1 // SPDX-License-Identifier: GPL-2.0-only
3 * mm/truncate.c - code for taking down pages from address_spaces
5 * Copyright (C) 2002, Linus Torvalds
7 * 10Sep2002 Andrew Morton
11 #include <linux/kernel.h>
12 #include <linux/backing-dev.h>
13 #include <linux/dax.h>
14 #include <linux/gfp.h>
16 #include <linux/swap.h>
17 #include <linux/export.h>
18 #include <linux/pagemap.h>
19 #include <linux/highmem.h>
20 #include <linux/pagevec.h>
21 #include <linux/task_io_accounting_ops.h>
22 #include <linux/buffer_head.h> /* grr. try_to_release_page,
24 #include <linux/shmem_fs.h>
25 #include <linux/rmap.h>
29 * Regular page slots are stabilized by the page lock even without the tree
30 * itself locked. These unlocked entries need verification under the tree
33 static inline void __clear_shadow_entry(struct address_space *mapping,
34 pgoff_t index, void *entry)
36 XA_STATE(xas, &mapping->i_pages, index);
38 xas_set_update(&xas, workingset_update_node);
39 if (xas_load(&xas) != entry)
41 xas_store(&xas, NULL);
44 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
47 spin_lock(&mapping->host->i_lock);
48 xa_lock_irq(&mapping->i_pages);
49 __clear_shadow_entry(mapping, index, entry);
50 xa_unlock_irq(&mapping->i_pages);
51 if (mapping_shrinkable(mapping))
52 inode_add_lru(mapping->host);
53 spin_unlock(&mapping->host->i_lock);
57 * Unconditionally remove exceptional entries. Usually called from truncate
58 * path. Note that the folio_batch may be altered by this function by removing
59 * exceptional entries similar to what folio_batch_remove_exceptionals() does.
61 static void truncate_folio_batch_exceptionals(struct address_space *mapping,
62 struct folio_batch *fbatch, pgoff_t *indices)
67 /* Handled by shmem itself */
68 if (shmem_mapping(mapping))
71 for (j = 0; j < folio_batch_count(fbatch); j++)
72 if (xa_is_value(fbatch->folios[j]))
75 if (j == folio_batch_count(fbatch))
78 dax = dax_mapping(mapping);
80 spin_lock(&mapping->host->i_lock);
81 xa_lock_irq(&mapping->i_pages);
84 for (i = j; i < folio_batch_count(fbatch); i++) {
85 struct folio *folio = fbatch->folios[i];
86 pgoff_t index = indices[i];
88 if (!xa_is_value(folio)) {
89 fbatch->folios[j++] = folio;
94 dax_delete_mapping_entry(mapping, index);
98 __clear_shadow_entry(mapping, index, folio);
102 xa_unlock_irq(&mapping->i_pages);
103 if (mapping_shrinkable(mapping))
104 inode_add_lru(mapping->host);
105 spin_unlock(&mapping->host->i_lock);
111 * Invalidate exceptional entry if easily possible. This handles exceptional
112 * entries for invalidate_inode_pages().
114 static int invalidate_exceptional_entry(struct address_space *mapping,
115 pgoff_t index, void *entry)
117 /* Handled by shmem itself, or for DAX we do nothing. */
118 if (shmem_mapping(mapping) || dax_mapping(mapping))
120 clear_shadow_entry(mapping, index, entry);
125 * Invalidate exceptional entry if clean. This handles exceptional entries for
126 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
128 static int invalidate_exceptional_entry2(struct address_space *mapping,
129 pgoff_t index, void *entry)
131 /* Handled by shmem itself */
132 if (shmem_mapping(mapping))
134 if (dax_mapping(mapping))
135 return dax_invalidate_mapping_entry_sync(mapping, index);
136 clear_shadow_entry(mapping, index, entry);
141 * folio_invalidate - Invalidate part or all of a folio.
142 * @folio: The folio which is affected.
143 * @offset: start of the range to invalidate
144 * @length: length of the range to invalidate
146 * folio_invalidate() is called when all or part of the folio has become
147 * invalidated by a truncate operation.
