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/shmem_fs.h>
23 #include <linux/rmap.h>
27 * Regular page slots are stabilized by the page lock even without the tree
28 * itself locked. These unlocked entries need verification under the tree
31 static inline void __clear_shadow_entry(struct address_space
*mapping
,
32 pgoff_t index
, void *entry
)
34 XA_STATE(xas
, &mapping
->i_pages
, index
);
36 xas_set_update(&xas
, workingset_update_node
);
37 if (xas_load(&xas
) != entry
)
39 xas_store(&xas
, NULL
);
42 static void clear_shadow_entry(struct address_space
*mapping
, pgoff_t index
,
45 spin_lock(&mapping
->host
->i_lock
);
46 xa_lock_irq(&mapping
->i_pages
);
47 __clear_shadow_entry(mapping
, index
, entry
);
48 xa_unlock_irq(&mapping
->i_pages
);
49 if (mapping_shrinkable(mapping
))
50 inode_add_lru(mapping
->host
);
51 spin_unlock(&mapping
->host
->i_lock
);
55 * Unconditionally remove exceptional entries. Usually called from truncate
56 * path. Note that the folio_batch may be altered by this function by removing
57 * exceptional entries similar to what folio_batch_remove_exceptionals() does.
59 static void truncate_folio_batch_exceptionals(struct address_space
*mapping
,
60 struct folio_batch
*fbatch
, pgoff_t
*indices
)
65 /* Handled by shmem itself */
66 if (shmem_mapping(mapping
))
69 for (j
= 0; j
< folio_batch_count(fbatch
); j
++)
70 if (xa_is_value(fbatch
->folios
[j
]))
73 if (j
== folio_batch_count(fbatch
))
76 dax
= dax_mapping(mapping
);
78 spin_lock(&mapping
->host
->i_lock
);
79 xa_lock_irq(&mapping
->i_pages
);
82 for (i
= j
; i
< folio_batch_count(fbatch
); i
++) {
83 struct folio
*folio
= fbatch
->folios
[i
];
84 pgoff_t index
= indices
[i
];
86 if (!xa_is_value(folio
)) {
87 fbatch
->folios
[j
++] = folio
;
92 dax_delete_mapping_entry(mapping
, index
);
96 __clear_shadow_entry(mapping
, index
, folio
);
100 xa_unlock_irq(&mapping
->i_pages
);
101 if (mapping_shrinkable(mapping
))
102 inode_add_lru(mapping
->host
);
103 spin_unlock(&mapping
->host
->i_lock
);
109 * Invalidate exceptional entry if easily possible. This handles exceptional
110 * entries for invalidate_inode_pages().
112 static int invalidate_exceptional_entry(struct address_space
*mapping
,
113 pgoff_t index
, void *entry
)
115 /* Handled by shmem itself, or for DAX we do nothing. */
116 if (shmem_mapping(mapping
) || dax_mapping(mapping
))
118 clear_shadow_entry(mapping
, index
, entry
);
123 * Invalidate exceptional entry if clean. This handles exceptional entries for
124 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
126 static int invalidate_exceptional_entry2(struct address_space
*mapping
,
127 pgoff_t index
, void *entry
)
129 /* Handled by shmem itself */
130 if (shmem_mapping(mapping
))
132 if (dax_mapping(mapping
))
133 return dax_invalidate_mapping_entry_sync(mapping
, index
);
134 clear_shadow_entry(mapping
, index
, entry
);
139 * folio_invalidate - Invalidate part or all of a folio.
140 * @folio: The folio which is affected.
141 * @offset: start of the range to invalidate
142 * @length: length of the range to invalidate
144 * folio_invalidate() is called when all or part of the folio has become
145 * invalidated by a truncate operation.
147 * folio_invalidate() does not have to release all buffers, but it must
148 * ensure that no dirty buffer is left outside @offset and that no I/O
149 * is underway against any of the blocks which are outside the truncation
150 * point. Because the caller is about to free (and possibly reuse) those
153 void folio_invalidate(struct folio
*folio
, size_t offset
, size_t length
)
155 const struct address_space_operations
*aops
= folio
->mapping
->a_ops
;
157 if (aops
->invalidate_folio
)
158 aops
->invalidate_folio(folio
, offset
, length
);
160 EXPORT_SYMBOL_GPL(folio_invalidate
);
163 * If truncate cannot remove the fs-private metadata from the page, the page
164 * becomes orphaned. It will be left on the LRU and may even be mapped into
165 * user pagetables if we're racing with filemap_fault().
