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.
170 static void truncate_cleanup_page(struct page *page)
172 if (page_mapped(page))
173 unmap_mapping_page(page);
175 if (page_has_private(page))
176 do_invalidatepage(page, 0, thp_size(page));
179 * Some filesystems seem to re-dirty the page even after
180 * the VM has canceled the dirty bit (eg ext3 journaling).
181 * Hence dirty accounting check is placed after invalidation.
183 cancel_dirty_page(page);
184 ClearPageMappedToDisk(page);
188 * This is for invalidate_mapping_pages(). That function can be called at
189 * any time, and is not supposed to throw away dirty pages. But pages can
190 * be marked dirty at any time too, so use remove_mapping which safely
191 * discards clean, unused pages.
193 * Returns non-zero if the page was successfully invalidated.
196 invalidate_complete_page(struct address_space *mapping, struct page *page)
200 if (page->mapping != mapping)
203 if (page_has_private(page) && !try_to_release_page(page, 0))
206 ret = remove_mapping(mapping, page);
211 int truncate_inode_page(struct address_space *mapping, struct page *page)
213 VM_BUG_ON_PAGE(PageTail(page), page);
215 if (page->mapping != mapping)
218 truncate_cleanup_page(page);
219 delete_from_page_cache(page);
224 * Used to get rid of pages on hardware memory corruption.
226 int generic_error_remove_page(struct address_space *mapping, struct page *page)
231 * Only punch for normal data pages for now.
232 * Handling other types like directories would need more auditing.
234 if (!S_ISREG(mapping->host->i_mode))
236 return truncate_inode_page(mapping, page);
238 EXPORT_SYMBOL(generic_error_remove_page);
241 * Safely invalidate one page from its pagecache mapping.
242 * It only drops clean, unused pages. The page must be locked.
244 * Returns 1 if the page is successfully invalidated, otherwise 0.
246 int invalidate_inode_page(struct page *page)
248 struct address_space *mapping = page_mapping(page);
251 if (PageDirty(page) || PageWriteback(page))
253 if (page_mapped(page))
255 return invalidate_complete_page(mapping, page);
259 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
260 * @mapping: mapping to truncate
261 * @lstart: offset from which to truncate
262 * @lend: offset to which to truncate (inclusive)
264 * Truncate the page cache, removing the pages that are between
265 * specified offsets (and zeroing out partial pages
266 * if lstart or lend + 1 is not page aligned).
268 * Truncate takes two passes - the first pass is nonblocking. It will not
269 * block on page locks and it will not block on writeback. The second pass
270 * will wait. This is to prevent as much IO as possible in the affected region.
271 * The first pass will remove most pages, so the search cost of the second pass
274 * We pass down the cache-hot hint to the page freeing code. Even if the
275 * mapping is large, it is probably the case that the final pages are the most
276 * recently touched, and freeing happens in ascending file offset order.
278 * Note that since ->invalidatepage() accepts range to invalidate
279 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
280 * page aligned properly.
282 void truncate_inode_pages_range(struct address_space *mapping,
283 loff_t lstart, loff_t lend)
285 pgoff_t start; /* inclusive */
286 pgoff_t end; /* exclusive */
287 unsigned int partial_start; /* inclusive */
288 unsigned int partial_end; /* exclusive */
290 pgoff_t indices[PAGEVEC_SIZE];
294 if (mapping_empty(mapping))
297 /* Offsets within partial pages */
298 partial_start = lstart & (PAGE_SIZE - 1);
299 partial_end = (lend + 1) & (PAGE_SIZE - 1);
302 * 'start' and 'end' always covers the range of pages to be fully
303 * truncated. Partial pages are covered with 'partial_start' at the
304 * start of the range and 'partial_end' at the end of the range.
305 * Note that 'end' is exclusive while 'lend' is inclusive.
307 start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
310 * lend == -1 indicates end-of-file so we have to set 'end'
311 * to the highest possible pgoff_t and since the type is
312 * unsigned we're using -1.
