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);
45 static void clear_shadow_entry(struct address_space *mapping, pgoff_t index,
48 xa_lock_irq(&mapping->i_pages);
49 __clear_shadow_entry(mapping, index, entry);
50 xa_unlock_irq(&mapping->i_pages);
54 * Unconditionally remove exceptional entries. Usually called from truncate
55 * path. Note that the pagevec may be altered by this function by removing
56 * exceptional entries similar to what pagevec_remove_exceptionals does.
58 static void truncate_exceptional_pvec_entries(struct address_space *mapping,
59 struct pagevec *pvec, pgoff_t *indices)
64 /* Handled by shmem itself */
65 if (shmem_mapping(mapping))
68 for (j = 0; j < pagevec_count(pvec); j++)
69 if (xa_is_value(pvec->pages[j]))
72 if (j == pagevec_count(pvec))
75 dax = dax_mapping(mapping);
77 xa_lock_irq(&mapping->i_pages);
79 for (i = j; i < pagevec_count(pvec); i++) {
80 struct page *page = pvec->pages[i];
81 pgoff_t index = indices[i];
83 if (!xa_is_value(page)) {
84 pvec->pages[j++] = page;
89 dax_delete_mapping_entry(mapping, index);
93 __clear_shadow_entry(mapping, index, page);
97 xa_unlock_irq(&mapping->i_pages);
102 * Invalidate exceptional entry if easily possible. This handles exceptional
103 * entries for invalidate_inode_pages().
105 static int invalidate_exceptional_entry(struct address_space *mapping,
106 pgoff_t index, void *entry)
108 /* Handled by shmem itself, or for DAX we do nothing. */
109 if (shmem_mapping(mapping) || dax_mapping(mapping))
111 clear_shadow_entry(mapping, index, entry);
116 * Invalidate exceptional entry if clean. This handles exceptional entries for
117 * invalidate_inode_pages2() so for DAX it evicts only clean entries.
119 static int invalidate_exceptional_entry2(struct address_space *mapping,
120 pgoff_t index, void *entry)
122 /* Handled by shmem itself */
123 if (shmem_mapping(mapping))
125 if (dax_mapping(mapping))
126 return dax_invalidate_mapping_entry_sync(mapping, index);
127 clear_shadow_entry(mapping, index, entry);
132 * do_invalidatepage - invalidate part or all of a page
133 * @page: the page which is affected
134 * @offset: start of the range to invalidate
135 * @length: length of the range to invalidate
137 * do_invalidatepage() is called when all or part of the page has become
138 * invalidated by a truncate operation.
140 * do_invalidatepage() does not have to release all buffers, but it must
141 * ensure that no dirty buffer is left outside @offset and that no I/O
142 * is underway against any of the blocks which are outside the truncation
143 * point. Because the caller is about to free (and possibly reuse) those
146 void do_invalidatepage(struct page *page, unsigned int offset,
149 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
151 invalidatepage = page->mapping->a_ops->invalidatepage;
154 invalidatepage = block_invalidatepage;
157 (*invalidatepage)(page, offset, length);
161 * If truncate cannot remove the fs-private metadata from the page, the page
162 * becomes orphaned. It will be left on the LRU and may even be mapped into
163 * user pagetables if we're racing with filemap_fault().
165 * We need to bail out if page->mapping is no longer equal to the original
166 * mapping. This happens a) when the VM reclaimed the page while we waited on
167 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
168 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
171 truncate_cleanup_page(struct address_space *mapping, struct page *page)
173 if (page_mapped(page)) {
174 unsigned int nr = thp_nr_pages(page);
175 unmap_mapping_pages(mapping, page->index, nr, false);
178 if (page_has_private(page))
179 do_invalidatepage(page, 0, thp_size(page));
182 * Some filesystems seem to re-dirty the page even after
183 * the VM has canceled the dirty bit (eg ext3 journaling).
184 * Hence dirty accounting check is placed after invalidation.
186 cancel_dirty_page(page);
187 ClearPageMappedToDisk(page);
191 * This is for invalidate_mapping_pages(). That function can be called at
192 * any time, and is not supposed to throw away dirty pages. But pages can
193 * be marked dirty at any time too, so use remove_mapping which safely
194 * discards clean, unused pages.
196 * Returns non-zero if the page was successfully invalidated.
