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/cleancache.h>
26 #include <linux/rmap.h>
30 * Regular page slots are stabilized by the page lock even without the tree
31 * itself locked. These unlocked entries need verification under the tree
34 static inline void __clear_shadow_entry(struct address_space *mapping,
35 pgoff_t index, void *entry)
37 XA_STATE(xas, &mapping->i_pages, index);
39 xas_set_update(&xas, workingset_update_node);
40 if (xas_load(&xas) != entry)
42 xas_store(&xas, NULL);
43 mapping->nrexceptional--;
46 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
49 xa_lock_irq(&mapping->i_pages);
50 __clear_shadow_entry(mapping, index, entry);
51 xa_unlock_irq(&mapping->i_pages);
55 * Unconditionally remove exceptional entries. Usually called from truncate
56 * path. Note that the pagevec may be altered by this function by removing
57 * exceptional entries similar to what pagevec_remove_exceptionals does.
59 static void truncate_exceptional_pvec_entries(struct address_space *mapping,
60 struct pagevec *pvec, pgoff_t *indices,
66 /* Handled by shmem itself */
67 if (shmem_mapping(mapping))
70 for (j = 0; j < pagevec_count(pvec); j++)
71 if (xa_is_value(pvec->pages[j]))
74 if (j == pagevec_count(pvec))
77 dax = dax_mapping(mapping);
78 lock = !dax && indices[j] < end;
80 xa_lock_irq(&mapping->i_pages);
82 for (i = j; i < pagevec_count(pvec); i++) {
83 struct page *page = pvec->pages[i];
84 pgoff_t index = indices[i];
86 if (!xa_is_value(page)) {
87 pvec->pages[j++] = page;
95 dax_delete_mapping_entry(mapping, index);
99 __clear_shadow_entry(mapping, index, page);
103 xa_unlock_irq(&mapping->i_pages);
108 * Invalidate exceptional entry if easily possible. This handles exceptional
109 * entries for invalidate_inode_pages().
111 static int invalidate_exceptional_entry(struct address_space *mapping,
112 pgoff_t index, void *entry)
114 /* Handled by shmem itself, or for DAX we do nothing. */
115 if (shmem_mapping(mapping) || dax_mapping(mapping))
117 clear_shadow_entry(mapping, index, entry);
122 * Invalidate exceptional entry if clean. This handles exceptional entries for
123 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
125 static int invalidate_exceptional_entry2(struct address_space *mapping,
126 pgoff_t index, void *entry)
128 /* Handled by shmem itself */
129 if (shmem_mapping(mapping))
131 if (dax_mapping(mapping))
132 return dax_invalidate_mapping_entry_sync(mapping, index);
133 clear_shadow_entry(mapping, index, entry);
138 * do_invalidatepage - invalidate part or all of a page
139 * @page: the page which is affected
140 * @offset: start of the range to invalidate
141 * @length: length of the range to invalidate
143 * do_invalidatepage() is called when all or part of the page has become
144 * invalidated by a truncate operation.
146 * do_invalidatepage() does not have to release all buffers, but it must
147 * ensure that no dirty buffer is left outside @offset and that no I/O
148 * is underway against any of the blocks which are outside the truncation
149 * point. Because the caller is about to free (and possibly reuse) those
152 void do_invalidatepage(struct page *page, unsigned int offset,
155 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
157 invalidatepage = page->mapping->a_ops->invalidatepage;
160 invalidatepage = block_invalidatepage;
163 (*invalidatepage)(page, offset, length);
167 * If truncate cannot remove the fs-private metadata from the page, the page
168 * becomes orphaned. It will be left on the LRU and may even be mapped into
169 * user pagetables if we're racing with filemap_fault().
