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 bale 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.
176 static void truncate_cleanup_page(struct page
*page
)
178 if (page_mapped(page
))
179 unmap_mapping_page(page
);
181 if (page_has_private(page
))
182 do_invalidatepage(page
, 0, PAGE_SIZE
);
185 * Some filesystems seem to re-dirty the page even after
186 * the VM has canceled the dirty bit (eg ext3 journaling).
187 * Hence dirty accounting check is placed after invalidation.
189 cancel_dirty_page(page
);
190 ClearPageMappedToDisk(page
);
194 * This is for invalidate_mapping_pages(). That function can be called at
195 * any time, and is not supposed to throw away dirty pages. But pages can
196 * be marked dirty at any time too, so use remove_mapping which safely
197 * discards clean, unused pages.
199 * Returns non-zero if the page was successfully invalidated.
202 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
206 if (page
->mapping
!= mapping
)
209 if (page_has_private(page
) && !try_to_release_page(page
, 0))
212 ret
= remove_mapping(mapping
, page
);
217 int truncate_inode_page(struct address_space
*mapping
, struct page
*page
)
219 VM_BUG_ON_PAGE(PageTail(page
), page
);
221 if (page
->mapping
!= mapping
)
224 truncate_cleanup_page(page
);
225 delete_from_page_cache(page
);
230 * Used to get rid of pages on hardware memory corruption.
232 int generic_error_remove_page(struct address_space
*mapping
, struct page
*page
)
237 * Only punch for normal data pages for now.
238 * Handling other types like directories would need more auditing.
240 if (!S_ISREG(mapping
->host
->i_mode
))
242 return truncate_inode_page(mapping
, page
);
244 EXPORT_SYMBOL(generic_error_remove_page
);
247 * Safely invalidate one page from its pagecache mapping.
248 * It only drops clean, unused pages. The page must be locked.
250 * Returns 1 if the page is successfully invalidated, otherwise 0.
252 int invalidate_inode_page(struct page
*page
)
254 struct address_space
*mapping
= page_mapping(page
);
257 if (PageDirty(page
) || PageWriteback(page
))
259 if (page_mapped(page
))
261 return invalidate_complete_page(mapping
, page
);
265 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
266 * @mapping: mapping to truncate
267 * @lstart: offset from which to truncate
268 * @lend: offset to which to truncate (inclusive)
270 * Truncate the page cache, removing the pages that are between
271 * specified offsets (and zeroing out partial pages
272 * if lstart or lend + 1 is not page aligned).
274 * Truncate takes two passes - the first pass is nonblocking. It will not
275 * block on page locks and it will not block on writeback. The second pass
276 * will wait. This is to prevent as much IO as possible in the affected region.
277 * The first pass will remove most pages, so the search cost of the second pass
280 * We pass down the cache-hot hint to the page freeing code. Even if the
281 * mapping is large, it is probably the case that the final pages are the most
282 * recently touched, and freeing happens in ascending file offset order.
284 * Note that since ->invalidatepage() accepts range to invalidate
285 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
286 * page aligned properly.
288 void truncate_inode_pages_range(struct address_space
*mapping
,
289 loff_t lstart
, loff_t lend
)
291 pgoff_t start
; /* inclusive */
292 pgoff_t end
; /* exclusive */
293 unsigned int partial_start
; /* inclusive */
294 unsigned int partial_end
; /* exclusive */
296 pgoff_t indices
[PAGEVEC_SIZE
];
300 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
303 /* Offsets within partial pages */
304 partial_start
= lstart
& (PAGE_SIZE
- 1);
305 partial_end
= (lend
+ 1) & (PAGE_SIZE
- 1);
308 * 'start' and 'end' always covers the range of pages to be fully
309 * truncated. Partial pages are covered with 'partial_start' at the
310 * start of the range and 'partial_end' at the end of the range.
311 * Note that 'end' is exclusive while 'lend' is inclusive.
313 start
= (lstart
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
316 * lend == -1 indicates end-of-file so we have to set 'end'
317 * to the highest possible pgoff_t and since the type is
318 * unsigned we're using -1.