149 * folio_invalidate() does not have to release all buffers, but it must
150 * ensure that no dirty buffer is left outside @offset and that no I/O
151 * is underway against any of the blocks which are outside the truncation
152 * point. Because the caller is about to free (and possibly reuse) those
155 void folio_invalidate(struct folio *folio, size_t offset, size_t length)
157 const struct address_space_operations *aops = folio->mapping->a_ops;
158 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
160 if (aops->invalidate_folio) {
161 aops->invalidate_folio(folio, offset, length);
165 invalidatepage = aops->invalidatepage;
167 (*invalidatepage)(&folio->page, offset, length);
169 EXPORT_SYMBOL_GPL(folio_invalidate);
172 * If truncate cannot remove the fs-private metadata from the page, the page
173 * becomes orphaned. It will be left on the LRU and may even be mapped into
174 * user pagetables if we're racing with filemap_fault().
176 * We need to bail out if page->mapping is no longer equal to the original
177 * mapping. This happens a) when the VM reclaimed the page while we waited on
178 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
179 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
181 static void truncate_cleanup_folio(struct folio *folio)
183 if (folio_mapped(folio))
184 unmap_mapping_folio(folio);
186 if (folio_has_private(folio))
187 folio_invalidate(folio, 0, folio_size(folio));
190 * Some filesystems seem to re-dirty the page even after
191 * the VM has canceled the dirty bit (eg ext3 journaling).
192 * Hence dirty accounting check is placed after invalidation.
194 folio_cancel_dirty(folio);
195 folio_clear_mappedtodisk(folio);
199 * This is for invalidate_mapping_pages(). That function can be called at
200 * any time, and is not supposed to throw away dirty pages. But pages can
201 * be marked dirty at any time too, so use remove_mapping which safely
202 * discards clean, unused pages.
204 * Returns non-zero if the page was successfully invalidated.
207 invalidate_complete_page(struct address_space *mapping, struct page *page)
210 if (page->mapping != mapping)
213 if (page_has_private(page) && !try_to_release_page(page, 0))
216 return remove_mapping(mapping, page);
219 int truncate_inode_folio(struct address_space *mapping, struct folio *folio)
221 if (folio->mapping != mapping)
224 truncate_cleanup_folio(folio);
225 filemap_remove_folio(folio);
230 * Handle partial folios. The folio may be entirely within the
231 * range if a split has raced with us. If not, we zero the part of the
232 * folio that's within the [start, end] range, and then split the folio if
233 * it's large. split_page_range() will discard pages which now lie beyond
234 * i_size, and we rely on the caller to discard pages which lie within a
235 * newly created hole.
237 * Returns false if splitting failed so the caller can avoid
238 * discarding the entire folio which is stubbornly unsplit.
240 bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end)
242 loff_t pos = folio_pos(folio);
243 unsigned int offset, length;
246 offset = start - pos;
249 length = folio_size(folio);
250 if (pos + length <= (u64)end)
251 length = length - offset;
253 length = end + 1 - pos - offset;
255 folio_wait_writeback(folio);
256 if (length == folio_size(folio)) {
257 truncate_inode_folio(folio->mapping, folio);
262 * We may be zeroing pages we're about to discard, but it avoids
263 * doing a complex calculation here, and then doing the zeroing
264 * anyway if the page split fails.
266 folio_zero_range(folio, offset, length);
268 if (folio_has_private(folio))
269 folio_invalidate(folio, offset, length);
270 if (!folio_test_large(folio))
272 if (split_huge_page(&folio->page) == 0)
274 if (folio_test_dirty(folio))
276 truncate_inode_folio(folio->mapping, folio);
281 * Used to get rid of pages on hardware memory corruption.
283 int generic_error_remove_page(struct address_space *mapping, struct page *page)
285 VM_BUG_ON_PAGE(PageTail(page), page);
290 * Only punch for normal data pages for now.
291 * Handling other types like directories would need more auditing.
293 if (!S_ISREG(mapping->host->i_mode))
295 return truncate_inode_folio(mapping, page_folio(page));
297 EXPORT_SYMBOL(generic_error_remove_page);
300 * Safely invalidate one page from its pagecache mapping.
