linux/mm/truncate.c
<<
>>
Prefs
   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
  27static 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);
  69unlock:
  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 */
  88void 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 */
 112static int
 113truncate_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 */
 140static int
 141invalidate_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
 156int truncate_inode_page(struct address_space *mapping, struct page *page)
 157{
 158        loff_t holelen;
 159        VM_BUG_ON_PAGE(PageTail(page), page);
 160
 161        holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE;
 162        if (page_mapped(page)) {
 163                unmap_mapping_range(mapping,
 164                                   (loff_t)page->index << PAGE_SHIFT,
 165                                   holelen, 0);
 166        }
 167        return truncate_complete_page(mapping, page);
 168}
 169
 170/*
 171 * Used to get rid of pages on hardware memory corruption.
 172 */
 173int generic_error_remove_page(struct address_space *mapping, struct page *page)
 174{
 175        if (!mapping)
 176                return -EINVAL;
 177        /*
 178         * Only punch for normal data pages for now.
 179         * Handling other types like directories would need more auditing.
 180         */
 181        if (!S_ISREG(mapping->host->i_mode))
 182                return -EIO;
 183        return truncate_inode_page(mapping, page);
 184}
 185EXPORT_SYMBOL(generic_error_remove_page);
 186
 187/*
 188 * Safely invalidate one page from its pagecache mapping.
 189 * It only drops clean, unused pages. The page must be locked.
 190 *
 191 * Returns 1 if the page is successfully invalidated, otherwise 0.
 192 */
 193int invalidate_inode_page(struct page *page)
 194{
 195        struct address_space *mapping = page_mapping(page);
 196        if (!mapping)
 197                return 0;
 198        if (PageDirty(page) || PageWriteback(page))
 199                return 0;
 200        if (page_mapped(page))
 201                return 0;
 202        return invalidate_complete_page(mapping, page);
 203}
 204
 205/**
 206 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
 207 * @mapping: mapping to truncate
 208 * @lstart: offset from which to truncate
 209 * @lend: offset to which to truncate (inclusive)
 210 *
 211 * Truncate the page cache, removing the pages that are between
 212 * specified offsets (and zeroing out partial pages
 213 * if lstart or lend + 1 is not page aligned).
 214 *
 215 * Truncate takes two passes - the first pass is nonblocking.  It will not
 216 * block on page locks and it will not block on writeback.  The second pass
 217 * will wait.  This is to prevent as much IO as possible in the affected region.
 218 * The first pass will remove most pages, so the search cost of the second pass
 219 * is low.
 220 *
 221 * We pass down the cache-hot hint to the page freeing code.  Even if the
 222 * mapping is large, it is probably the case that the final pages are the most
 223 * recently touched, and freeing happens in ascending file offset order.
 224 *
 225 * Note that since ->invalidatepage() accepts range to invalidate
 226 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
 227 * page aligned properly.
 228 */
 229void truncate_inode_pages_range(struct address_space *mapping,
 230                                loff_t lstart, loff_t lend)
 231{
 232        pgoff_t         start;          /* inclusive */
 233        pgoff_t         end;            /* exclusive */
 234        unsigned int    partial_start;  /* inclusive */
 235        unsigned int    partial_end;    /* exclusive */
 236        struct pagevec  pvec;
 237        pgoff_t         indices[PAGEVEC_SIZE];
 238        pgoff_t         index;
 239        int             i;
 240
 241        cleancache_invalidate_inode(mapping);
 242        if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
 243                return;
 244
 245        /* Offsets within partial pages */
 246        partial_start = lstart & (PAGE_SIZE - 1);
 247        partial_end = (lend + 1) & (PAGE_SIZE - 1);
 248
 249        /*
 250         * 'start' and 'end' always covers the range of pages to be fully
 251         * truncated. Partial pages are covered with 'partial_start' at the
 252         * start of the range and 'partial_end' at the end of the range.
 253         * Note that 'end' is exclusive while 'lend' is inclusive.
 254         */
 255        start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
 256        if (lend == -1)
 257                /*
 258                 * lend == -1 indicates end-of-file so we have to set 'end'
 259                 * to the highest possible pgoff_t and since the type is
 260                 * unsigned we're using -1.