316 end = (lend + 1) >> PAGE_SHIFT;
320 while (index < end && find_lock_entries(mapping, index, end - 1,
322 index = indices[pagevec_count(&pvec) - 1] + 1;
323 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
324 for (i = 0; i < pagevec_count(&pvec); i++)
325 truncate_cleanup_page(pvec.pages[i]);
326 delete_from_page_cache_batch(mapping, &pvec);
327 for (i = 0; i < pagevec_count(&pvec); i++)
328 unlock_page(pvec.pages[i]);
329 pagevec_release(&pvec);
334 struct page *page = find_lock_page(mapping, start - 1);
336 unsigned int top = PAGE_SIZE;
338 /* Truncation within a single page */
342 wait_on_page_writeback(page);
343 zero_user_segment(page, partial_start, top);
344 cleancache_invalidate_page(mapping, page);
345 if (page_has_private(page))
346 do_invalidatepage(page, partial_start,
347 top - partial_start);
353 struct page *page = find_lock_page(mapping, end);
355 wait_on_page_writeback(page);
356 zero_user_segment(page, 0, partial_end);
357 cleancache_invalidate_page(mapping, page);
358 if (page_has_private(page))
359 do_invalidatepage(page, 0,
366 * If the truncation happened within a single page no pages
367 * will be released, just zeroed, so we can bail out now.
375 if (!find_get_entries(mapping, index, end - 1, &pvec,
377 /* If all gone from start onwards, we're done */
380 /* Otherwise restart to make sure all gone */
385 for (i = 0; i < pagevec_count(&pvec); i++) {
386 struct page *page = pvec.pages[i];
388 /* We rely upon deletion not changing page->index */
391 if (xa_is_value(page))
395 WARN_ON(page_to_index(page) != index);
396 wait_on_page_writeback(page);
397 truncate_inode_page(mapping, page);
400 truncate_exceptional_pvec_entries(mapping, &pvec, indices);
401 pagevec_release(&pvec);
406 cleancache_invalidate_inode(mapping);
408 EXPORT_SYMBOL(truncate_inode_pages_range);
411 * truncate_inode_pages - truncate *all* the pages from an offset
412 * @mapping: mapping to truncate
413 * @lstart: offset from which to truncate
415 * Called under (and serialised by) inode->i_mutex.
417 * Note: When this function returns, there can be a page in the process of
418 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
419 * mapping->nrpages can be non-zero when this function returns even after
420 * truncation of the whole mapping.
422 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
424 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
426 EXPORT_SYMBOL(truncate_inode_pages);
429 * truncate_inode_pages_final - truncate *all* pages before inode dies
430 * @mapping: mapping to truncate
432 * Called under (and serialized by) inode->i_mutex.
434 * Filesystems have to use this in the .evict_inode path to inform the
435 * VM that this is the final truncate and the inode is going away.
437 void truncate_inode_pages_final(struct address_space *mapping)
440 * Page reclaim can not participate in regular inode lifetime
441 * management (can't call iput()) and thus can race with the
442 * inode teardown. Tell it when the address space is exiting,
443 * so that it does not install eviction information after the
444 * final truncate has begun.
446 mapping_set_exiting(mapping);
448 if (!mapping_empty(mapping)) {
450 * As truncation uses a lockless tree lookup, cycle
451 * the tree lock to make sure any ongoing tree
452 * modification that does not see AS_EXITING is
453 * completed before starting the final truncate.
455 xa_lock_irq(&mapping->i_pages);
456 xa_unlock_irq(&mapping->i_pages);
460 * Cleancache needs notification even if there are no pages or shadow
463 truncate_inode_pages(mapping, 0);
465 EXPORT_SYMBOL(truncate_inode_pages_final);
467 static unsigned long __invalidate_mapping_pages(struct address_space *mapping,
468 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
470 pgoff_t indices[PAGEVEC_SIZE];
472 pgoff_t index = start;
474 unsigned long count = 0;
478 while (find_lock_entries(mapping, index, end, &pvec, indices)) {
479 for (i = 0; i < pagevec_count(&pvec); i++) {
480 struct page *page = pvec.pages[i];
482 /* We rely upon deletion not changing page->index */
485 if (xa_is_value(page)) {
486 invalidate_exceptional_entry(mapping, index,
490 index += thp_nr_pages(page) - 1;
492 ret = invalidate_inode_page(page);
495 * Invalidation is a hint that the page is no longer
496 * of interest and try to speed up its reclaim.
499 deactivate_file_page(page);
500 /* It is likely on the pagevec of a remote CPU */
506 pagevec_remove_exceptionals(&pvec);
507 pagevec_release(&pvec);
515 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
516 * @mapping: the address_space which holds the pages to invalidate
517 * @start: the offset 'from' which to invalidate
518 * @end: the offset 'to' which to invalidate (inclusive)
520 * This function only removes the unlocked pages, if you want to
521 * remove all the pages of one inode, you must call truncate_inode_pages.