167 * We need to bail out if page->mapping is no longer equal to the original
168 * mapping. This happens a) when the VM reclaimed the page while we waited on
169 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
170 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
172 static void truncate_cleanup_folio(struct folio
*folio
)
174 if (folio_mapped(folio
))
175 unmap_mapping_folio(folio
);
177 if (folio_has_private(folio
))
178 folio_invalidate(folio
, 0, folio_size(folio
));
181 * Some filesystems seem to re-dirty the page even after
182 * the VM has canceled the dirty bit (eg ext3 journaling).
183 * Hence dirty accounting check is placed after invalidation.
185 folio_cancel_dirty(folio
);
186 folio_clear_mappedtodisk(folio
);
189 int truncate_inode_folio(struct address_space
*mapping
, struct folio
*folio
)
191 if (folio
->mapping
!= mapping
)
194 truncate_cleanup_folio(folio
);
195 filemap_remove_folio(folio
);
200 * Handle partial folios. The folio may be entirely within the
201 * range if a split has raced with us. If not, we zero the part of the
202 * folio that's within the [start, end] range, and then split the folio if
203 * it's large. split_page_range() will discard pages which now lie beyond
204 * i_size, and we rely on the caller to discard pages which lie within a
205 * newly created hole.
207 * Returns false if splitting failed so the caller can avoid
208 * discarding the entire folio which is stubbornly unsplit.
210 bool truncate_inode_partial_folio(struct folio
*folio
, loff_t start
, loff_t end
)
212 loff_t pos
= folio_pos(folio
);
213 unsigned int offset
, length
;
216 offset
= start
- pos
;
219 length
= folio_size(folio
);
220 if (pos
+ length
<= (u64
)end
)
221 length
= length
- offset
;
223 length
= end
+ 1 - pos
- offset
;
225 folio_wait_writeback(folio
);
226 if (length
== folio_size(folio
)) {
227 truncate_inode_folio(folio
->mapping
, folio
);
232 * We may be zeroing pages we're about to discard, but it avoids
233 * doing a complex calculation here, and then doing the zeroing
234 * anyway if the page split fails.
236 folio_zero_range(folio
, offset
, length
);
238 if (folio_has_private(folio
))
239 folio_invalidate(folio
, offset
, length
);
240 if (!folio_test_large(folio
))
242 if (split_folio(folio
) == 0)
244 if (folio_test_dirty(folio
))
246 truncate_inode_folio(folio
->mapping
, folio
);
251 * Used to get rid of pages on hardware memory corruption.
253 int generic_error_remove_folio(struct address_space
*mapping
,
259 * Only punch for normal data pages for now.
260 * Handling other types like directories would need more auditing.
262 if (!S_ISREG(mapping
->host
->i_mode
))
264 return truncate_inode_folio(mapping
, folio
);
266 EXPORT_SYMBOL(generic_error_remove_folio
);
269 * mapping_evict_folio() - Remove an unused folio from the page-cache.
270 * @mapping: The mapping this folio belongs to.
271 * @folio: The folio to remove.
273 * Safely remove one folio from the page cache.
274 * It only drops clean, unused folios.
276 * Context: Folio must be locked.
277 * Return: The number of pages successfully removed.
279 long mapping_evict_folio(struct address_space
*mapping
, struct folio
*folio
)
281 /* The page may have been truncated before it was locked */
284 if (folio_test_dirty(folio
) || folio_test_writeback(folio
))
286 /* The refcount will be elevated if any page in the folio is mapped */
287 if (folio_ref_count(folio
) >
288 folio_nr_pages(folio
) + folio_has_private(folio
) + 1)
290 if (!filemap_release_folio(folio
, 0))
293 return remove_mapping(mapping
, folio
);
297 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
298 * @mapping: mapping to truncate
299 * @lstart: offset from which to truncate
300 * @lend: offset to which to truncate (inclusive)
302 * Truncate the page cache, removing the pages that are between
303 * specified offsets (and zeroing out partial pages
304 * if lstart or lend + 1 is not page aligned).