301 * It only drops clean, unused pages. The page must be locked.
303 * Returns 1 if the page is successfully invalidated, otherwise 0.
305 int invalidate_inode_page(struct page *page)
307 struct address_space *mapping = page_mapping(page);
310 if (PageDirty(page) || PageWriteback(page))
312 if (page_mapped(page))
314 return invalidate_complete_page(mapping, page);
318 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
319 * @mapping: mapping to truncate
320 * @lstart: offset from which to truncate
321 * @lend: offset to which to truncate (inclusive)
323 * Truncate the page cache, removing the pages that are between
324 * specified offsets (and zeroing out partial pages
325 * if lstart or lend + 1 is not page aligned).
327 * Truncate takes two passes - the first pass is nonblocking. It will not
328 * block on page locks and it will not block on writeback. The second pass
329 * will wait. This is to prevent as much IO as possible in the affected region.
330 * The first pass will remove most pages, so the search cost of the second pass
333 * We pass down the cache-hot hint to the page freeing code. Even if the
334 * mapping is large, it is probably the case that the final pages are the most
335 * recently touched, and freeing happens in ascending file offset order.
337 * Note that since ->invalidatepage() accepts range to invalidate
338 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
339 * page aligned properly.
341 void truncate_inode_pages_range(struct address_space *mapping,
342 loff_t lstart, loff_t lend)
344 pgoff_t start; /* inclusive */
345 pgoff_t end; /* exclusive */
346 struct folio_batch fbatch;
347 pgoff_t indices[PAGEVEC_SIZE];
353 if (mapping_empty(mapping))
357 * 'start' and 'end' always covers the range of pages to be fully
358 * truncated. Partial pages are covered with 'partial_start' at the
359 * start of the range and 'partial_end' at the end of the range.
360 * Note that 'end' is exclusive while 'lend' is inclusive.
362 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
365 * lend == -1 indicates end-of-file so we have to set 'end'
366 * to the highest possible pgoff_t and since the type is
367 * unsigned we're using -1.
371 end = (lend + 1) >> PAGE_SHIFT;
373 folio_batch_init(&fbatch);
375 while (index < end && find_lock_entries(mapping, index, end - 1,
377 index = indices[folio_batch_count(&fbatch) - 1] + 1;
378 truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
379 for (i = 0; i < folio_batch_count(&fbatch); i++)
380 truncate_cleanup_folio(fbatch.folios[i]);
381 delete_from_page_cache_batch(mapping, &fbatch);
382 for (i = 0; i < folio_batch_count(&fbatch); i++)
383 folio_unlock(fbatch.folios[i]);
384 folio_batch_release(&fbatch);
388 same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT);
389 folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0);
391 same_folio = lend < folio_pos(folio) + folio_size(folio);
392 if (!truncate_inode_partial_folio(folio, lstart, lend)) {
393 start = folio->index + folio_nr_pages(folio);
403 folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT,
406 if (!truncate_inode_partial_folio(folio, lstart, lend))
413 while (index < end) {
415 if (!find_get_entries(mapping, index, end - 1, &fbatch,
417 /* If all gone from start onwards, we're done */
420 /* Otherwise restart to make sure all gone */
425 for (i = 0; i < folio_batch_count(&fbatch); i++) {
426 struct folio *folio = fbatch.folios[i];
428 /* We rely upon deletion not changing page->index */
431 if (xa_is_value(folio))
435 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
436 folio_wait_writeback(folio);
437 truncate_inode_folio(mapping, folio);
439 index = folio_index(folio) + folio_nr_pages(folio) - 1;
441 truncate_folio_batch_exceptionals(mapping, &fbatch, indices);
442 folio_batch_release(&fbatch);
446 EXPORT_SYMBOL(truncate_inode_pages_range);
449 * truncate_inode_pages - truncate *all* the pages from an offset
450 * @mapping: mapping to truncate
451 * @lstart: offset from which to truncate
453 * Called under (and serialised by) inode->i_rwsem and
454 * mapping->invalidate_lock.
456 * Note: When this function returns, there can be a page in the process of
457 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
458 * mapping->nrpages can be non-zero when this function returns even after
459 * truncation of the whole mapping.