 261                 */
 262                end = -1;
 263        else
 264                end = (lend + 1) >> PAGE_SHIFT;
 265
 266        pagevec_init(&pvec, 0);
 267        index = start;
 268        while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
 269                        min(end - index, (pgoff_t)PAGEVEC_SIZE),
 270                        indices)) {
 271                for (i = 0; i < pagevec_count(&pvec); i++) {
 272                        struct page *page = pvec.pages[i];
 273
 274                        /* We rely upon deletion not changing page->index */
 275                        index = indices[i];
 276                        if (index >= end)
 277                                break;
 278
 279                        if (radix_tree_exceptional_entry(page)) {
 280                                clear_exceptional_entry(mapping, index, page);
 281                                continue;
 282                        }
 283
 284                        if (!trylock_page(page))
 285                                continue;
 286                        WARN_ON(page_to_index(page) != index);
 287                        if (PageWriteback(page)) {
 288                                unlock_page(page);
 289                                continue;
 290                        }
 291                        truncate_inode_page(mapping, page);
 292                        unlock_page(page);
 293                }
 294                pagevec_remove_exceptionals(&pvec);
 295                pagevec_release(&pvec);
 296                cond_resched();
 297                index++;
 298        }
 299
 300        if (partial_start) {
 301                struct page *page = find_lock_page(mapping, start - 1);
 302                if (page) {
 303                        unsigned int top = PAGE_SIZE;
 304                        if (start > end) {
 305                                /* Truncation within a single page */
 306                                top = partial_end;
 307                                partial_end = 0;
 308                        }
 309                        wait_on_page_writeback(page);
 310                        zero_user_segment(page, partial_start, top);
 311                        cleancache_invalidate_page(mapping, page);
 312                        if (page_has_private(page))
 313                                do_invalidatepage(page, partial_start,
 314                                                  top - partial_start);
 315                        unlock_page(page);
 316                        put_page(page);
 317                }
 318        }
 319        if (partial_end) {
 320                struct page *page = find_lock_page(mapping, end);
 321                if (page) {
 322                        wait_on_page_writeback(page);
 323                        zero_user_segment(page, 0, partial_end);
 324                        cleancache_invalidate_page(mapping, page);
 325                        if (page_has_private(page))
 326                                do_invalidatepage(page, 0,
 327                                                  partial_end);
 328                        unlock_page(page);
 329                        put_page(page);
 330                }
 331        }
 332        /*
 333         * If the truncation happened within a single page no pages
 334         * will be released, just zeroed, so we can bail out now.
 335         */
 336        if (start >= end)
 337                return;
 338
 339        index = start;
 340        for ( ; ; ) {
 341                cond_resched();
 342                if (!pagevec_lookup_entries(&pvec, mapping, index,
 343                        min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
 344                        /* If all gone from start onwards, we're done */
 345                        if (index == start)
 346                                break;
 347                        /* Otherwise restart to make sure all gone */
 348                        index = start;
 349                        continue;
 350                }
 351                if (index == start && indices[0] >= end) {
 352                        /* All gone out of hole to be punched, we're done */
 353                        pagevec_remove_exceptionals(&pvec);
 354                        pagevec_release(&pvec);
 355                        break;
 356                }
 357                for (i = 0; i < pagevec_count(&pvec); i++) {
 358                        struct page *page = pvec.pages[i];
 359
 360                        /* We rely upon deletion not changing page->index */
 361                        index = indices[i];
 362                        if (index >= end) {
 363                                /* Restart punch to make sure all gone */
 364                                index = start - 1;
 365                                break;
 366                        }
 367
 368                        if (radix_tree_exceptional_entry(page)) {
 369                                clear_exceptional_entry(mapping, index, page);
 370                                continue;
 371                        }
 372
 373                        lock_page(page);
 374                        WARN_ON(page_to_index(page) != index);
 375                        wait_on_page_writeback(page);
 376                        truncate_inode_page(mapping, page);
 377                        unlock_page(page);
 378                }
 379                pagevec_remove_exceptionals(&pvec);
 380                pagevec_release(&pvec);
 381                index++;
 382        }
 383        cleancache_invalidate_inode(mapping);
 384}
 385EXPORT_SYMBOL(truncate_inode_pages_range);
 386
 387/**
 388 * truncate_inode_pages - truncate *all* the pages from an offset
 389 * @mapping: mapping to truncate
 390 * @lstart: offset from which to truncate
 391 *
 392 * Called under (and serialised by) inode->i_mutex.