322 end
= (lend
+ 1) >> PAGE_SHIFT
;
326 while (index
< end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
327 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
330 * Pagevec array has exceptional entries and we may also fail
331 * to lock some pages. So we store pages that can be deleted
334 struct pagevec locked_pvec
;
336 pagevec_init(&locked_pvec
);
337 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
338 struct page
*page
= pvec
.pages
[i
];
340 /* We rely upon deletion not changing page->index */
345 if (xa_is_value(page
))
348 if (!trylock_page(page
))
350 WARN_ON(page_to_index(page
) != index
);
351 if (PageWriteback(page
)) {
355 if (page
->mapping
!= mapping
) {
359 pagevec_add(&locked_pvec
, page
);
361 for (i
= 0; i
< pagevec_count(&locked_pvec
); i
++)
362 truncate_cleanup_page(locked_pvec
.pages
[i
]);
363 delete_from_page_cache_batch(mapping
, &locked_pvec
);
364 for (i
= 0; i
< pagevec_count(&locked_pvec
); i
++)
365 unlock_page(locked_pvec
.pages
[i
]);
366 truncate_exceptional_pvec_entries(mapping
, &pvec
, indices
, end
);
367 pagevec_release(&pvec
);
372 struct page
*page
= find_lock_page(mapping
, start
- 1);
374 unsigned int top
= PAGE_SIZE
;
376 /* Truncation within a single page */
380 wait_on_page_writeback(page
);
381 zero_user_segment(page
, partial_start
, top
);
382 cleancache_invalidate_page(mapping
, page
);
383 if (page_has_private(page
))
384 do_invalidatepage(page
, partial_start
,
385 top
- partial_start
);
391 struct page
*page
= find_lock_page(mapping
, end
);
393 wait_on_page_writeback(page
);
394 zero_user_segment(page
, 0, partial_end
);
395 cleancache_invalidate_page(mapping
, page
);
396 if (page_has_private(page
))
397 do_invalidatepage(page
, 0,
404 * If the truncation happened within a single page no pages
405 * will be released, just zeroed, so we can bail out now.
413 if (!pagevec_lookup_entries(&pvec
, mapping
, index
,
414 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
), indices
)) {
415 /* If all gone from start onwards, we're done */
418 /* Otherwise restart to make sure all gone */
422 if (index
== start
&& indices
[0] >= end
) {
423 /* All gone out of hole to be punched, we're done */
424 pagevec_remove_exceptionals(&pvec
);
425 pagevec_release(&pvec
);
429 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
430 struct page
*page
= pvec
.pages
[i
];
432 /* We rely upon deletion not changing page->index */
435 /* Restart punch to make sure all gone */
440 if (xa_is_value(page
))
444 WARN_ON(page_to_index(page
) != index
);
445 wait_on_page_writeback(page
);
446 truncate_inode_page(mapping
, page
);
449 truncate_exceptional_pvec_entries(mapping
, &pvec
, indices
, end
);
450 pagevec_release(&pvec
);
455 cleancache_invalidate_inode(mapping
);
457 EXPORT_SYMBOL(truncate_inode_pages_range
);
460 * truncate_inode_pages - truncate *all* the pages from an offset
461 * @mapping: mapping to truncate
462 * @lstart: offset from which to truncate
464 * Called under (and serialised by) inode->i_mutex.
466 * Note: When this function returns, there can be a page in the process of
467 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
468 * mapping->nrpages can be non-zero when this function returns even after
469 * truncation of the whole mapping.
471 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
473 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
475 EXPORT_SYMBOL(truncate_inode_pages
);
478 * truncate_inode_pages_final - truncate *all* pages before inode dies
479 * @mapping: mapping to truncate
481 * Called under (and serialized by) inode->i_mutex.
483 * Filesystems have to use this in the .evict_inode path to inform the
484 * VM that this is the final truncate and the inode is going away.