461 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
463 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
465 EXPORT_SYMBOL(truncate_inode_pages);
468 * truncate_inode_pages_final - truncate *all* pages before inode dies
469 * @mapping: mapping to truncate
471 * Called under (and serialized by) inode->i_rwsem.
473 * Filesystems have to use this in the .evict_inode path to inform the
474 * VM that this is the final truncate and the inode is going away.
476 void truncate_inode_pages_final(struct address_space *mapping)
479 * Page reclaim can not participate in regular inode lifetime
480 * management (can't call iput()) and thus can race with the
481 * inode teardown. Tell it when the address space is exiting,
482 * so that it does not install eviction information after the
483 * final truncate has begun.
485 mapping_set_exiting(mapping);
487 if (!mapping_empty(mapping)) {
489 * As truncation uses a lockless tree lookup, cycle
490 * the tree lock to make sure any ongoing tree
491 * modification that does not see AS_EXITING is
492 * completed before starting the final truncate.
494 xa_lock_irq(&mapping->i_pages);
495 xa_unlock_irq(&mapping->i_pages);
498 truncate_inode_pages(mapping, 0);
500 EXPORT_SYMBOL(truncate_inode_pages_final);
502 static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
503 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
505 pgoff_t indices[PAGEVEC_SIZE];
506 struct folio_batch fbatch;
507 pgoff_t index = start;
509 unsigned long count = 0;
512 folio_batch_init(&fbatch);
513 while (find_lock_entries(mapping, index, end, &fbatch, indices)) {
514 for (i = 0; i < folio_batch_count(&fbatch); i++) {
515 struct page *page = &fbatch.folios[i]->page;
517 /* We rely upon deletion not changing page->index */
520 if (xa_is_value(page)) {
521 count += invalidate_exceptional_entry(mapping,
526 index += thp_nr_pages(page) - 1;
528 ret = invalidate_inode_page(page);
531 * Invalidation is a hint that the page is no longer
532 * of interest and try to speed up its reclaim.
535 deactivate_file_page(page);
536 /* It is likely on the pagevec of a remote CPU */
542 folio_batch_remove_exceptionals(&fbatch);
543 folio_batch_release(&fbatch);
551 * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode
552 * @mapping: the address_space which holds the cache to invalidate
553 * @start: the offset 'from' which to invalidate
554 * @end: the offset 'to' which to invalidate (inclusive)
556 * This function removes pages that are clean, unmapped and unlocked,
557 * as well as shadow entries. It will not block on IO activity.
559 * If you want to remove all the pages of one inode, regardless of
560 * their use and writeback state, use truncate_inode_pages().
562 * Return: the number of the cache entries that were invalidated
564 unsigned long invalidate_mapping_pages(struct address_space *mapping,
565 pgoff_t start, pgoff_t end)
567 return __invalidate_mapping_pages(mapping, start, end, NULL);
569 EXPORT_SYMBOL(invalidate_mapping_pages);
572 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
573 * @mapping: the address_space which holds the pages to invalidate
574 * @start: the offset 'from' which to invalidate
575 * @end: the offset 'to' which to invalidate (inclusive)
576 * @nr_pagevec: invalidate failed page number for caller
578 * This helper is similar to invalidate_mapping_pages(), except that it accounts
579 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
580 * will be used by the caller.
582 void invalidate_mapping_pagevec(struct address_space *mapping,
583 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
585 __invalidate_mapping_pages(mapping, start, end, nr_pagevec);
589 * This is like invalidate_complete_page(), except it ignores the page's
590 * refcount. We do this because invalidate_inode_pages2() needs stronger
591 * invalidation guarantees, and cannot afford to leave pages behind because
592 * shrink_page_list() has a temp ref on them, or because they're transiently
593 * sitting in the lru_cache_add() pagevecs.