171 * We need to bail out if page->mapping is no longer equal to the original
172 * mapping. This happens a) when the VM reclaimed the page while we waited on
173 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
174 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
177 truncate_cleanup_page(struct address_space *mapping, struct page *page)
179 if (page_mapped(page)) {
180 unsigned int nr = thp_nr_pages(page);
181 unmap_mapping_pages(mapping, page->index, nr, false);
184 if (page_has_private(page))
185 do_invalidatepage(page, 0, thp_size(page));
188 * Some filesystems seem to re-dirty the page even after
189 * the VM has canceled the dirty bit (eg ext3 journaling).
190 * Hence dirty accounting check is placed after invalidation.
192 cancel_dirty_page(page);
193 ClearPageMappedToDisk(page);
197 * This is for invalidate_mapping_pages(). That function can be called at
198 * any time, and is not supposed to throw away dirty pages. But pages can
199 * be marked dirty at any time too, so use remove_mapping which safely
200 * discards clean, unused pages.
202 * Returns non-zero if the page was successfully invalidated.
205 invalidate_complete_page(struct address_space *mapping, struct page *page)
209 if (page->mapping != mapping)
212 if (page_has_private(page) && !try_to_release_page(page, 0))
215 ret = remove_mapping(mapping, page);
220 int truncate_inode_page(struct address_space *mapping, struct page *page)
222 VM_BUG_ON_PAGE(PageTail(page), page);
224 if (page->mapping != mapping)
227 truncate_cleanup_page(mapping, page);
228 delete_from_page_cache(page);
233 * Used to get rid of pages on hardware memory corruption.
235 int generic_error_remove_page(struct address_space *mapping, struct page *page)
240 * Only punch for normal data pages for now.
241 * Handling other types like directories would need more auditing.
243 if (!S_ISREG(mapping->host->i_mode))
245 return truncate_inode_page(mapping, page);
247 EXPORT_SYMBOL(generic_error_remove_page);
250 * Safely invalidate one page from its pagecache mapping.
251 * It only drops clean, unused pages. The page must be locked.
253 * Returns 1 if the page is successfully invalidated, otherwise 0.
255 int invalidate_inode_page(struct page *page)
257 struct address_space *mapping = page_mapping(page);
260 if (PageDirty(page) || PageWriteback(page))
262 if (page_mapped(page))
264 return invalidate_complete_page(mapping, page);
268 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
269 * @mapping: mapping to truncate
270 * @lstart: offset from which to truncate
271 * @lend: offset to which to truncate (inclusive)
273 * Truncate the page cache, removing the pages that are between
274 * specified offsets (and zeroing out partial pages
275 * if lstart or lend + 1 is not page aligned).
277 * Truncate takes two passes - the first pass is nonblocking. It will not
278 * block on page locks and it will not block on writeback. The second pass
279 * will wait. This is to prevent as much IO as possible in the affected region.
280 * The first pass will remove most pages, so the search cost of the second pass
283 * We pass down the cache-hot hint to the page freeing code. Even if the
284 * mapping is large, it is probably the case that the final pages are the most
285 * recently touched, and freeing happens in ascending file offset order.
287 * Note that since ->invalidatepage() accepts range to invalidate
288 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
289 * page aligned properly.
291 void truncate_inode_pages_range(struct address_space *mapping,
292 loff_t lstart, loff_t lend)
294 pgoff_t start; /* inclusive */
295 pgoff_t end; /* exclusive */
296 unsigned int partial_start; /* inclusive */
297 unsigned int partial_end; /* exclusive */
299 pgoff_t indices[PAGEVEC_SIZE];
303 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
306 /* Offsets within partial pages */
307 partial_start = lstart & (PAGE_SIZE - 1);
308 partial_end = (lend + 1) & (PAGE_SIZE - 1);
311 * 'start' and 'end' always covers the range of pages to be fully
312 * truncated. Partial pages are covered with 'partial_start' at the
313 * start of the range and 'partial_end' at the end of the range.
314 * Note that 'end' is exclusive while 'lend' is inclusive.
316 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
319 * lend == -1 indicates end-of-file so we have to set 'end'
320 * to the highest possible pgoff_t and since the type is
321 * unsigned we're using -1.