306 * Truncate takes two passes - the first pass is nonblocking. It will not
307 * block on page locks and it will not block on writeback. The second pass
308 * will wait. This is to prevent as much IO as possible in the affected region.
309 * The first pass will remove most pages, so the search cost of the second pass
312 * We pass down the cache-hot hint to the page freeing code. Even if the
313 * mapping is large, it is probably the case that the final pages are the most
314 * recently touched, and freeing happens in ascending file offset order.
316 * Note that since ->invalidate_folio() accepts range to invalidate
317 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
318 * page aligned properly.
320 void truncate_inode_pages_range(struct address_space
*mapping
,
321 loff_t lstart
, loff_t lend
)
323 pgoff_t start
; /* inclusive */
324 pgoff_t end
; /* exclusive */
325 struct folio_batch fbatch
;
326 pgoff_t indices
[PAGEVEC_SIZE
];
332 if (mapping_empty(mapping
))
336 * 'start' and 'end' always covers the range of pages to be fully
337 * truncated. Partial pages are covered with 'partial_start' at the
338 * start of the range and 'partial_end' at the end of the range.
339 * Note that 'end' is exclusive while 'lend' is inclusive.
341 start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
344 * lend == -1 indicates end-of-file so we have to set 'end'
345 * to the highest possible pgoff_t and since the type is
346 * unsigned we're using -1.
350 end
= (lend
+ 1) >> PAGE_SHIFT
;
352 folio_batch_init(&fbatch
);
354 while (index
< end
&& find_lock_entries(mapping
, &index
, end
- 1,
356 truncate_folio_batch_exceptionals(mapping
, &fbatch
, indices
);
357 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++)
358 truncate_cleanup_folio(fbatch
.folios
[i
]);
359 delete_from_page_cache_batch(mapping
, &fbatch
);
360 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++)
361 folio_unlock(fbatch
.folios
[i
]);
362 folio_batch_release(&fbatch
);
366 same_folio
= (lstart
>> PAGE_SHIFT
) == (lend
>> PAGE_SHIFT
);
367 folio
= __filemap_get_folio(mapping
, lstart
>> PAGE_SHIFT
, FGP_LOCK
, 0);
368 if (!IS_ERR(folio
)) {
369 same_folio
= lend
< folio_pos(folio
) + folio_size(folio
);
370 if (!truncate_inode_partial_folio(folio
, lstart
, lend
)) {
371 start
= folio_next_index(folio
);
381 folio
= __filemap_get_folio(mapping
, lend
>> PAGE_SHIFT
,
383 if (!IS_ERR(folio
)) {
384 if (!truncate_inode_partial_folio(folio
, lstart
, lend
))
392 while (index
< end
) {
394 if (!find_get_entries(mapping
, &index
, end
- 1, &fbatch
,
396 /* If all gone from start onwards, we're done */
399 /* Otherwise restart to make sure all gone */
404 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
405 struct folio
*folio
= fbatch
.folios
[i
];
407 /* We rely upon deletion not changing page->index */
409 if (xa_is_value(folio
))
413 VM_BUG_ON_FOLIO(!folio_contains(folio
, indices
[i
]), folio
);
414 folio_wait_writeback(folio
);
415 truncate_inode_folio(mapping
, folio
);
418 truncate_folio_batch_exceptionals(mapping
, &fbatch
, indices
);
419 folio_batch_release(&fbatch
);
422 EXPORT_SYMBOL(truncate_inode_pages_range
);
425 * truncate_inode_pages - truncate *all* the pages from an offset
426 * @mapping: mapping to truncate
427 * @lstart: offset from which to truncate
429 * Called under (and serialised by) inode->i_rwsem and
430 * mapping->invalidate_lock.
432 * Note: When this function returns, there can be a page in the process of
433 * deletion (inside __filemap_remove_folio()) in the specified range. Thus
434 * mapping->nrpages can be non-zero when this function returns even after
435 * truncation of the whole mapping.
437 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
439 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
441 EXPORT_SYMBOL(truncate_inode_pages
);
444 * truncate_inode_pages_final - truncate *all* pages before inode dies
445 * @mapping: mapping to truncate
447 * Called under (and serialized by) inode->i_rwsem.