486 void truncate_inode_pages_final(struct address_space
*mapping
)
488 unsigned long nrexceptional
;
489 unsigned long nrpages
;
492 * Page reclaim can not participate in regular inode lifetime
493 * management (can't call iput()) and thus can race with the
494 * inode teardown. Tell it when the address space is exiting,
495 * so that it does not install eviction information after the
496 * final truncate has begun.
498 mapping_set_exiting(mapping
);
501 * When reclaim installs eviction entries, it increases
502 * nrexceptional first, then decreases nrpages. Make sure we see
503 * this in the right order or we might miss an entry.
505 nrpages
= mapping
->nrpages
;
507 nrexceptional
= mapping
->nrexceptional
;
509 if (nrpages
|| nrexceptional
) {
511 * As truncation uses a lockless tree lookup, cycle
512 * the tree lock to make sure any ongoing tree
513 * modification that does not see AS_EXITING is
514 * completed before starting the final truncate.
516 xa_lock_irq(&mapping
->i_pages
);
517 xa_unlock_irq(&mapping
->i_pages
);
521 * Cleancache needs notification even if there are no pages or shadow
524 truncate_inode_pages(mapping
, 0);
526 EXPORT_SYMBOL(truncate_inode_pages_final
);
529 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
530 * @mapping: the address_space which holds the pages to invalidate
531 * @start: the offset 'from' which to invalidate
532 * @end: the offset 'to' which to invalidate (inclusive)
534 * This function only removes the unlocked pages, if you want to
535 * remove all the pages of one inode, you must call truncate_inode_pages.
537 * invalidate_mapping_pages() will not block on IO activity. It will not
538 * invalidate pages which are dirty, locked, under writeback or mapped into
541 * Return: the number of the pages that were invalidated
543 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
544 pgoff_t start
, pgoff_t end
)
546 pgoff_t indices
[PAGEVEC_SIZE
];
548 pgoff_t index
= start
;
550 unsigned long count
= 0;
554 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
555 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
557 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
558 struct page
*page
= pvec
.pages
[i
];
560 /* We rely upon deletion not changing page->index */
565 if (xa_is_value(page
)) {
566 invalidate_exceptional_entry(mapping
, index
,
571 if (!trylock_page(page
))
574 WARN_ON(page_to_index(page
) != index
);
576 /* Middle of THP: skip */
577 if (PageTransTail(page
)) {
580 } else if (PageTransHuge(page
)) {
581 index
+= HPAGE_PMD_NR
- 1;
582 i
+= HPAGE_PMD_NR
- 1;
584 * 'end' is in the middle of THP. Don't
585 * invalidate the page as the part outside of
586 * 'end' could be still useful.
593 /* Take a pin outside pagevec */
597 * Drop extra pins before trying to invalidate
600 pagevec_remove_exceptionals(&pvec
);
601 pagevec_release(&pvec
);
604 ret
= invalidate_inode_page(page
);
607 * Invalidation is a hint that the page is no longer
608 * of interest and try to speed up its reclaim.
611 deactivate_file_page(page
);
612 if (PageTransHuge(page
))
616 pagevec_remove_exceptionals(&pvec
);
617 pagevec_release(&pvec
);
623 EXPORT_SYMBOL(invalidate_mapping_pages
);
626 * This is like invalidate_complete_page(), except it ignores the page's
627 * refcount. We do this because invalidate_inode_pages2() needs stronger
628 * invalidation guarantees, and cannot afford to leave pages behind because
629 * shrink_page_list() has a temp ref on them, or because they're transiently
630 * sitting in the lru_cache_add() pagevecs.