325 end = (lend + 1) >> PAGE_SHIFT;
329 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
330 min(end - index, (pgoff_t)PAGEVEC_SIZE),
333 * Pagevec array has exceptional entries and we may also fail
334 * to lock some pages. So we store pages that can be deleted
337 struct pagevec locked_pvec;
339 pagevec_init(&locked_pvec);
340 for (i = 0; i < pagevec_count(&pvec); i++) {
341 struct page *page = pvec.pages[i];
343 /* We rely upon deletion not changing page->index */
348 if (xa_is_value(page))
351 if (!trylock_page(page))
353 WARN_ON(page_to_index(page) != index);
354 if (PageWriteback(page)) {
358 if (page->mapping != mapping) {
362 pagevec_add(&locked_pvec, page);
364 for (i = 0; i < pagevec_count(&locked_pvec); i++)
365 truncate_cleanup_page(mapping, locked_pvec.pages[i]);
366 delete_from_page_cache_batch(mapping, &locked_pvec);
367 for (i = 0; i < pagevec_count(&locked_pvec); i++)
368 unlock_page(locked_pvec.pages[i]);
369 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
370 pagevec_release(&pvec);
375 struct page *page = find_lock_page(mapping, start - 1);
377 unsigned int top = PAGE_SIZE;
379 /* Truncation within a single page */
383 wait_on_page_writeback(page);
384 zero_user_segment(page, partial_start, top);
385 cleancache_invalidate_page(mapping, page);
386 if (page_has_private(page))
387 do_invalidatepage(page, partial_start,
388 top - partial_start);
394 struct page *page = find_lock_page(mapping, end);
396 wait_on_page_writeback(page);
397 zero_user_segment(page, 0, partial_end);
398 cleancache_invalidate_page(mapping, page);
399 if (page_has_private(page))
400 do_invalidatepage(page, 0,
407 * If the truncation happened within a single page no pages
408 * will be released, just zeroed, so we can bail out now.
416 if (!pagevec_lookup_entries(&pvec, mapping, index,
417 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
418 /* If all gone from start onwards, we're done */
421 /* Otherwise restart to make sure all gone */
425 if (index == start && indices[0] >= end) {
426 /* All gone out of hole to be punched, we're done */
427 pagevec_remove_exceptionals(&pvec);
428 pagevec_release(&pvec);
432 for (i = 0; i < pagevec_count(&pvec); i++) {
433 struct page *page = pvec.pages[i];
435 /* We rely upon deletion not changing page->index */
438 /* Restart punch to make sure all gone */
443 if (xa_is_value(page))
447 WARN_ON(page_to_index(page) != index);
448 wait_on_page_writeback(page);
449 truncate_inode_page(mapping, page);
452 truncate_exceptional_pvec_entries(mapping, &pvec, indices, end);
453 pagevec_release(&pvec);
458 cleancache_invalidate_inode(mapping);
460 EXPORT_SYMBOL(truncate_inode_pages_range);
463 * truncate_inode_pages - truncate *all* the pages from an offset
464 * @mapping: mapping to truncate
465 * @lstart: offset from which to truncate
467 * Called under (and serialised by) inode->i_mutex.
469 * Note: When this function returns, there can be a page in the process of
470 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
471 * mapping->nrpages can be non-zero when this function returns even after
472 * truncation of the whole mapping.
474 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
476 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
478 EXPORT_SYMBOL(truncate_inode_pages);
481 * truncate_inode_pages_final - truncate *all* pages before inode dies
482 * @mapping: mapping to truncate
484 * Called under (and serialized by) inode->i_mutex.
486 * Filesystems have to use this in the .evict_inode path to inform the
487 * VM that this is the final truncate and the inode is going away.
489 void truncate_inode_pages_final(struct address_space *mapping)
491 unsigned long nrexceptional;
492 unsigned long nrpages;
495 * Page reclaim can not participate in regular inode lifetime
496 * management (can't call iput()) and thus can race with the
497 * inode teardown. Tell it when the address space is exiting,
498 * so that it does not install eviction information after the
499 * final truncate has begun.