 393 *
 394 * Note: When this function returns, there can be a page in the process of
 395 * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
 396 * mapping->nrpages can be non-zero when this function returns even after
 397 * truncation of the whole mapping.
 398 */
 399void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 400{
 401        truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
 402}
 403EXPORT_SYMBOL(truncate_inode_pages);
 404
 405/**
 406 * truncate_inode_pages_final - truncate *all* pages before inode dies
 407 * @mapping: mapping to truncate
 408 *
 409 * Called under (and serialized by) inode->i_mutex.
 410 *
 411 * Filesystems have to use this in the .evict_inode path to inform the
 412 * VM that this is the final truncate and the inode is going away.
 413 */
 414void truncate_inode_pages_final(struct address_space *mapping)
 415{
 416        unsigned long nrexceptional;
 417        unsigned long nrpages;
 418
 419        /*
 420         * Page reclaim can not participate in regular inode lifetime
 421         * management (can't call iput()) and thus can race with the
 422         * inode teardown.  Tell it when the address space is exiting,
 423         * so that it does not install eviction information after the
 424         * final truncate has begun.
 425         */
 426        mapping_set_exiting(mapping);
 427
 428        /*
 429         * When reclaim installs eviction entries, it increases
 430         * nrexceptional first, then decreases nrpages.  Make sure we see
 431         * this in the right order or we might miss an entry.
 432         */
 433        nrpages = mapping->nrpages;
 434        smp_rmb();
 435        nrexceptional = mapping->nrexceptional;
 436
 437        if (nrpages || nrexceptional) {
 438                /*
 439                 * As truncation uses a lockless tree lookup, cycle
 440                 * the tree lock to make sure any ongoing tree
 441                 * modification that does not see AS_EXITING is
 442                 * completed before starting the final truncate.
 443                 */
 444                spin_lock_irq(&mapping->tree_lock);
 445                spin_unlock_irq(&mapping->tree_lock);
 446
 447                truncate_inode_pages(mapping, 0);
 448        }
 449}
 450EXPORT_SYMBOL(truncate_inode_pages_final);
 451
 452/**
 453 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 454 * @mapping: the address_space which holds the pages to invalidate
 455 * @start: the offset 'from' which to invalidate
 456 * @end: the offset 'to' which to invalidate (inclusive)
 457 *
 458 * This function only removes the unlocked pages, if you want to
 459 * remove all the pages of one inode, you must call truncate_inode_pages.
 460 *
 461 * invalidate_mapping_pages() will not block on IO activity. It will not
 462 * invalidate pages which are dirty, locked, under writeback or mapped into
 463 * pagetables.
 464 */
 465unsigned long invalidate_mapping_pages(struct address_space *mapping,
 466                pgoff_t start, pgoff_t end)
 467{
 468        pgoff_t indices[PAGEVEC_SIZE];
 469        struct pagevec pvec;
 470        pgoff_t index = start;
 471        unsigned long ret;
 472        unsigned long count = 0;
 473        int i;
 474
 475        pagevec_init(&pvec, 0);
 476        while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
 477                        min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 478                        indices)) {
 479                for (i = 0; i < pagevec_count(&pvec); i++) {
 480                        struct page *page = pvec.pages[i];
 481
 482                        /* We rely upon deletion not changing page->index */
 483                        index = indices[i];
 484                        if (index > end)
 485                                break;
 486
 487                        if (radix_tree_exceptional_entry(page)) {
 488                                clear_exceptional_entry(mapping, index, page);
 489                                continue;
 490                        }
 491
 492                        if (!trylock_page(page))
 493                                continue;
 494
 495                        WARN_ON(page_to_index(page) != index);
 496
 497                        /* Middle of THP: skip */
 498                        if (PageTransTail(page)) {
 499                                unlock_page(page);
 500                                continue;
 501                        } else if (PageTransHuge(page)) {
 502                                index += HPAGE_PMD_NR - 1;
 503                                i += HPAGE_PMD_NR - 1;
 504                                /* 'end' is in the middle of THP */
 505                                if (index ==  round_down(end, HPAGE_PMD_NR))
 506                                        continue;
 507                        }
 508
 509                        ret = invalidate_inode_page(page);
 510                        unlock_page(page);
 511                        /*
 512                         * Invalidation is a hint that the page is no longer
 513                         * of interest and try to speed up its reclaim.