595 static int invalidate_complete_folio2(struct address_space *mapping,
598 if (folio->mapping != mapping)
601 if (folio_has_private(folio) &&
602 !filemap_release_folio(folio, GFP_KERNEL))
605 spin_lock(&mapping->host->i_lock);
606 xa_lock_irq(&mapping->i_pages);
607 if (folio_test_dirty(folio))
610 BUG_ON(folio_has_private(folio));
611 __filemap_remove_folio(folio, NULL);
612 xa_unlock_irq(&mapping->i_pages);
613 if (mapping_shrinkable(mapping))
614 inode_add_lru(mapping->host);
615 spin_unlock(&mapping->host->i_lock);
617 filemap_free_folio(mapping, folio);
620 xa_unlock_irq(&mapping->i_pages);
621 spin_unlock(&mapping->host->i_lock);
625 static int do_launder_folio(struct address_space *mapping, struct folio *folio)
627 if (!folio_test_dirty(folio))
629 if (folio->mapping != mapping || mapping->a_ops->launder_page == NULL)
631 return mapping->a_ops->launder_page(&folio->page);
635 * invalidate_inode_pages2_range - remove range of pages from an address_space
636 * @mapping: the address_space
637 * @start: the page offset 'from' which to invalidate
638 * @end: the page offset 'to' which to invalidate (inclusive)
640 * Any pages which are found to be mapped into pagetables are unmapped prior to
643 * Return: -EBUSY if any pages could not be invalidated.
645 int invalidate_inode_pages2_range(struct address_space *mapping,
646 pgoff_t start, pgoff_t end)
648 pgoff_t indices[PAGEVEC_SIZE];
649 struct folio_batch fbatch;
654 int did_range_unmap = 0;
656 if (mapping_empty(mapping))
659 folio_batch_init(&fbatch);
661 while (find_get_entries(mapping, index, end, &fbatch, indices)) {
662 for (i = 0; i < folio_batch_count(&fbatch); i++) {
663 struct folio *folio = fbatch.folios[i];
665 /* We rely upon deletion not changing folio->index */
668 if (xa_is_value(folio)) {
669 if (!invalidate_exceptional_entry2(mapping,
675 if (!did_range_unmap && folio_mapped(folio)) {
677 * If folio is mapped, before taking its lock,
678 * zap the rest of the file in one hit.
680 unmap_mapping_pages(mapping, index,
681 (1 + end - index), false);
686 VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
687 if (folio->mapping != mapping) {
691 folio_wait_writeback(folio);
693 if (folio_mapped(folio))
694 unmap_mapping_folio(folio);
695 BUG_ON(folio_mapped(folio));
697 ret2 = do_launder_folio(mapping, folio);
699 if (!invalidate_complete_folio2(mapping, folio))
706 folio_batch_remove_exceptionals(&fbatch);
707 folio_batch_release(&fbatch);
712 * For DAX we invalidate page tables after invalidating page cache. We
713 * could invalidate page tables while invalidating each entry however
714 * that would be expensive. And doing range unmapping before doesn't
715 * work as we have no cheap way to find whether page cache entry didn't
716 * get remapped later.
718 if (dax_mapping(mapping)) {
719 unmap_mapping_pages(mapping, start, end - start + 1, false);
723 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
726 * invalidate_inode_pages2 - remove all pages from an address_space
727 * @mapping: the address_space
729 * Any pages which are found to be mapped into pagetables are unmapped prior to
732 * Return: -EBUSY if any pages could not be invalidated.
734 int invalidate_inode_pages2(struct address_space *mapping)
736 return invalidate_inode_pages2_range(mapping, 0, -1);
738 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
741 * truncate_pagecache - unmap and remove pagecache that has been truncated
743 * @newsize: new file size
745 * inode's new i_size must already be written before truncate_pagecache
748 * This function should typically be called before the filesystem
749 * releases resources associated with the freed range (eg. deallocates
750 * blocks). This way, pagecache will always stay logically coherent
751 * with on-disk format, and the filesystem would not have to deal with
752 * situations such as writepage being called for a page that has already
753 * had its underlying blocks deallocated.
755 void truncate_pagecache(struct inode *inode, loff_t newsize)
757 struct address_space *mapping = inode->i_mapping;
758 loff_t holebegin = round_up(newsize, PAGE_SIZE);
761 * unmap_mapping_range is called twice, first simply for
762 * efficiency so that truncate_inode_pages does fewer
763 * single-page unmaps. However after this first call, and
764 * before truncate_inode_pages finishes, it is possible for
765 * private pages to be COWed, which remain after
766 * truncate_inode_pages finishes, hence the second
767 * unmap_mapping_range call must be made for correctness.
