          * If the truncation happened within a single page no pages
 330          * will be released, just zeroed, so we can bail out now.
 331          */
 332         if (start >= end)
 333                 return;
 334
 335         index = start;
 336         for ( ; ; ) {
 337                 cond_resched();
 338                 if (!pagevec_lookup_entries(&pvec, mapping, index,
 339                         min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
 340                         /* If all gone from start onwards, we're done */
 341                         if (index == start)
 342                                 break;
 343                         /* Otherwise restart to make sure all gone */
 344                         index = start;
 345                         continue;
 346                 }
 347                 if (index == start && indices[0] >= end) {
 348                         /* All gone out of hole to be punched, we're done */
 349                         pagevec_remove_exceptionals(&pvec);
 350                         pagevec_release(&pvec);
 351                         break;
 352                 }
 353                 for (i = 0; i < pagevec_count(&pvec); i++) {
 354                         struct page *page = pvec.pages[i];
 355
 356                         /* We rely upon deletion not changing page->index */
 357                         index = indices[i];
 358                         if (index >= end) {
 359                                 /* Restart punch to make sure all gone */
 360                                 index = start - 1;
 361                                 break;
 362                         }
 363
 364                         if (radix_tree_exceptional_entry(page)) {
 365                                 clear_exceptional_entry(mapping, index, page);
 366                                 continue;
 367                         }
 368
 369                         lock_page(page);
 370                         WARN_ON(page->index != index);
 371                         wait_on_page_writeback(page);
 372                         truncate_inode_page(mapping, page);
 373                         unlock_page(page);
 374                 }
 375                 pagevec_remove_exceptionals(&pvec);
 376                 pagevec_release(&pvec);
 377                 index++;
 378         }
 379         cleancache_invalidate_inode(mapping);
 380 }
 381 EXPORT_SYMBOL(truncate_inode_pages_range);
 382
 383 /**
 384  * truncate_inode_pages - truncate *all* the pages from an offset
 385  * @mapping: mapping to truncate
 386  * @lstart: offset from which to truncate
 387  *
 388  * Called under (and serialised by) inode->i_mutex.
 389  *
 390  * Note: When this function returns, there can be a page in the process of
 391  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
 392  * mapping->nrpages can be non-zero when this function returns even after
 393  * truncation of the whole mapping.
 394  */
 395 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 396 {
 397         truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
 398 }
 399 EXPORT_SYMBOL(truncate_inode_pages);
 400
 401 /**
 402  * truncate_inode_pages_final - truncate *all* pages before inode dies
 403  * @mapping: mapping to truncate
 404  *
 405  * Called under (and serialized by) inode->i_mutex.
 406  *
 407  * Filesystems have to use this in the .evict_inode path to inform the
 408  * VM that this is the final truncate and the inode is going away.
 409  */
 410 void truncate_inode_pages_final(struct address_space *mapping)
 411 {
 412         unsigned long nrexceptional;
 413         unsigned long nrpages;
 414
 415         /*
 416          * Page reclaim can not participate in regular inode lifetime
 417          * management (can't call iput()) and thus can race with the
 418          * inode teardown.  Tell it when the address space is exiting,
 419          * so that it does not install eviction information after the
 420          * final truncate has begun.
 421          */
 422         mapping_set_exiting(mapping);
 423
 424         /*
 425          * When reclaim installs eviction entries, it increases
 426          * nrexceptional first, then decreases nrpages.  Make sure we see
 427          * this in the right order or we might miss an entry.
 428          */
 429         nrpages = mapping->nrpages;
 430         smp_rmb();
 431         nrexceptional = mapping->nrexceptional;
 432
 433         if (nrpages || nrexceptional) {
 434                 /*
 435                  * As truncation uses a lockless tree lookup, cycle
 436                  * the tree lock to make sure any ongoing tree
 437                  * modification that does not see AS_EXITING is
 438                  * completed before starting the final truncate.
 439                  */
 440                 spin_lock_irq(&mapping->tree_lock);
 441                 spin_unlock_irq(&mapping->tree_lock);
 442
 443                 truncate_inode_pages(mapping, 0);
 444         }
 445 }
 446 EXPORT_SYMBOL(truncate_inode_pages_final);
 447
 448 /**
 449  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 450  * @mapping: the address_space which holds the pages to invalidate
 451  * @start: the offset 'from' which to invalidate
 452  * @end: the offset 'to' which to invalidate (inclusive)
 453  *
 454  * This function only removes the unlocked pages, if you want to
 455  * remove all the pages of one inode, you must call truncate_inode_pages.