 514                         */
 515                        if (!ret)
 516                                deactivate_file_page(page);
 517                        count += ret;
 518                }
 519                pagevec_remove_exceptionals(&pvec);
 520                pagevec_release(&pvec);
 521                cond_resched();
 522                index++;
 523        }
 524        return count;
 525}
 526EXPORT_SYMBOL(invalidate_mapping_pages);
 527
 528/*
 529 * This is like invalidate_complete_page(), except it ignores the page's
 530 * refcount.  We do this because invalidate_inode_pages2() needs stronger
 531 * invalidation guarantees, and cannot afford to leave pages behind because
 532 * shrink_page_list() has a temp ref on them, or because they're transiently
 533 * sitting in the lru_cache_add() pagevecs.
 534 */
 535static int
 536invalidate_complete_page2(struct address_space *mapping, struct page *page)
 537{
 538        unsigned long flags;
 539
 540        if (page->mapping != mapping)
 541                return 0;
 542
 543        if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
 544                return 0;
 545
 546        spin_lock_irqsave(&mapping->tree_lock, flags);
 547        if (PageDirty(page))
 548                goto failed;
 549
 550        BUG_ON(page_has_private(page));
 551        __delete_from_page_cache(page, NULL);
 552        spin_unlock_irqrestore(&mapping->tree_lock, flags);
 553
 554        if (mapping->a_ops->freepage)
 555                mapping->a_ops->freepage(page);
 556
 557        put_page(page); /* pagecache ref */
 558        return 1;
 559failed:
 560        spin_unlock_irqrestore(&mapping->tree_lock, flags);
 561        return 0;
 562}
 563
 564static int do_launder_page(struct address_space *mapping, struct page *page)
 565{
 566        if (!PageDirty(page))
 567                return 0;
 568        if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
 569                return 0;
 570        return mapping->a_ops->launder_page(page);
 571}
 572
 573/**
 574 * invalidate_inode_pages2_range - remove range of pages from an address_space
 575 * @mapping: the address_space
 576 * @start: the page offset 'from' which to invalidate
 577 * @end: the page offset 'to' which to invalidate (inclusive)
 578 *
 579 * Any pages which are found to be mapped into pagetables are unmapped prior to
 580 * invalidation.
 581 *
 582 * Returns -EBUSY if any pages could not be invalidated.
 583 */
 584int invalidate_inode_pages2_range(struct address_space *mapping,
 585                                  pgoff_t start, pgoff_t end)
 586{
 587        pgoff_t indices[PAGEVEC_SIZE];
 588        struct pagevec pvec;
 589        pgoff_t index;
 590        int i;
 591        int ret = 0;
 592        int ret2 = 0;
 593        int did_range_unmap = 0;
 594
 595        cleancache_invalidate_inode(mapping);
 596        pagevec_init(&pvec, 0);
 597        index = start;
 598        while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
 599                        min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 600                        indices)) {
 601                for (i = 0; i < pagevec_count(&pvec); i++) {
 602                        struct page *page = pvec.pages[i];
 603
 604                        /* We rely upon deletion not changing page->index */
 605                        index = indices[i];
 606                        if (index > end)
 607                                break;
 608
 609                        if (radix_tree_exceptional_entry(page)) {
 610                                clear_exceptional_entry(mapping, index, page);
 611                                continue;
 612                        }
 613
 614                        lock_page(page);
 615                        WARN_ON(page_to_index(page) != index);
 616                        if (page->mapping != mapping) {
 617                                unlock_page(page);
 618                                continue;
 619                        }
 620                        wait_on_page_writeback(page);
 621                        if (page_mapped(page)) {
 622                                if (!did_range_unmap) {
 623                                        /*
 624                                         * Zap the rest of the file in one hit.