199 invalidate_complete_page(struct address_space *mapping, struct page *page)
203 if (page->mapping != mapping)
206 if (page_has_private(page) && !try_to_release_page(page, 0))
209 ret = remove_mapping(mapping, page);
214 int truncate_inode_page(struct address_space *mapping, struct page *page)
216 VM_BUG_ON_PAGE(PageTail(page), page);
218 if (page->mapping != mapping)
221 truncate_cleanup_page(mapping, page);
222 delete_from_page_cache(page);
227 * Used to get rid of pages on hardware memory corruption.
229 int generic_error_remove_page(struct address_space *mapping, struct page *page)
234 * Only punch for normal data pages for now.
235 * Handling other types like directories would need more auditing.
237 if (!S_ISREG(mapping->host->i_mode))
239 return truncate_inode_page(mapping, page);
241 EXPORT_SYMBOL(generic_error_remove_page);
244 * Safely invalidate one page from its pagecache mapping.
245 * It only drops clean, unused pages. The page must be locked.
247 * Returns 1 if the page is successfully invalidated, otherwise 0.
249 int invalidate_inode_page(struct page *page)
251 struct address_space *mapping = page_mapping(page);
254 if (PageDirty(page) || PageWriteback(page))
256 if (page_mapped(page))
258 return invalidate_complete_page(mapping, page);
262 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
263 * @mapping: mapping to truncate
264 * @lstart: offset from which to truncate
265 * @lend: offset to which to truncate (inclusive)
267 * Truncate the page cache, removing the pages that are between
268 * specified offsets (and zeroing out partial pages
269 * if lstart or lend + 1 is not page aligned).
271 * Truncate takes two passes - the first pass is nonblocking. It will not
272 * block on page locks and it will not block on writeback. The second pass
273 * will wait. This is to prevent as much IO as possible in the affected region.
274 * The first pass will remove most pages, so the search cost of the second pass
277 * We pass down the cache-hot hint to the page freeing code. Even if the
278 * mapping is large, it is probably the case that the final pages are the most
279 * recently touched, and freeing happens in ascending file offset order.
281 * Note that since ->invalidatepage() accepts range to invalidate
282 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
283 * page aligned properly.
285 void truncate_inode_pages_range(struct address_space *mapping,
286 loff_t lstart, loff_t lend)
288 pgoff_t start; /* inclusive */
289 pgoff_t end; /* exclusive */
290 unsigned int partial_start; /* inclusive */
291 unsigned int partial_end; /* exclusive */
293 pgoff_t indices[PAGEVEC_SIZE];
297 if (mapping_empty(mapping))
300 /* Offsets within partial pages */
301 partial_start = lstart & (PAGE_SIZE - 1);
302 partial_end = (lend + 1) & (PAGE_SIZE - 1);
305 * 'start' and 'end' always covers the range of pages to be fully
306 * truncated. Partial pages are covered with 'partial_start' at the
307 * start of the range and 'partial_end' at the end of the range.
308 * Note that 'end' is exclusive while 'lend' is inclusive.
310 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
313 * lend == -1 indicates end-of-file so we have to set 'end'
314 * to the highest possible pgoff_t and since the type is
315 * unsigned we're using -1.
319 end = (lend + 1) >> PAGE_SHIFT;
323 while (index < end && find_lock_entries(mapping, index, end - 1,
325 index = indices[pagevec_count(&pvec) - 1] + 1;
326 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
327 for (i = 0; i < pagevec_count(&pvec); i++)
328 truncate_cleanup_page(mapping, pvec.pages[i]);
329 delete_from_page_cache_batch(mapping, &pvec);
330 for (i = 0; i < pagevec_count(&pvec); i++)
331 unlock_page(pvec.pages[i]);
332 pagevec_release(&pvec);
337 struct page *page = find_lock_page(mapping, start - 1);
339 unsigned int top = PAGE_SIZE;
341 /* Truncation within a single page */
345 wait_on_page_writeback(page);
346 zero_user_segment(page, partial_start, top);
347 cleancache_invalidate_page(mapping, page);
348 if (page_has_private(page))
349 do_invalidatepage(page, partial_start,
350 top - partial_start);
356 struct page *page = find_lock_page(mapping, end);
358 wait_on_page_writeback(page);
359 zero_user_segment(page, 0, partial_end);
360 cleancache_invalidate_page(mapping, page);
361 if (page_has_private(page))
362 do_invalidatepage(page, 0,
369 * If the truncation happened within a single page no pages
370 * will be released, just zeroed, so we can bail out now.