 456  *
 457  * invalidate_mapping_pages() will not block on IO activity. It will not
 458  * invalidate pages which are dirty, locked, under writeback or mapped into
 459  * pagetables.
 460  */
 461 unsigned long invalidate_mapping_pages(struct address_space *mapping,
 462                 pgoff_t start, pgoff_t end)
 463 {
 464         pgoff_t indices[PAGEVEC_SIZE];
 465         struct pagevec pvec;
 466         pgoff_t index = start;
 467         unsigned long ret;
 468         unsigned long count = 0;
 469         int i;
 470
 471         pagevec_init(&pvec, 0);
 472         while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
 473                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 474                         indices)) {
 475                 for (i = 0; i < pagevec_count(&pvec); i++) {
 476                         struct page *page = pvec.pages[i];
 477
 478                         /* We rely upon deletion not changing page->index */
 479                         index = indices[i];
 480                         if (index > end)
 481                                 break;
 482
 483                         if (radix_tree_exceptional_entry(page)) {
 484                                 clear_exceptional_entry(mapping, index, page);
 485                                 continue;
 486                         }
 487
 488                         if (!trylock_page(page))
 489                                 continue;
 490                         WARN_ON(page->index != index);
 491                         ret = invalidate_inode_page(page);
 492                         unlock_page(page);
 493                         /*
 494                          * Invalidation is a hint that the page is no longer
 495                          * of interest and try to speed up its reclaim.
 496                          */
 497                         if (!ret)
 498                                 deactivate_file_page(page);
 499                         count += ret;
 500                 }
 501                 pagevec_remove_exceptionals(&pvec);
 502                 pagevec_release(&pvec);
 503                 cond_resched();
 504                 index++;
 505         }
 506         return count;
 507 }
 508 EXPORT_SYMBOL(invalidate_mapping_pages);
 509
 510 /*
 511  * This is like invalidate_complete_page(), except it ignores the page's
 512  * refcount.  We do this because invalidate_inode_pages2() needs stronger
 513  * invalidation guarantees, and cannot afford to leave pages behind because
 514  * shrink_page_list() has a temp ref on them, or because they're transiently
 515  * sitting in the lru_cache_add() pagevecs.
 516  */
 517 static int
 518 invalidate_complete_page2(struct address_space *mapping, struct page *page)
 519 {
 520         unsigned long flags;
 521
 522         if (page->mapping != mapping)
 523                 return 0;
 524
 525         if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
 526                 return 0;
 527
 528         spin_lock_irqsave(&mapping->tree_lock, flags);
 529         if (PageDirty(page))
 530                 goto failed;
 531
 532         BUG_ON(page_has_private(page));
 533         __delete_from_page_cache(page, NULL);
 534         spin_unlock_irqrestore(&mapping->tree_lock, flags);
 535
 536         if (mapping->a_ops->freepage)
 537                 mapping->a_ops->freepage(page);
 538
 539         put_page(page); /* pagecache ref */
 540         return 1;
 541 failed:
 542         spin_unlock_irqrestore(&mapping->tree_lock, flags);
 543         return 0;
 544 }
 545
 546 static int do_launder_page(struct address_space *mapping, struct page *page)
 547 {
 548         if (!PageDirty(page))
 549                 return 0;
 550         if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
 551                 return 0;
 552         return mapping->a_ops->launder_page(page);
 553 }
 554
 555 /**
 556  * invalidate_inode_pages2_range - remove range of pages from an address_space
 557  * @mapping: the address_space
 558  * @start: the page offset 'from' which to invalidate
 559  * @end: the page offset 'to' which to invalidate (inclusive)
 560  *
 561  * Any pages which are found to be mapped into pagetables are unmapped prior to
 562  * invalidation.
 563  *
 564  * Returns -EBUSY if any pages could not be invalidated.