501 mapping_set_exiting(mapping);
504 * When reclaim installs eviction entries, it increases
505 * nrexceptional first, then decreases nrpages. Make sure we see
506 * this in the right order or we might miss an entry.
508 nrpages = mapping->nrpages;
510 nrexceptional = mapping->nrexceptional;
512 if (nrpages || nrexceptional) {
514 * As truncation uses a lockless tree lookup, cycle
515 * the tree lock to make sure any ongoing tree
516 * modification that does not see AS_EXITING is
517 * completed before starting the final truncate.
519 xa_lock_irq(&mapping->i_pages);
520 xa_unlock_irq(&mapping->i_pages);
524 * Cleancache needs notification even if there are no pages or shadow
527 truncate_inode_pages(mapping, 0);
529 EXPORT_SYMBOL(truncate_inode_pages_final);
531 static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
532 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
534 pgoff_t indices[PAGEVEC_SIZE];
536 pgoff_t index = start;
538 unsigned long count = 0;
542 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
543 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
545 for (i = 0; i < pagevec_count(&pvec); i++) {
546 struct page *page = pvec.pages[i];
548 /* We rely upon deletion not changing page->index */
553 if (xa_is_value(page)) {
554 invalidate_exceptional_entry(mapping, index,
559 if (!trylock_page(page))
562 WARN_ON(page_to_index(page) != index);
564 /* Middle of THP: skip */
565 if (PageTransTail(page)) {
568 } else if (PageTransHuge(page)) {
569 index += HPAGE_PMD_NR - 1;
570 i += HPAGE_PMD_NR - 1;
572 * 'end' is in the middle of THP. Don't
573 * invalidate the page as the part outside of
574 * 'end' could be still useful.
581 /* Take a pin outside pagevec */
585 * Drop extra pins before trying to invalidate
588 pagevec_remove_exceptionals(&pvec);
589 pagevec_release(&pvec);
592 ret = invalidate_inode_page(page);
595 * Invalidation is a hint that the page is no longer
596 * of interest and try to speed up its reclaim.
599 deactivate_file_page(page);
600 /* It is likely on the pagevec of a remote CPU */
605 if (PageTransHuge(page))
609 pagevec_remove_exceptionals(&pvec);
610 pagevec_release(&pvec);
618 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
619 * @mapping: the address_space which holds the pages to invalidate
620 * @start: the offset 'from' which to invalidate
621 * @end: the offset 'to' which to invalidate (inclusive)
623 * This function only removes the unlocked pages, if you want to
624 * remove all the pages of one inode, you must call truncate_inode_pages.
626 * invalidate_mapping_pages() will not block on IO activity. It will not
627 * invalidate pages which are dirty, locked, under writeback or mapped into
630 * Return: the number of the pages that were invalidated
632 unsigned long invalidate_mapping_pages(struct address_space *mapping,
633 pgoff_t start, pgoff_t end)
635 return __invalidate_mapping_pages(mapping, start, end, NULL);
637 EXPORT_SYMBOL(invalidate_mapping_pages);
640 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
641 * @mapping: the address_space which holds the pages to invalidate
642 * @start: the offset 'from' which to invalidate
643 * @end: the offset 'to' which to invalidate (inclusive)
644 * @nr_pagevec: invalidate failed page number for caller
646 * This helper is similar to invalidate_mapping_pages(), except that it accounts
647 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
648 * will be used by the caller.
650 void invalidate_mapping_pagevec(struct address_space *mapping,
651 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
653 __invalidate_mapping_pages(mapping, start, end, nr_pagevec);
657 * This is like invalidate_complete_page(), except it ignores the page's
658 * refcount. We do this because invalidate_inode_pages2() needs stronger
659 * invalidation guarantees, and cannot afford to leave pages behind because
660 * shrink_page_list() has a temp ref on them, or because they're transiently
661 * sitting in the lru_cache_add() pagevecs.