633 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
637 if (page
->mapping
!= mapping
)
640 if (page_has_private(page
) && !try_to_release_page(page
, GFP_KERNEL
))
643 xa_lock_irqsave(&mapping
->i_pages
, flags
);
647 BUG_ON(page_has_private(page
));
648 __delete_from_page_cache(page
, NULL
);
649 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
651 if (mapping
->a_ops
->freepage
)
652 mapping
->a_ops
->freepage(page
);
654 put_page(page
); /* pagecache ref */
657 xa_unlock_irqrestore(&mapping
->i_pages
, flags
);
661 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
663 if (!PageDirty(page
))
665 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
667 return mapping
->a_ops
->launder_page(page
);
671 * invalidate_inode_pages2_range - remove range of pages from an address_space
672 * @mapping: the address_space
673 * @start: the page offset 'from' which to invalidate
674 * @end: the page offset 'to' which to invalidate (inclusive)
676 * Any pages which are found to be mapped into pagetables are unmapped prior to
679 * Return: -EBUSY if any pages could not be invalidated.
681 int invalidate_inode_pages2_range(struct address_space
*mapping
,
682 pgoff_t start
, pgoff_t end
)
684 pgoff_t indices
[PAGEVEC_SIZE
];
690 int did_range_unmap
= 0;
692 if (mapping
->nrpages
== 0 && mapping
->nrexceptional
== 0)
697 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
698 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
700 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
701 struct page
*page
= pvec
.pages
[i
];
703 /* We rely upon deletion not changing page->index */
708 if (xa_is_value(page
)) {
709 if (!invalidate_exceptional_entry2(mapping
,
715 if (!did_range_unmap
&& page_mapped(page
)) {
717 * If page is mapped, before taking its lock,
718 * zap the rest of the file in one hit.
720 unmap_mapping_pages(mapping
, index
,
721 (1 + end
- index
), false);
726 WARN_ON(page_to_index(page
) != index
);
727 if (page
->mapping
!= mapping
) {
731 wait_on_page_writeback(page
);
733 if (page_mapped(page
))
734 unmap_mapping_page(page
);
735 BUG_ON(page_mapped(page
));
737 ret2
= do_launder_page(mapping
, page
);
739 if (!invalidate_complete_page2(mapping
, page
))
746 pagevec_remove_exceptionals(&pvec
);
747 pagevec_release(&pvec
);
752 * For DAX we invalidate page tables after invalidating page cache. We
753 * could invalidate page tables while invalidating each entry however
754 * that would be expensive. And doing range unmapping before doesn't
755 * work as we have no cheap way to find whether page cache entry didn't
756 * get remapped later.
758 if (dax_mapping(mapping
)) {
759 unmap_mapping_pages(mapping
, start
, end
- start
+ 1, false);
762 cleancache_invalidate_inode(mapping
);
765 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
768 * invalidate_inode_pages2 - remove all pages from an address_space
769 * @mapping: the address_space
771 * Any pages which are found to be mapped into pagetables are unmapped prior to
774 * Return: -EBUSY if any pages could not be invalidated.
776 int invalidate_inode_pages2(struct address_space
*mapping
)
778 return invalidate_inode_pages2_range(mapping
, 0, -1);
780 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);
783 * truncate_pagecache - unmap and remove pagecache that has been truncated
785 * @newsize: new file size
787 * inode's new i_size must already be written before truncate_pagecache
790 * This function should typically be called before the filesystem
791 * releases resources associated with the freed range (eg. deallocates
792 * blocks). This way, pagecache will always stay logically coherent
793 * with on-disk format, and the filesystem would not have to deal with
794 * situations such as writepage being called for a page that has already
795 * had its underlying blocks deallocated.
797 void truncate_pagecache(struct inode
*inode
, loff_t newsize
)
799 struct address_space
*mapping
= inode
->i_mapping
;
800 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
803 * unmap_mapping_range is called twice, first simply for
804 * efficiency so that truncate_inode_pages does fewer
805 * single-page unmaps. However after this first call, and
806 * before truncate_inode_pages finishes, it is possible for
807 * private pages to be COWed, which remain after
808 * truncate_inode_pages finishes, hence the second
809 * unmap_mapping_range call must be made for correctness.