 565  */
 566 int invalidate_inode_pages2_range(struct address_space *mapping,
 567                                   pgoff_t start, pgoff_t end)
 568 {
 569         pgoff_t indices[PAGEVEC_SIZE];
 570         struct pagevec pvec;
 571         pgoff_t index;
 572         int i;
 573         int ret = 0;
 574         int ret2 = 0;
 575         int did_range_unmap = 0;
 576
 577         cleancache_invalidate_inode(mapping);
 578         pagevec_init(&pvec, 0);
 579         index = start;
 580         while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
 581                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 582                         indices)) {
 583                 for (i = 0; i < pagevec_count(&pvec); i++) {
 584                         struct page *page = pvec.pages[i];
 585
 586                         /* We rely upon deletion not changing page->index */
 587                         index = indices[i];
 588                         if (index > end)
 589                                 break;
 590
 591                         if (radix_tree_exceptional_entry(page)) {
 592                                 clear_exceptional_entry(mapping, index, page);
 593                                 continue;
 594                         }
 595
 596                         lock_page(page);
 597                         WARN_ON(page->index != index);
 598                         if (page->mapping != mapping) {
 599                                 unlock_page(page);
 600                                 continue;
 601                         }
 602                         wait_on_page_writeback(page);
 603                         if (page_mapped(page)) {
 604                                 if (!did_range_unmap) {
 605                                         /*
 606                                          * Zap the rest of the file in one hit.
 607                                          */
 608                                         unmap_mapping_range(mapping,
 609                                            (loff_t)index << PAGE_SHIFT,
 610                                            (loff_t)(1 + end - index)
 611                                                          << PAGE_SHIFT,
 612                                                          0);
 613                                         did_range_unmap = 1;
 614                                 } else {
 615                                         /*
 616                                          * Just zap this page
 617                                          */
 618                                         unmap_mapping_range(mapping,
 619                                            (loff_t)index << PAGE_SHIFT,
 620                                            PAGE_SIZE, 0);
 621                                 }
 622                         }
 623                         BUG_ON(page_mapped(page));
 624                         ret2 = do_launder_page(mapping, page);
 625                         if (ret2 == 0) {
 626                                 if (!invalidate_complete_page2(mapping, page))
 627                                         ret2 = -EBUSY;
 628                         }
 629                         if (ret2 < 0)
 630                                 ret = ret2;
 631                         unlock_page(page);
 632                 }
 633                 pagevec_remove_exceptionals(&pvec);
 634                 pagevec_release(&pvec);
 635                 cond_resched();
 636                 index++;
 637         }
 638         cleancache_invalidate_inode(mapping);
 639         return ret;
 640 }
 641 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
 642
 643 /**
 644  * invalidate_inode_pages2 - remove all pages from an address_space
 645  * @mapping: the address_space
 646  *
 647  * Any pages which are found to be mapped into pagetables are unmapped prior to
 648  * invalidation.
 649  *
 650  * Returns -EBUSY if any pages could not be invalidated.
 651  */
 652 int invalidate_inode_pages2(struct address_space *mapping)
 653 {
 654         return invalidate_inode_pages2_range(mapping, 0, -1);
 655 }
 656 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
 657
 658 /**
 659  * truncate_pagecache - unmap and remove pagecache that has been truncated
 660  * @inode: inode
 661  * @newsize: new file size
 662  *
 663  * inode's new i_size must already be written before truncate_pagecache
 664  * is called.
 665  *
 666  * This function should typically be called before the filesystem
 667  * releases resources associated with the freed range (eg. deallocates
 668  * blocks). This way, pagecache will always stay logically coherent
 669  * with on-disk format, and the filesystem would not have to deal with
 670  * situations such as writepage being called for a page that has already
 671  * had its underlying blocks deallocated.
 672  */
 673 void truncate_pagecache(struct inode *inode, loff_t newsize)
 674 {
 675         struct address_space *mapping = inode->i_mapping;
 676         loff_t holebegin = round_up(newsize, PAGE_SIZE);
 677
 678         /*
 679          * unmap_mapping_range is called twice, first simply for
 680          * efficiency so that truncate_inode_pages does fewer
 681          * single-page unmaps.  However after this first call, and
 682          * before truncate_inode_pages finishes, it is possible for
 683          * private pages to be COWed, which remain after
 684          * truncate_inode_pages finishes, hence the second
 685          * unmap_mapping_range call must be made for correctness.
 686          */
 687         unmap_mapping_range(mapping, holebegin, 0, 1);
 688         truncate_inode_pages(mapping, newsize);
 689         unmap_mapping_range(mapping, holebegin, 0, 1);
 690 }
 691 EXPORT_SYMBOL(truncate_pagecache);
 692
 693 /**
 694  * truncate_setsize - update inode and pagecache for a new file size
 695  * @inode: inode
 696  * @newsize: new file size
 697  *
 698  * truncate_setsize updates i_size and performs pagecache truncation (if
 699  * necessary) to @newsize. It will be typically be called from the filesystem's
 700  * setattr function when ATTR_SIZE is passed in.