523 * invalidate_mapping_pages() will not block on IO activity. It will not
524 * invalidate pages which are dirty, locked, under writeback or mapped into
527 * Return: the number of the pages that were invalidated
529 unsigned long invalidate_mapping_pages(struct address_space *mapping,
530 pgoff_t start, pgoff_t end)
532 return __invalidate_mapping_pages(mapping, start, end, NULL);
534 EXPORT_SYMBOL(invalidate_mapping_pages);
537 * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode
538 * @mapping: the address_space which holds the pages to invalidate
539 * @start: the offset 'from' which to invalidate
540 * @end: the offset 'to' which to invalidate (inclusive)
541 * @nr_pagevec: invalidate failed page number for caller
543 * This helper is similar to invalidate_mapping_pages(), except that it accounts
544 * for pages that are likely on a pagevec and counts them in @nr_pagevec, which
545 * will be used by the caller.
547 void invalidate_mapping_pagevec(struct address_space *mapping,
548 pgoff_t start, pgoff_t end, unsigned long *nr_pagevec)
550 __invalidate_mapping_pages(mapping, start, end, nr_pagevec);
554 * This is like invalidate_complete_page(), except it ignores the page's
555 * refcount. We do this because invalidate_inode_pages2() needs stronger
556 * invalidation guarantees, and cannot afford to leave pages behind because
557 * shrink_page_list() has a temp ref on them, or because they're transiently
558 * sitting in the lru_cache_add() pagevecs.
561 invalidate_complete_page2(struct address_space *mapping, struct page *page)
565 if (page->mapping != mapping)
568 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
571 xa_lock_irqsave(&mapping->i_pages, flags);
575 BUG_ON(page_has_private(page));
576 __delete_from_page_cache(page, NULL);
577 xa_unlock_irqrestore(&mapping->i_pages, flags);
579 if (mapping->a_ops->freepage)
580 mapping->a_ops->freepage(page);
582 put_page(page); /* pagecache ref */
585 xa_unlock_irqrestore(&mapping->i_pages, flags);
589 static int do_launder_page(struct address_space *mapping, struct page *page)
591 if (!PageDirty(page))
593 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
595 return mapping->a_ops->launder_page(page);
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];
618 int did_range_unmap = 0;
620 if (mapping_empty(mapping))
625 while (find_get_entries(mapping, index, end, &pvec, indices)) {
626 for (i = 0; i < pagevec_count(&pvec); i++) {
627 struct page *page = pvec.pages[i];
629 /* We rely upon deletion not changing page->index */
632 if (xa_is_value(page)) {
633 if (!invalidate_exceptional_entry2(mapping,
639 if (!did_range_unmap && page_mapped(page)) {
641 * If page is mapped, before taking its lock,
642 * zap the rest of the file in one hit.
644 unmap_mapping_pages(mapping, index,
645 (1 + end - index), false);
650 WARN_ON(page_to_index(page) != index);
651 if (page->mapping != mapping) {
655 wait_on_page_writeback(page);
657 if (page_mapped(page))
658 unmap_mapping_page(page);
659 BUG_ON(page_mapped(page));
661 ret2 = do_launder_page(mapping, page);
663 if (!invalidate_complete_page2(mapping, page))
670 pagevec_remove_exceptionals(&pvec);
671 pagevec_release(&pvec);
676 * For DAX we invalidate page tables after invalidating page cache. We
677 * could invalidate page tables while invalidating each entry however
678 * that would be expensive. And doing range unmapping before doesn't
679 * work as we have no cheap way to find whether page cache entry didn't
680 * get remapped later.
682 if (dax_mapping(mapping)) {
683 unmap_mapping_pages(mapping, start, end - start + 1, false);
686 cleancache_invalidate_inode(mapping);
689 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
692 * invalidate_inode_pages2 - remove all pages from an address_space
693 * @mapping: the address_space
695 * Any pages which are found to be mapped into pagetables are unmapped prior to
698 * Return: -EBUSY if any pages could not be invalidated.
700 int invalidate_inode_pages2(struct address_space *mapping)
702 return invalidate_inode_pages2_range(mapping, 0, -1);
704 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
707 * truncate_pagecache - unmap and remove pagecache that has been truncated
709 * @newsize: new file size
711 * inode's new i_size must already be written before truncate_pagecache
714 * This function should typically be called before the filesystem
715 * releases resources associated with the freed range (eg. deallocates
716 * blocks). This way, pagecache will always stay logically coherent
717 * with on-disk format, and the filesystem would not have to deal with
718 * situations such as writepage being called for a page that has already
719 * had its underlying blocks deallocated.