378 if (!find_get_entries(mapping, index, end - 1, &pvec,
380 /* If all gone from start onwards, we're done */
383 /* Otherwise restart to make sure all gone */
388 for (i = 0; i < pagevec_count(&pvec); i++) {
389 struct page *page = pvec.pages[i];
391 /* We rely upon deletion not changing page->index */
394 if (xa_is_value(page))
398 WARN_ON(page_to_index(page) != index);
399 wait_on_page_writeback(page);
400 truncate_inode_page(mapping, page);
403 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
404 pagevec_release(&pvec);
409 cleancache_invalidate_inode(mapping);
411 EXPORT_SYMBOL(truncate_inode_pages_range);
414 * truncate_inode_pages - truncate *all* the pages from an offset
415 * @mapping: mapping to truncate
416 * @lstart: offset from which to truncate
418 * Called under (and serialised by) inode->i_mutex.
420 * Note: When this function returns, there can be a page in the process of
421 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
422 * mapping->nrpages can be non-zero when this function returns even after
423 * truncation of the whole mapping.
425 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
427 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
429 EXPORT_SYMBOL(truncate_inode_pages);
432 * truncate_inode_pages_final - truncate *all* pages before inode dies
433 * @mapping: mapping to truncate
435 * Called under (and serialized by) inode->i_mutex.
437 * Filesystems have to use this in the .evict_inode path to inform the
438 * VM that this is the final truncate and the inode is going away.
440 void truncate_inode_pages_final(struct address_space *mapping)
443 * Page reclaim can not participate in regular inode lifetime
444 * management (can't call iput()) and thus can race with the
445 * inode teardown. Tell it when the address space is exiting,
446 * so that it does not install eviction information after the
447 * final truncate has begun.
449 mapping_set_exiting(mapping);
451 if (!mapping_empty(mapping)) {
453 * As truncation uses a lockless tree lookup, cycle
454 * the tree lock to make sure any ongoing tree
455 * modification that does not see AS_EXITING is
456 * completed before starting the final truncate.
458 xa_lock_irq(&mapping->i_pages);
459 xa_unlock_irq(&mapping->i_pages);
463 * Cleancache needs notification even if there are no pages or shadow
466 truncate_inode_pages(mapping, 0);
468 EXPORT_SYMBOL(truncate_inode_pages_final);
470 static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
471 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
473 pgoff_t indices[PAGEVEC_SIZE];
475 pgoff_t index = start;
477 unsigned long count = 0;
481 while (find_lock_entries(mapping, index, end, &pvec, indices)) {
482 for (i = 0; i < pagevec_count(&pvec); i++) {
483 struct page *page = pvec.pages[i];
485 /* We rely upon deletion not changing page->index */
488 if (xa_is_value(page)) {
489 invalidate_exceptional_entry(mapping, index,
493 index += thp_nr_pages(page) - 1;
495 ret = invalidate_inode_page(page);
498 * Invalidation is a hint that the page is no longer
499 * of interest and try to speed up its reclaim.
502 deactivate_file_page(page);
503 /* It is likely on the pagevec of a remote CPU */
509 pagevec_remove_exceptionals(&pvec);
510 pagevec_release(&pvec);
518 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
519 * @mapping: the address_space which holds the pages to invalidate
520 * @start: the offset 'from' which to invalidate
521 * @end: the offset 'to' which to invalidate (inclusive)
523 * This function only removes the unlocked pages, if you want to
524 * remove all the pages of one inode, you must call truncate_inode_pages.
526 * invalidate_mapping_pages() will not block on IO activity. It will not
527 * invalidate pages which are dirty, locked, under writeback or mapped into
530 * Return: the number of the pages that were invalidated
532 unsigned long invalidate_mapping_pages(struct address_space *mapping,
533 pgoff_t start, pgoff_t end)
535 return __invalidate_mapping_pages(mapping, start, end, NULL);
537 EXPORT_SYMBOL(invalidate_mapping_pages);
540 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
541 * @mapping: the address_space which holds the pages to invalidate
542 * @start: the offset 'from' which to invalidate
543 * @end: the offset 'to' which to invalidate (inclusive)
544 * @nr_pagevec: invalidate failed page number for caller
546 * This helper is similar to invalidate_mapping_pages(), except that it accounts
547 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
548 * will be used by the caller.