811 unmap_mapping_range(mapping
, holebegin
, 0, 1);
812 truncate_inode_pages(mapping
, newsize
);
813 unmap_mapping_range(mapping
, holebegin
, 0, 1);
815 EXPORT_SYMBOL(truncate_pagecache
);
818 * truncate_setsize - update inode and pagecache for a new file size
820 * @newsize: new file size
822 * truncate_setsize updates i_size and performs pagecache truncation (if
823 * necessary) to @newsize. It will be typically be called from the filesystem's
824 * setattr function when ATTR_SIZE is passed in.
826 * Must be called with a lock serializing truncates and writes (generally
827 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
828 * specific block truncation has been performed.
830 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
832 loff_t oldsize
= inode
->i_size
;
834 i_size_write(inode
, newsize
);
835 if (newsize
> oldsize
)
836 pagecache_isize_extended(inode
, oldsize
, newsize
);
837 truncate_pagecache(inode
, newsize
);
839 EXPORT_SYMBOL(truncate_setsize
);
842 * pagecache_isize_extended - update pagecache after extension of i_size
843 * @inode: inode for which i_size was extended
844 * @from: original inode size
845 * @to: new inode size
847 * Handle extension of inode size either caused by extending truncate or by
848 * write starting after current i_size. We mark the page straddling current
849 * i_size RO so that page_mkwrite() is called on the nearest write access to
850 * the page. This way filesystem can be sure that page_mkwrite() is called on
851 * the page before user writes to the page via mmap after the i_size has been
854 * The function must be called after i_size is updated so that page fault
855 * coming after we unlock the page will already see the new i_size.
856 * The function must be called while we still hold i_mutex - this not only
857 * makes sure i_size is stable but also that userspace cannot observe new
858 * i_size value before we are prepared to store mmap writes at new inode size.
860 void pagecache_isize_extended(struct inode
*inode
, loff_t from
, loff_t to
)
862 int bsize
= i_blocksize(inode
);
867 WARN_ON(to
> inode
->i_size
);
869 if (from
>= to
|| bsize
== PAGE_SIZE
)
871 /* Page straddling @from will not have any hole block created? */
872 rounded_from
= round_up(from
, bsize
);
873 if (to
<= rounded_from
|| !(rounded_from
& (PAGE_SIZE
- 1)))
876 index
= from
>> PAGE_SHIFT
;
877 page
= find_lock_page(inode
->i_mapping
, index
);
878 /* Page not cached? Nothing to do */
882 * See clear_page_dirty_for_io() for details why set_page_dirty()
885 if (page_mkclean(page
))
886 set_page_dirty(page
);
890 EXPORT_SYMBOL(pagecache_isize_extended
);
893 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
895 * @lstart: offset of beginning of hole
896 * @lend: offset of last byte of hole
898 * This function should typically be called before the filesystem
899 * releases resources associated with the freed range (eg. deallocates
900 * blocks). This way, pagecache will always stay logically coherent
901 * with on-disk format, and the filesystem would not have to deal with
902 * situations such as writepage being called for a page that has already
903 * had its underlying blocks deallocated.
905 void truncate_pagecache_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
907 struct address_space
*mapping
= inode
->i_mapping
;
908 loff_t unmap_start
= round_up(lstart
, PAGE_SIZE
);
909 loff_t unmap_end
= round_down(1 + lend
, PAGE_SIZE
) - 1;
911 * This rounding is currently just for example: unmap_mapping_range
912 * expands its hole outwards, whereas we want it to contract the hole
913 * inwards. However, existing callers of truncate_pagecache_range are
914 * doing their own page rounding first. Note that unmap_mapping_range
915 * allows holelen 0 for all, and we allow lend -1 for end of file.
919 * Unlike in truncate_pagecache, unmap_mapping_range is called only
920 * once (before truncating pagecache), and without "even_cows" flag:
921 * hole-punching should not remove private COWed pages from the hole.
923 if ((u64
)unmap_end
> (u64
)unmap_start
)
924 unmap_mapping_range(mapping
, unmap_start
,
925 1 + unmap_end
- unmap_start
, 0);
926 truncate_inode_pages_range(mapping
, lstart
, lend
);
928 EXPORT_SYMBOL(truncate_pagecache_range
);