664 invalidate_complete_page2(struct address_space *mapping, struct page *page)
668 if (page->mapping != mapping)
671 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
674 xa_lock_irqsave(&mapping->i_pages, flags);
678 BUG_ON(page_has_private(page));
679 __delete_from_page_cache(page, NULL);
680 xa_unlock_irqrestore(&mapping->i_pages, flags);
682 if (mapping->a_ops->freepage)
683 mapping->a_ops->freepage(page);
685 put_page(page); /* pagecache ref */
688 xa_unlock_irqrestore(&mapping->i_pages, flags);
692 static int do_launder_page(struct address_space *mapping, struct page *page)
694 if (!PageDirty(page))
696 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
698 return mapping->a_ops->launder_page(page);
702 * invalidate_inode_pages2_range - remove range of pages from an address_space
703 * @mapping: the address_space
704 * @start: the page offset 'from' which to invalidate
705 * @end: the page offset 'to' which to invalidate (inclusive)
707 * Any pages which are found to be mapped into pagetables are unmapped prior to
710 * Return: -EBUSY if any pages could not be invalidated.
712 int invalidate_inode_pages2_range(struct address_space *mapping,
713 pgoff_t start, pgoff_t end)
715 pgoff_t indices[PAGEVEC_SIZE];
721 int did_range_unmap = 0;
723 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
728 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
729 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
731 for (i = 0; i < pagevec_count(&pvec); i++) {
732 struct page *page = pvec.pages[i];
734 /* We rely upon deletion not changing page->index */
739 if (xa_is_value(page)) {
740 if (!invalidate_exceptional_entry2(mapping,
747 WARN_ON(page_to_index(page) != index);
748 if (page->mapping != mapping) {
752 wait_on_page_writeback(page);
753 if (page_mapped(page)) {
754 if (!did_range_unmap) {
756 * Zap the rest of the file in one hit.
758 unmap_mapping_pages(mapping, index,
759 (1 + end - index), false);
765 unmap_mapping_pages(mapping, index,
769 BUG_ON(page_mapped(page));
770 ret2 = do_launder_page(mapping, page);
772 if (!invalidate_complete_page2(mapping, page))
779 pagevec_remove_exceptionals(&pvec);
780 pagevec_release(&pvec);
785 * For DAX we invalidate page tables after invalidating page cache. We
786 * could invalidate page tables while invalidating each entry however
787 * that would be expensive. And doing range unmapping before doesn't
788 * work as we have no cheap way to find whether page cache entry didn't
789 * get remapped later.
791 if (dax_mapping(mapping)) {
792 unmap_mapping_pages(mapping, start, end - start + 1, false);
795 cleancache_invalidate_inode(mapping);
798 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
801 * invalidate_inode_pages2 - remove all pages from an address_space
802 * @mapping: the address_space
804 * Any pages which are found to be mapped into pagetables are unmapped prior to
807 * Return: -EBUSY if any pages could not be invalidated.
809 int invalidate_inode_pages2(struct address_space *mapping)
811 return invalidate_inode_pages2_range(mapping, 0, -1);
813 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
816 * truncate_pagecache - unmap and remove pagecache that has been truncated
818 * @newsize: new file size
820 * inode's new i_size must already be written before truncate_pagecache
823 * This function should typically be called before the filesystem
824 * releases resources associated with the freed range (eg. deallocates
825 * blocks). This way, pagecache will always stay logically coherent
826 * with on-disk format, and the filesystem would not have to deal with
827 * situations such as writepage being called for a page that has already
828 * had its underlying blocks deallocated.
830 void truncate_pagecache(struct inode *inode, loff_t newsize)
832 struct address_space *mapping = inode->i_mapping;
833 loff_t holebegin = round_up(newsize, PAGE_SIZE);
836 * unmap_mapping_range is called twice, first simply for
837 * efficiency so that truncate_inode_pages does fewer
838 * single-page unmaps. However after this first call, and
839 * before truncate_inode_pages finishes, it is possible for
840 * private pages to be COWed, which remain after
841 * truncate_inode_pages finishes, hence the second
842 * unmap_mapping_range call must be made for correctness.