449 * Filesystems have to use this in the .evict_inode path to inform the
450 * VM that this is the final truncate and the inode is going away.
452 void truncate_inode_pages_final(struct address_space
*mapping
)
455 * Page reclaim can not participate in regular inode lifetime
456 * management (can't call iput()) and thus can race with the
457 * inode teardown. Tell it when the address space is exiting,
458 * so that it does not install eviction information after the
459 * final truncate has begun.
461 mapping_set_exiting(mapping
);
463 if (!mapping_empty(mapping
)) {
465 * As truncation uses a lockless tree lookup, cycle
466 * the tree lock to make sure any ongoing tree
467 * modification that does not see AS_EXITING is
468 * completed before starting the final truncate.
470 xa_lock_irq(&mapping
->i_pages
);
471 xa_unlock_irq(&mapping
->i_pages
);
474 truncate_inode_pages(mapping
, 0);
476 EXPORT_SYMBOL(truncate_inode_pages_final
);
479 * mapping_try_invalidate - Invalidate all the evictable folios of one inode
480 * @mapping: the address_space which holds the folios to invalidate
481 * @start: the offset 'from' which to invalidate
482 * @end: the offset 'to' which to invalidate (inclusive)
483 * @nr_failed: How many folio invalidations failed
485 * This function is similar to invalidate_mapping_pages(), except that it
486 * returns the number of folios which could not be evicted in @nr_failed.
488 unsigned long mapping_try_invalidate(struct address_space
*mapping
,
489 pgoff_t start
, pgoff_t end
, unsigned long *nr_failed
)
491 pgoff_t indices
[PAGEVEC_SIZE
];
492 struct folio_batch fbatch
;
493 pgoff_t index
= start
;
495 unsigned long count
= 0;
498 folio_batch_init(&fbatch
);
499 while (find_lock_entries(mapping
, &index
, end
, &fbatch
, indices
)) {
500 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
501 struct folio
*folio
= fbatch
.folios
[i
];
503 /* We rely upon deletion not changing folio->index */
505 if (xa_is_value(folio
)) {
506 count
+= invalidate_exceptional_entry(mapping
,
511 ret
= mapping_evict_folio(mapping
, folio
);
514 * Invalidation is a hint that the folio is no longer
515 * of interest and try to speed up its reclaim.
518 deactivate_file_folio(folio
);
519 /* Likely in the lru cache of a remote CPU */
525 folio_batch_remove_exceptionals(&fbatch
);
526 folio_batch_release(&fbatch
);
533 * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode
534 * @mapping: the address_space which holds the cache to invalidate
535 * @start: the offset 'from' which to invalidate
536 * @end: the offset 'to' which to invalidate (inclusive)
538 * This function removes pages that are clean, unmapped and unlocked,
539 * as well as shadow entries. It will not block on IO activity.
541 * If you want to remove all the pages of one inode, regardless of
542 * their use and writeback state, use truncate_inode_pages().
544 * Return: The number of indices that had their contents invalidated
546 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
547 pgoff_t start
, pgoff_t end
)
549 return mapping_try_invalidate(mapping
, start
, end
, NULL
);
551 EXPORT_SYMBOL(invalidate_mapping_pages
);
554 * This is like mapping_evict_folio(), except it ignores the folio's
555 * refcount. We do this because invalidate_inode_pages2() needs stronger
556 * invalidation guarantees, and cannot afford to leave folios behind because
557 * shrink_page_list() has a temp ref on them, or because they're transiently
558 * sitting in the folio_add_lru() caches.
560 static int invalidate_complete_folio2(struct address_space
*mapping
,
563 if (folio
->mapping
!= mapping
)
566 if (!filemap_release_folio(folio
, GFP_KERNEL
))
569 spin_lock(&mapping
->host
->i_lock
);
570 xa_lock_irq(&mapping
->i_pages
);
571 if (folio_test_dirty(folio
))
574 BUG_ON(folio_has_private(folio
));
575 __filemap_remove_folio(folio
, NULL
);
576 xa_unlock_irq(&mapping
->i_pages
);
577 if (mapping_shrinkable(mapping
))
578 inode_add_lru(mapping
->host
);
579 spin_unlock(&mapping
->host
->i_lock
);
581 filemap_free_folio(mapping
, folio
);
584 xa_unlock_irq(&mapping
->i_pages
);
585 spin_unlock(&mapping
->host
->i_lock
);
589 static int folio_launder(struct address_space
*mapping
, struct folio
*folio
)
591 if (!folio_test_dirty(folio
))
593 if (folio
->mapping
!= mapping
|| mapping
->a_ops
->launder_folio
== NULL
)
595 return mapping
->a_ops
->launder_folio(folio
);
599 * invalidate_inode_pages2_range - remove range of pages from an address_space
600 * @mapping: the address_space
601 * @start: the page offset 'from' which to invalidate
602 * @end: the page offset 'to' which to invalidate (inclusive)
604 * Any pages which are found to be mapped into pagetables are unmapped prior to
607 * Return: -EBUSY if any pages could not be invalidated.