550 void invalidate_mapping_pagevec(struct address_space *mapping,
551 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
553 __invalidate_mapping_pages(mapping, start, end, nr_pagevec);
557 * This is like invalidate_complete_page(), except it ignores the page's
558 * refcount. We do this because invalidate_inode_pages2() needs stronger
559 * invalidation guarantees, and cannot afford to leave pages behind because
560 * shrink_page_list() has a temp ref on them, or because they're transiently
561 * sitting in the lru_cache_add() pagevecs.
564 invalidate_complete_page2(struct address_space *mapping, struct page *page)
568 if (page->mapping != mapping)
571 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
574 xa_lock_irqsave(&mapping->i_pages, flags);
578 BUG_ON(page_has_private(page));
579 __delete_from_page_cache(page, NULL);
580 xa_unlock_irqrestore(&mapping->i_pages, flags);
582 if (mapping->a_ops->freepage)
583 mapping->a_ops->freepage(page);
585 put_page(page); /* pagecache ref */
588 xa_unlock_irqrestore(&mapping->i_pages, flags);
592 static int do_launder_page(struct address_space *mapping, struct page *page)
594 if (!PageDirty(page))
596 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
598 return mapping->a_ops->launder_page(page);
602 * invalidate_inode_pages2_range - remove range of pages from an address_space
603 * @mapping: the address_space
604 * @start: the page offset 'from' which to invalidate
605 * @end: the page offset 'to' which to invalidate (inclusive)
607 * Any pages which are found to be mapped into pagetables are unmapped prior to
610 * Return: -EBUSY if any pages could not be invalidated.
612 int invalidate_inode_pages2_range(struct address_space *mapping,
613 pgoff_t start, pgoff_t end)
615 pgoff_t indices[PAGEVEC_SIZE];
621 int did_range_unmap = 0;
623 if (mapping_empty(mapping))
628 while (find_get_entries(mapping, index, end, &pvec, indices)) {
629 for (i = 0; i < pagevec_count(&pvec); i++) {
630 struct page *page = pvec.pages[i];
632 /* We rely upon deletion not changing page->index */
635 if (xa_is_value(page)) {
636 if (!invalidate_exceptional_entry2(mapping,
643 WARN_ON(page_to_index(page) != index);
644 if (page->mapping != mapping) {
648 wait_on_page_writeback(page);
649 if (page_mapped(page)) {
650 if (!did_range_unmap) {
652 * Zap the rest of the file in one hit.
654 unmap_mapping_pages(mapping, index,
655 (1 + end - index), false);
661 unmap_mapping_pages(mapping, index,
665 BUG_ON(page_mapped(page));
666 ret2 = do_launder_page(mapping, page);
668 if (!invalidate_complete_page2(mapping, page))
675 pagevec_remove_exceptionals(&pvec);
676 pagevec_release(&pvec);
681 * For DAX we invalidate page tables after invalidating page cache. We
682 * could invalidate page tables while invalidating each entry however
683 * that would be expensive. And doing range unmapping before doesn't
684 * work as we have no cheap way to find whether page cache entry didn't
685 * get remapped later.
687 if (dax_mapping(mapping)) {
688 unmap_mapping_pages(mapping, start, end - start + 1, false);
691 cleancache_invalidate_inode(mapping);
694 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
697 * invalidate_inode_pages2 - remove all pages from an address_space
698 * @mapping: the address_space
700 * Any pages which are found to be mapped into pagetables are unmapped prior to
703 * Return: -EBUSY if any pages could not be invalidated.
705 int invalidate_inode_pages2(struct address_space *mapping)
707 return invalidate_inode_pages2_range(mapping, 0, -1);
709 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
712 * truncate_pagecache - unmap and remove pagecache that has been truncated
714 * @newsize: new file size
716 * inode's new i_size must already be written before truncate_pagecache
719 * This function should typically be called before the filesystem
720 * releases resources associated with the freed range (eg. deallocates
721 * blocks). This way, pagecache will always stay logically coherent
722 * with on-disk format, and the filesystem would not have to deal with
723 * situations such as writepage being called for a page that has already
724 * had its underlying blocks deallocated.