 625                                         */
 626                                        unmap_mapping_range(mapping,
 627                                           (loff_t)index << PAGE_SHIFT,
 628                                           (loff_t)(1 + end - index)
 629                                                         << PAGE_SHIFT,
 630                                                         0);
 631                                        did_range_unmap = 1;
 632                                } else {
 633                                        /*
 634                                         * Just zap this page
 635                                         */
 636                                        unmap_mapping_range(mapping,
 637                                           (loff_t)index << PAGE_SHIFT,
 638                                           PAGE_SIZE, 0);
 639                                }
 640                        }
 641                        BUG_ON(page_mapped(page));
 642                        ret2 = do_launder_page(mapping, page);
 643                        if (ret2 == 0) {
 644                                if (!invalidate_complete_page2(mapping, page))
 645                                        ret2 = -EBUSY;
 646                        }
 647                        if (ret2 < 0)
 648                                ret = ret2;
 649                        unlock_page(page);
 650                }
 651                pagevec_remove_exceptionals(&pvec);
 652                pagevec_release(&pvec);
 653                cond_resched();
 654                index++;
 655        }
 656        cleancache_invalidate_inode(mapping);
 657        return ret;
 658}
 659EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
 660
 661/**
 662 * invalidate_inode_pages2 - remove all pages from an address_space
 663 * @mapping: the address_space
 664 *
 665 * Any pages which are found to be mapped into pagetables are unmapped prior to
 666 * invalidation.
 667 *
 668 * Returns -EBUSY if any pages could not be invalidated.
 669 */
 670int invalidate_inode_pages2(struct address_space *mapping)
 671{
 672        return invalidate_inode_pages2_range(mapping, 0, -1);
 673}
 674EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
 675
 676/**
 677 * truncate_pagecache - unmap and remove pagecache that has been truncated
 678 * @inode: inode
 679 * @newsize: new file size
 680 *
 681 * inode's new i_size must already be written before truncate_pagecache
 682 * is called.
 683 *
 684 * This function should typically be called before the filesystem
 685 * releases resources associated with the freed range (eg. deallocates
 686 * blocks). This way, pagecache will always stay logically coherent
 687 * with on-disk format, and the filesystem would not have to deal with
 688 * situations such as writepage being called for a page that has already
 689 * had its underlying blocks deallocated.
 690 */
 691void truncate_pagecache(struct inode *inode, loff_t newsize)
 692{
 693        struct address_space *mapping = inode->i_mapping;
 694        loff_t holebegin = round_up(newsize, PAGE_SIZE);
 695
 696        /*
 697         * unmap_mapping_range is called twice, first simply for
 698         * efficiency so that truncate_inode_pages does fewer
 699         * single-page unmaps.  However after this first call, and
 700         * before truncate_inode_pages finishes, it is possible for
 701         * private pages to be COWed, which remain after
 702         * truncate_inode_pages finishes, hence the second
 703         * unmap_mapping_range call must be made for correctness.
 704         */
 705        unmap_mapping_range(mapping, holebegin, 0, 1);
 706        truncate_inode_pages(mapping, newsize);
 707        unmap_mapping_range(mapping, holebegin, 0, 1);
 708}
 709EXPORT_SYMBOL(truncate_pagecache);
 710
 711/**
 712 * truncate_setsize - update inode and pagecache for a new file size
 713 * @inode: inode
 714 * @newsize: new file size
 715 *
 716 * truncate_setsize updates i_size and performs pagecache truncation (if
 717 * necessary) to @newsize. It will be typically be called from the filesystem's
 718 * setattr function when ATTR_SIZE is passed in.