609 int invalidate_inode_pages2_range(struct address_space
*mapping
,
610 pgoff_t start
, pgoff_t end
)
612 pgoff_t indices
[PAGEVEC_SIZE
];
613 struct folio_batch fbatch
;
618 int did_range_unmap
= 0;
620 if (mapping_empty(mapping
))
623 folio_batch_init(&fbatch
);
625 while (find_get_entries(mapping
, &index
, end
, &fbatch
, indices
)) {
626 for (i
= 0; i
< folio_batch_count(&fbatch
); i
++) {
627 struct folio
*folio
= fbatch
.folios
[i
];
629 /* We rely upon deletion not changing folio->index */
631 if (xa_is_value(folio
)) {
632 if (!invalidate_exceptional_entry2(mapping
,
638 if (!did_range_unmap
&& folio_mapped(folio
)) {
640 * If folio is mapped, before taking its lock,
641 * zap the rest of the file in one hit.
643 unmap_mapping_pages(mapping
, indices
[i
],
644 (1 + end
- indices
[i
]), false);
649 if (unlikely(folio
->mapping
!= mapping
)) {
653 VM_BUG_ON_FOLIO(!folio_contains(folio
, indices
[i
]), folio
);
654 folio_wait_writeback(folio
);
656 if (folio_mapped(folio
))
657 unmap_mapping_folio(folio
);
658 BUG_ON(folio_mapped(folio
));
660 ret2
= folio_launder(mapping
, folio
);
662 if (!invalidate_complete_folio2(mapping
, folio
))
669 folio_batch_remove_exceptionals(&fbatch
);
670 folio_batch_release(&fbatch
);
674 * For DAX we invalidate page tables after invalidating page cache. We
675 * could invalidate page tables while invalidating each entry however
676 * that would be expensive. And doing range unmapping before doesn't
677 * work as we have no cheap way to find whether page cache entry didn't
678 * get remapped later.
680 if (dax_mapping(mapping
)) {
681 unmap_mapping_pages(mapping
, start
, end
- start
+ 1, false);
685 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
688 * invalidate_inode_pages2 - remove all pages from an address_space
689 * @mapping: the address_space
691 * Any pages which are found to be mapped into pagetables are unmapped prior to
694 * Return: -EBUSY if any pages could not be invalidated.
696 int invalidate_inode_pages2(struct address_space
*mapping
)
698 return invalidate_inode_pages2_range(mapping
, 0, -1);
700 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);
703 * truncate_pagecache - unmap and remove pagecache that has been truncated
705 * @newsize: new file size
707 * inode's new i_size must already be written before truncate_pagecache
710 * This function should typically be called before the filesystem
711 * releases resources associated with the freed range (eg. deallocates
712 * blocks). This way, pagecache will always stay logically coherent
713 * with on-disk format, and the filesystem would not have to deal with
714 * situations such as writepage being called for a page that has already
715 * had its underlying blocks deallocated.
717 void truncate_pagecache(struct inode
*inode
, loff_t newsize
)
719 struct address_space
*mapping
= inode
->i_mapping
;
720 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
723 * unmap_mapping_range is called twice, first simply for
724 * efficiency so that truncate_inode_pages does fewer
725 * single-page unmaps. However after this first call, and
726 * before truncate_inode_pages finishes, it is possible for
727 * private pages to be COWed, which remain after
728 * truncate_inode_pages finishes, hence the second
729 * unmap_mapping_range call must be made for correctness.