721 void truncate_pagecache(struct inode *inode, loff_t newsize)
723 struct address_space *mapping = inode->i_mapping;
724 loff_t holebegin = round_up(newsize, PAGE_SIZE);
727 * unmap_mapping_range is called twice, first simply for
728 * efficiency so that truncate_inode_pages does fewer
729 * single-page unmaps. However after this first call, and
730 * before truncate_inode_pages finishes, it is possible for
731 * private pages to be COWed, which remain after
732 * truncate_inode_pages finishes, hence the second
733 * unmap_mapping_range call must be made for correctness.
735 unmap_mapping_range(mapping, holebegin, 0, 1);
736 truncate_inode_pages(mapping, newsize);
737 unmap_mapping_range(mapping, holebegin, 0, 1);
739 EXPORT_SYMBOL(truncate_pagecache);
742 * truncate_setsize - update inode and pagecache for a new file size
744 * @newsize: new file size
746 * truncate_setsize updates i_size and performs pagecache truncation (if
747 * necessary) to @newsize. It will be typically be called from the filesystem's
748 * setattr function when ATTR_SIZE is passed in.
750 * Must be called with a lock serializing truncates and writes (generally
751 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
752 * specific block truncation has been performed.
754 void truncate_setsize(struct inode *inode, loff_t newsize)
756 loff_t oldsize = inode->i_size;
758 i_size_write(inode, newsize);
759 if (newsize > oldsize)
760 pagecache_isize_extended(inode, oldsize, newsize);
761 truncate_pagecache(inode, newsize);
763 EXPORT_SYMBOL(truncate_setsize);
766 * pagecache_isize_extended - update pagecache after extension of i_size
767 * @inode: inode for which i_size was extended
768 * @from: original inode size
769 * @to: new inode size
771 * Handle extension of inode size either caused by extending truncate or by
772 * write starting after current i_size. We mark the page straddling current
773 * i_size RO so that page_mkwrite() is called on the nearest write access to
774 * the page. This way filesystem can be sure that page_mkwrite() is called on
775 * the page before user writes to the page via mmap after the i_size has been
778 * The function must be called after i_size is updated so that page fault
779 * coming after we unlock the page will already see the new i_size.
780 * The function must be called while we still hold i_mutex - this not only
781 * makes sure i_size is stable but also that userspace cannot observe new
782 * i_size value before we are prepared to store mmap writes at new inode size.
784 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
786 int bsize = i_blocksize(inode);
791 WARN_ON(to > inode->i_size);
793 if (from >= to || bsize == PAGE_SIZE)
795 /* Page straddling @from will not have any hole block created? */
796 rounded_from = round_up(from, bsize);
797 if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1)))
800 index = from >> PAGE_SHIFT;
801 page = find_lock_page(inode->i_mapping, index);
802 /* Page not cached? Nothing to do */
806 * See clear_page_dirty_for_io() for details why set_page_dirty()
809 if (page_mkclean(page))
810 set_page_dirty(page);
814 EXPORT_SYMBOL(pagecache_isize_extended);
817 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
819 * @lstart: offset of beginning of hole
820 * @lend: offset of last byte of hole
822 * This function should typically be called before the filesystem
823 * releases resources associated with the freed range (eg. deallocates
824 * blocks). This way, pagecache will always stay logically coherent
825 * with on-disk format, and the filesystem would not have to deal with
826 * situations such as writepage being called for a page that has already
827 * had its underlying blocks deallocated.
829 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
831 struct address_space *mapping = inode->i_mapping;
832 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
833 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
835 * This rounding is currently just for example: unmap_mapping_range
836 * expands its hole outwards, whereas we want it to contract the hole
837 * inwards. However, existing callers of truncate_pagecache_range are
838 * doing their own page rounding first. Note that unmap_mapping_range
839 * allows holelen 0 for all, and we allow lend -1 for end of file.
843 * Unlike in truncate_pagecache, unmap_mapping_range is called only
844 * once (before truncating pagecache), and without "even_cows" flag:
845 * hole-punching should not remove private COWed pages from the hole.
847 if ((u64)unmap_end > (u64)unmap_start)
848 unmap_mapping_range(mapping, unmap_start,
849 1 + unmap_end - unmap_start, 0);
850 truncate_inode_pages_range(mapping, lstart, lend);
852 EXPORT_SYMBOL(truncate_pagecache_range);