726 void truncate_pagecache(struct inode *inode, loff_t newsize)
728 struct address_space *mapping = inode->i_mapping;
729 loff_t holebegin = round_up(newsize, PAGE_SIZE);
732 * unmap_mapping_range is called twice, first simply for
733 * efficiency so that truncate_inode_pages does fewer
734 * single-page unmaps. However after this first call, and
735 * before truncate_inode_pages finishes, it is possible for
736 * private pages to be COWed, which remain after
737 * truncate_inode_pages finishes, hence the second
738 * unmap_mapping_range call must be made for correctness.
740 unmap_mapping_range(mapping, holebegin, 0, 1);
741 truncate_inode_pages(mapping, newsize);
742 unmap_mapping_range(mapping, holebegin, 0, 1);
744 EXPORT_SYMBOL(truncate_pagecache);
747 * truncate_setsize - update inode and pagecache for a new file size
749 * @newsize: new file size
751 * truncate_setsize updates i_size and performs pagecache truncation (if
752 * necessary) to @newsize. It will be typically be called from the filesystem's
753 * setattr function when ATTR_SIZE is passed in.
755 * Must be called with a lock serializing truncates and writes (generally
756 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
757 * specific block truncation has been performed.
759 void truncate_setsize(struct inode *inode, loff_t newsize)
761 loff_t oldsize = inode->i_size;
763 i_size_write(inode, newsize);
764 if (newsize > oldsize)
765 pagecache_isize_extended(inode, oldsize, newsize);
766 truncate_pagecache(inode, newsize);
768 EXPORT_SYMBOL(truncate_setsize);
771 * pagecache_isize_extended - update pagecache after extension of i_size
772 * @inode: inode for which i_size was extended
773 * @from: original inode size
774 * @to: new inode size
776 * Handle extension of inode size either caused by extending truncate or by
777 * write starting after current i_size. We mark the page straddling current
778 * i_size RO so that page_mkwrite() is called on the nearest write access to
779 * the page. This way filesystem can be sure that page_mkwrite() is called on
780 * the page before user writes to the page via mmap after the i_size has been
783 * The function must be called after i_size is updated so that page fault
784 * coming after we unlock the page will already see the new i_size.
785 * The function must be called while we still hold i_mutex - this not only
786 * makes sure i_size is stable but also that userspace cannot observe new
787 * i_size value before we are prepared to store mmap writes at new inode size.
789 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
791 int bsize = i_blocksize(inode);
796 WARN_ON(to > inode->i_size);
798 if (from >= to || bsize == PAGE_SIZE)
800 /* Page straddling @from will not have any hole block created? */
801 rounded_from = round_up(from, bsize);
802 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
805 index = from >> PAGE_SHIFT;
806 page = find_lock_page(inode->i_mapping, index);
807 /* Page not cached? Nothing to do */
811 * See clear_page_dirty_for_io() for details why set_page_dirty()
814 if (page_mkclean(page))
815 set_page_dirty(page);
819 EXPORT_SYMBOL(pagecache_isize_extended);
822 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
824 * @lstart: offset of beginning of hole
825 * @lend: offset of last byte of hole
827 * This function should typically be called before the filesystem
828 * releases resources associated with the freed range (eg. deallocates
829 * blocks). This way, pagecache will always stay logically coherent
830 * with on-disk format, and the filesystem would not have to deal with
831 * situations such as writepage being called for a page that has already
832 * had its underlying blocks deallocated.
834 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
836 struct address_space *mapping = inode->i_mapping;
837 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
838 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
840 * This rounding is currently just for example: unmap_mapping_range
841 * expands its hole outwards, whereas we want it to contract the hole
842 * inwards. However, existing callers of truncate_pagecache_range are
843 * doing their own page rounding first. Note that unmap_mapping_range
844 * allows holelen 0 for all, and we allow lend -1 for end of file.
848 * Unlike in truncate_pagecache, unmap_mapping_range is called only
849 * once (before truncating pagecache), and without "even_cows" flag:
850 * hole-punching should not remove private COWed pages from the hole.
852 if ((u64)unmap_end > (u64)unmap_start)
853 unmap_mapping_range(mapping, unmap_start,
854 1 + unmap_end - unmap_start, 0);
855 truncate_inode_pages_range(mapping, lstart, lend);
857 EXPORT_SYMBOL(truncate_pagecache_range);