769 unmap_mapping_range(mapping, holebegin, 0, 1);
770 truncate_inode_pages(mapping, newsize);
771 unmap_mapping_range(mapping, holebegin, 0, 1);
773 EXPORT_SYMBOL(truncate_pagecache);
776 * truncate_setsize - update inode and pagecache for a new file size
778 * @newsize: new file size
780 * truncate_setsize updates i_size and performs pagecache truncation (if
781 * necessary) to @newsize. It will be typically be called from the filesystem's
782 * setattr function when ATTR_SIZE is passed in.
784 * Must be called with a lock serializing truncates and writes (generally
785 * i_rwsem but e.g. xfs uses a different lock) and before all filesystem
786 * specific block truncation has been performed.
788 void truncate_setsize(struct inode *inode, loff_t newsize)
790 loff_t oldsize = inode->i_size;
792 i_size_write(inode, newsize);
793 if (newsize > oldsize)
794 pagecache_isize_extended(inode, oldsize, newsize);
795 truncate_pagecache(inode, newsize);
797 EXPORT_SYMBOL(truncate_setsize);
800 * pagecache_isize_extended - update pagecache after extension of i_size
801 * @inode: inode for which i_size was extended
802 * @from: original inode size
803 * @to: new inode size
805 * Handle extension of inode size either caused by extending truncate or by
806 * write starting after current i_size. We mark the page straddling current
807 * i_size RO so that page_mkwrite() is called on the nearest write access to
808 * the page. This way filesystem can be sure that page_mkwrite() is called on
809 * the page before user writes to the page via mmap after the i_size has been
812 * The function must be called after i_size is updated so that page fault
813 * coming after we unlock the page will already see the new i_size.
814 * The function must be called while we still hold i_rwsem - this not only
815 * makes sure i_size is stable but also that userspace cannot observe new
816 * i_size value before we are prepared to store mmap writes at new inode size.
818 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
820 int bsize = i_blocksize(inode);
825 WARN_ON(to > inode->i_size);
827 if (from >= to || bsize == PAGE_SIZE)
829 /* Page straddling @from will not have any hole block created? */
830 rounded_from = round_up(from, bsize);
831 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
834 index = from >> PAGE_SHIFT;
835 page = find_lock_page(inode->i_mapping, index);
836 /* Page not cached? Nothing to do */
840 * See clear_page_dirty_for_io() for details why set_page_dirty()
843 if (page_mkclean(page))
844 set_page_dirty(page);
848 EXPORT_SYMBOL(pagecache_isize_extended);
851 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
853 * @lstart: offset of beginning of hole
854 * @lend: offset of last byte of hole
856 * This function should typically be called before the filesystem
857 * releases resources associated with the freed range (eg. deallocates
858 * blocks). This way, pagecache will always stay logically coherent
859 * with on-disk format, and the filesystem would not have to deal with
860 * situations such as writepage being called for a page that has already
861 * had its underlying blocks deallocated.
863 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
865 struct address_space *mapping = inode->i_mapping;
866 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
867 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
869 * This rounding is currently just for example: unmap_mapping_range
870 * expands its hole outwards, whereas we want it to contract the hole
871 * inwards. However, existing callers of truncate_pagecache_range are
872 * doing their own page rounding first. Note that unmap_mapping_range
873 * allows holelen 0 for all, and we allow lend -1 for end of file.
877 * Unlike in truncate_pagecache, unmap_mapping_range is called only
878 * once (before truncating pagecache), and without "even_cows" flag:
879 * hole-punching should not remove private COWed pages from the hole.
881 if ((u64)unmap_end > (u64)unmap_start)
882 unmap_mapping_range(mapping, unmap_start,
883 1 + unmap_end - unmap_start, 0);
884 truncate_inode_pages_range(mapping, lstart, lend);
886 EXPORT_SYMBOL(truncate_pagecache_range);