 719 *
 720 * Must be called with a lock serializing truncates and writes (generally
 721 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
 722 * specific block truncation has been performed.
 723 */
 724void truncate_setsize(struct inode *inode, loff_t newsize)
 725{
 726        loff_t oldsize = inode->i_size;
 727
 728        i_size_write(inode, newsize);
 729        if (newsize > oldsize)
 730                pagecache_isize_extended(inode, oldsize, newsize);
 731        truncate_pagecache(inode, newsize);
 732}
 733EXPORT_SYMBOL(truncate_setsize);
 734
 735/**
 736 * pagecache_isize_extended - update pagecache after extension of i_size
 737 * @inode:      inode for which i_size was extended
 738 * @from:       original inode size
 739 * @to:         new inode size
 740 *
 741 * Handle extension of inode size either caused by extending truncate or by
 742 * write starting after current i_size. We mark the page straddling current
 743 * i_size RO so that page_mkwrite() is called on the nearest write access to
 744 * the page.  This way filesystem can be sure that page_mkwrite() is called on
 745 * the page before user writes to the page via mmap after the i_size has been
 746 * changed.
 747 *
 748 * The function must be called after i_size is updated so that page fault
 749 * coming after we unlock the page will already see the new i_size.
 750 * The function must be called while we still hold i_mutex - this not only
 751 * makes sure i_size is stable but also that userspace cannot observe new
 752 * i_size value before we are prepared to store mmap writes at new inode size.
 753 */
 754void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
 755{
 756        int bsize = 1 << inode->i_blkbits;
 757        loff_t rounded_from;
 758        struct page *page;
 759        pgoff_t index;
 760
 761        WARN_ON(to > inode->i_size);
 762
 763        if (from >= to || bsize == PAGE_SIZE)
 764                return;
 765        /* Page straddling @from will not have any hole block created? */
 766        rounded_from = round_up(from, bsize);
 767        if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
 768                return;
 769
 770        index = from >> PAGE_SHIFT;
 771        page = find_lock_page(inode->i_mapping, index);
 772        /* Page not cached? Nothing to do */
 773        if (!page)
 774                return;
 775        /*
 776         * See clear_page_dirty_for_io() for details why set_page_dirty()
 777         * is needed.
 778         */
 779        if (page_mkclean(page))
 780                set_page_dirty(page);
 781        unlock_page(page);
 782        put_page(page);
 783}
 784EXPORT_SYMBOL(pagecache_isize_extended);
 785
 786/**
 787 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
 788 * @inode: inode
 789 * @lstart: offset of beginning of hole
 790 * @lend: offset of last byte of hole
 791 *
 792 * This function should typically be called before the filesystem
 793 * releases resources associated with the freed range (eg. deallocates
 794 * blocks). This way, pagecache will always stay logically coherent
 795 * with on-disk format, and the filesystem would not have to deal with
 796 * situations such as writepage being called for a page that has already
 797 * had its underlying blocks deallocated.
 798 */
 799void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
 800{
 801        struct address_space *mapping = inode->i_mapping;
 802        loff_t unmap_start = round_up(lstart, PAGE_SIZE);
 803        loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
 804        /*
 805         * This rounding is currently just for example: unmap_mapping_range
 806         * expands its hole outwards, whereas we want it to contract the hole
 807         * inwards.  However, existing callers of truncate_pagecache_range are
 808         * doing their own page rounding first.  Note that unmap_mapping_range
 809         * allows holelen 0 for all, and we allow lend -1 for end of file.
 810         */
 811
 812        /*
 813         * Unlike in truncate_pagecache, unmap_mapping_range is called only
 814         * once (before truncating pagecache), and without "even_cows" flag:
 815         * hole-punching should not remove private COWed pages from the hole.
 816         */
 817        if ((u64)unmap_end > (u64)unmap_start)
 818                unmap_mapping_range(mapping, unmap_start,
 819                                    1 + unmap_end - unmap_start, 0);
 820        truncate_inode_pages_range(mapping, lstart, lend);
 821}
 822EXPORT_SYMBOL(truncate_pagecache_range);
 823