 701  *
 702  * Must be called with a lock serializing truncates and writes (generally
 703  * i_mutex but e.g. xfs uses a different lock) and before all filesystem
 704  * specific block truncation has been performed.
 705  */
 706 void truncate_setsize(struct inode *inode, loff_t newsize)
 707 {
 708         loff_t oldsize = inode->i_size;
 709
 710         i_size_write(inode, newsize);
 711         if (newsize > oldsize)
 712                 pagecache_isize_extended(inode, oldsize, newsize);
 713         truncate_pagecache(inode, newsize);
 714 }
 715 EXPORT_SYMBOL(truncate_setsize);
 716
 717 /**
 718  * pagecache_isize_extended - update pagecache after extension of i_size
 719  * @inode:      inode for which i_size was extended
 720  * @from:       original inode size
 721  * @to:         new inode size
 722  *
 723  * Handle extension of inode size either caused by extending truncate or by
 724  * write starting after current i_size. We mark the page straddling current
 725  * i_size RO so that page_mkwrite() is called on the nearest write access to
 726  * the page.  This way filesystem can be sure that page_mkwrite() is called on
 727  * the page before user writes to the page via mmap after the i_size has been
 728  * changed.
 729  *
 730  * The function must be called after i_size is updated so that page fault
 731  * coming after we unlock the page will already see the new i_size.
 732  * The function must be called while we still hold i_mutex - this not only
 733  * makes sure i_size is stable but also that userspace cannot observe new
 734  * i_size value before we are prepared to store mmap writes at new inode size.
 735  */
 736 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
 737 {
 738         int bsize = 1 << inode->i_blkbits;
 739         loff_t rounded_from;
 740         struct page *page;
 741         pgoff_t index;
 742
 743         WARN_ON(to > inode->i_size);
 744
 745         if (from >= to || bsize == PAGE_SIZE)
 746                 return;
 747         /* Page straddling @from will not have any hole block created? */
 748         rounded_from = round_up(from, bsize);
 749         if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
 750                 return;
 751
 752         index = from >> PAGE_SHIFT;
 753         page = find_lock_page(inode->i_mapping, index);
 754         /* Page not cached? Nothing to do */
 755         if (!page)
 756                 return;
 757         /*
 758          * See clear_page_dirty_for_io() for details why set_page_dirty()
 759          * is needed.
 760          */
 761         if (page_mkclean(page))
 762                 set_page_dirty(page);
 763         unlock_page(page);
 764         put_page(page);
 765 }
 766 EXPORT_SYMBOL(pagecache_isize_extended);
 767
 768 /**
 769  * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
 770  * @inode: inode
 771  * @lstart: offset of beginning of hole
 772  * @lend: offset of last byte of hole
 773  *
 774  * This function should typically be called before the filesystem
 775  * releases resources associated with the freed range (eg. deallocates
 776  * blocks). This way, pagecache will always stay logically coherent
 777  * with on-disk format, and the filesystem would not have to deal with
 778  * situations such as writepage being called for a page that has already
 779  * had its underlying blocks deallocated.
 780  */
 781 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
 782 {
 783         struct address_space *mapping = inode->i_mapping;
 784         loff_t unmap_start = round_up(lstart, PAGE_SIZE);
 785         loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
 786         /*
 787          * This rounding is currently just for example: unmap_mapping_range
 788          * expands its hole outwards, whereas we want it to contract the hole
 789          * inwards.  However, existing callers of truncate_pagecache_range are
 790          * doing their own page rounding first.  Note that unmap_mapping_range
 791          * allows holelen 0 for all, and we allow lend -1 for end of file.
 792          */
 793
 794         /*
 795          * Unlike in truncate_pagecache, unmap_mapping_range is called only
 796          * once (before truncating pagecache), and without "even_cows" flag:
 797          * hole-punching should not remove private COWed pages from the hole.
 798          */
 799         if ((u64)unmap_end > (u64)unmap_start)
 800                 unmap_mapping_range(mapping, unmap_start,
 801                                     1 + unmap_end - unmap_start, 0);
 802         truncate_inode_pages_range(mapping, lstart, lend);
 803 }
 804 EXPORT_SYMBOL(truncate_pagecache_range);