844 unmap_mapping_range(mapping, holebegin, 0, 1);
845 truncate_inode_pages(mapping, newsize);
846 unmap_mapping_range(mapping, holebegin, 0, 1);
848 EXPORT_SYMBOL(truncate_pagecache);
851 * truncate_setsize - update inode and pagecache for a new file size
853 * @newsize: new file size
855 * truncate_setsize updates i_size and performs pagecache truncation (if
856 * necessary) to @newsize. It will be typically be called from the filesystem's
857 * setattr function when ATTR_SIZE is passed in.
859 * Must be called with a lock serializing truncates and writes (generally
860 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
861 * specific block truncation has been performed.
863 void truncate_setsize(struct inode *inode, loff_t newsize)
865 loff_t oldsize = inode->i_size;
867 i_size_write(inode, newsize);
868 if (newsize > oldsize)
869 pagecache_isize_extended(inode, oldsize, newsize);
870 truncate_pagecache(inode, newsize);
872 EXPORT_SYMBOL(truncate_setsize);
875 * pagecache_isize_extended - update pagecache after extension of i_size
876 * @inode: inode for which i_size was extended
877 * @from: original inode size
878 * @to: new inode size
880 * Handle extension of inode size either caused by extending truncate or by
881 * write starting after current i_size. We mark the page straddling current
882 * i_size RO so that page_mkwrite() is called on the nearest write access to
883 * the page. This way filesystem can be sure that page_mkwrite() is called on
884 * the page before user writes to the page via mmap after the i_size has been
887 * The function must be called after i_size is updated so that page fault
888 * coming after we unlock the page will already see the new i_size.
889 * The function must be called while we still hold i_mutex - this not only
890 * makes sure i_size is stable but also that userspace cannot observe new
891 * i_size value before we are prepared to store mmap writes at new inode size.
893 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
895 int bsize = i_blocksize(inode);
900 WARN_ON(to > inode->i_size);
902 if (from >= to || bsize == PAGE_SIZE)
904 /* Page straddling @from will not have any hole block created? */
905 rounded_from = round_up(from, bsize);
906 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
909 index = from >> PAGE_SHIFT;
910 page = find_lock_page(inode->i_mapping, index);
911 /* Page not cached? Nothing to do */
915 * See clear_page_dirty_for_io() for details why set_page_dirty()
918 if (page_mkclean(page))
919 set_page_dirty(page);
923 EXPORT_SYMBOL(pagecache_isize_extended);
926 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
928 * @lstart: offset of beginning of hole
929 * @lend: offset of last byte of hole
931 * This function should typically be called before the filesystem
932 * releases resources associated with the freed range (eg. deallocates
933 * blocks). This way, pagecache will always stay logically coherent
934 * with on-disk format, and the filesystem would not have to deal with
935 * situations such as writepage being called for a page that has already
936 * had its underlying blocks deallocated.
938 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
940 struct address_space *mapping = inode->i_mapping;
941 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
942 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
944 * This rounding is currently just for example: unmap_mapping_range
945 * expands its hole outwards, whereas we want it to contract the hole
946 * inwards. However, existing callers of truncate_pagecache_range are
947 * doing their own page rounding first. Note that unmap_mapping_range
948 * allows holelen 0 for all, and we allow lend -1 for end of file.
952 * Unlike in truncate_pagecache, unmap_mapping_range is called only
953 * once (before truncating pagecache), and without "even_cows" flag:
954 * hole-punching should not remove private COWed pages from the hole.
956 if ((u64)unmap_end > (u64)unmap_start)
957 unmap_mapping_range(mapping, unmap_start,
958 1 + unmap_end - unmap_start, 0);
959 truncate_inode_pages_range(mapping, lstart, lend);
961 EXPORT_SYMBOL(truncate_pagecache_range);