731 unmap_mapping_range(mapping
, holebegin
, 0, 1);
732 truncate_inode_pages(mapping
, newsize
);
733 unmap_mapping_range(mapping
, holebegin
, 0, 1);
735 EXPORT_SYMBOL(truncate_pagecache
);
738 * truncate_setsize - update inode and pagecache for a new file size
740 * @newsize: new file size
742 * truncate_setsize updates i_size and performs pagecache truncation (if
743 * necessary) to @newsize. It will be typically be called from the filesystem's
744 * setattr function when ATTR_SIZE is passed in.
746 * Must be called with a lock serializing truncates and writes (generally
747 * i_rwsem but e.g. xfs uses a different lock) and before all filesystem
748 * specific block truncation has been performed.
750 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
752 loff_t oldsize
= inode
->i_size
;
754 i_size_write(inode
, newsize
);
755 if (newsize
> oldsize
)
756 pagecache_isize_extended(inode
, oldsize
, newsize
);
757 truncate_pagecache(inode
, newsize
);
759 EXPORT_SYMBOL(truncate_setsize
);
762 * pagecache_isize_extended - update pagecache after extension of i_size
763 * @inode: inode for which i_size was extended
764 * @from: original inode size
765 * @to: new inode size
767 * Handle extension of inode size either caused by extending truncate or by
768 * write starting after current i_size. We mark the page straddling current
769 * i_size RO so that page_mkwrite() is called on the nearest write access to
770 * the page. This way filesystem can be sure that page_mkwrite() is called on
771 * the page before user writes to the page via mmap after the i_size has been
774 * The function must be called after i_size is updated so that page fault
775 * coming after we unlock the page will already see the new i_size.
776 * The function must be called while we still hold i_rwsem - this not only
777 * makes sure i_size is stable but also that userspace cannot observe new
778 * i_size value before we are prepared to store mmap writes at new inode size.
780 void pagecache_isize_extended(struct inode
*inode
, loff_t from
, loff_t to
)
782 int bsize
= i_blocksize(inode
);
787 WARN_ON(to
> inode
->i_size
);
789 if (from
>= to
|| bsize
== PAGE_SIZE
)
791 /* Page straddling @from will not have any hole block created? */
792 rounded_from
= round_up(from
, bsize
);
793 if (to
<= rounded_from
|| !(rounded_from
& (PAGE_SIZE
- 1)))
796 index
= from
>> PAGE_SHIFT
;
797 page
= find_lock_page(inode
->i_mapping
, index
);
798 /* Page not cached? Nothing to do */
802 * See clear_page_dirty_for_io() for details why set_page_dirty()
805 if (page_mkclean(page
))
806 set_page_dirty(page
);
810 EXPORT_SYMBOL(pagecache_isize_extended
);
813 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
815 * @lstart: offset of beginning of hole
816 * @lend: offset of last byte of hole
818 * This function should typically be called before the filesystem
819 * releases resources associated with the freed range (eg. deallocates
820 * blocks). This way, pagecache will always stay logically coherent
821 * with on-disk format, and the filesystem would not have to deal with
822 * situations such as writepage being called for a page that has already
823 * had its underlying blocks deallocated.
825 void truncate_pagecache_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
827 struct address_space
*mapping
= inode
->i_mapping
;
828 loff_t unmap_start
= round_up(lstart
, PAGE_SIZE
);
829 loff_t unmap_end
= round_down(1 + lend
, PAGE_SIZE
) - 1;
831 * This rounding is currently just for example: unmap_mapping_range
832 * expands its hole outwards, whereas we want it to contract the hole
833 * inwards. However, existing callers of truncate_pagecache_range are
834 * doing their own page rounding first. Note that unmap_mapping_range
835 * allows holelen 0 for all, and we allow lend -1 for end of file.
839 * Unlike in truncate_pagecache, unmap_mapping_range is called only
840 * once (before truncating pagecache), and without "even_cows" flag:
841 * hole-punching should not remove private COWed pages from the hole.
843 if ((u64
)unmap_end
> (u64
)unmap_start
)
844 unmap_mapping_range(mapping
, unmap_start
,
845 1 + unmap_end
- unmap_start
, 0);
846 truncate_inode_pages_range(mapping
, lstart
, lend
);
848 EXPORT_SYMBOL(truncate_pagecache_range
);