2 * mm/truncate.c - code for taking down pages from address_spaces
4 * Copyright (C) 2002, Linus Torvalds
6 * 10Sep2002 Andrew Morton
10 #include <linux/kernel.h>
11 #include <linux/backing-dev.h>
12 #include <linux/dax.h>
13 #include <linux/gfp.h>
15 #include <linux/swap.h>
16 #include <linux/export.h>
17 #include <linux/pagemap.h>
18 #include <linux/highmem.h>
19 #include <linux/pagevec.h>
20 #include <linux/task_io_accounting_ops.h>
21 #include <linux/buffer_head.h> /* grr. try_to_release_page,
23 #include <linux/cleancache.h>
24 #include <linux/rmap.h>
27 static void clear_exceptional_entry(struct address_space *mapping,
28 pgoff_t index, void *entry)
30 struct radix_tree_node *node;
33 /* Handled by shmem itself */
34 if (shmem_mapping(mapping))
37 spin_lock_irq(&mapping->tree_lock);
39 if (dax_mapping(mapping)) {
40 if (radix_tree_delete_item(&mapping->page_tree, index, entry))
41 mapping->nrexceptional--;
44 * Regular page slots are stabilized by the page lock even
45 * without the tree itself locked. These unlocked entries
46 * need verification under the tree lock.
48 if (!__radix_tree_lookup(&mapping->page_tree, index, &node,
53 radix_tree_replace_slot(slot, NULL);
54 mapping->nrexceptional--;
57 workingset_node_shadows_dec(node);
59 * Don't track node without shadow entries.
61 * Avoid acquiring the list_lru lock if already untracked.
62 * The list_empty() test is safe as node->private_list is
63 * protected by mapping->tree_lock.
65 if (!workingset_node_shadows(node) &&
66 !list_empty(&node->private_list))
67 list_lru_del(&workingset_shadow_nodes,
69 __radix_tree_delete_node(&mapping->page_tree, node);
72 spin_unlock_irq(&mapping->tree_lock);
76 * do_invalidatepage - invalidate part or all of a page
77 * @page: the page which is affected
78 * @offset: start of the range to invalidate
79 * @length: length of the range to invalidate
81 * do_invalidatepage() is called when all or part of the page has become
82 * invalidated by a truncate operation.
84 * do_invalidatepage() does not have to release all buffers, but it must
85 * ensure that no dirty buffer is left outside @offset and that no I/O
86 * is underway against any of the blocks which are outside the truncation
87 * point. Because the caller is about to free (and possibly reuse) those
90 void do_invalidatepage(struct page *page, unsigned int offset,
93 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
95 invalidatepage = page->mapping->a_ops->invalidatepage;
98 invalidatepage = block_invalidatepage;
101 (*invalidatepage)(page, offset, length);
105 * If truncate cannot remove the fs-private metadata from the page, the page
106 * becomes orphaned. It will be left on the LRU and may even be mapped into
107 * user pagetables if we're racing with filemap_fault().
109 * We need to bale out if page->mapping is no longer equal to the original
110 * mapping. This happens a) when the VM reclaimed the page while we waited on
111 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
112 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
115 truncate_complete_page(struct address_space *mapping, struct page *page)
117 if (page->mapping != mapping)
120 if (page_has_private(page))
121 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
124 * Some filesystems seem to re-dirty the page even after
125 * the VM has canceled the dirty bit (eg ext3 journaling).
126 * Hence dirty accounting check is placed after invalidation.
128 cancel_dirty_page(page);
129 ClearPageMappedToDisk(page);
130 delete_from_page_cache(page);
135 * This is for invalidate_mapping_pages(). That function can be called at
136 * any time, and is not supposed to throw away dirty pages. But pages can
137 * be marked dirty at any time too, so use remove_mapping which safely
138 * discards clean, unused pages.
140 * Returns non-zero if the page was successfully invalidated.
143 invalidate_complete_page(struct address_space *mapping, struct page *page)
147 if (page->mapping != mapping)
150 if (page_has_private(page) && !try_to_release_page(page, 0))
153 ret = remove_mapping(mapping, page);
158 int truncate_inode_page(struct address_space *mapping, struct page *page)
160 if (page_mapped(page)) {
161 unmap_mapping_range(mapping,
162 (loff_t)page->index << PAGE_CACHE_SHIFT,
165 return truncate_complete_page(mapping, page);
169 * Used to get rid of pages on hardware memory corruption.
171 int generic_error_remove_page(struct address_space *mapping, struct page *page)
176 * Only punch for normal data pages for now.
177 * Handling other types like directories would need more auditing.
179 if (!S_ISREG(mapping->host->i_mode))
181 return truncate_inode_page(mapping, page);
183 EXPORT_SYMBOL(generic_error_remove_page);
186 * Safely invalidate one page from its pagecache mapping.
187 * It only drops clean, unused pages. The page must be locked.
189 * Returns 1 if the page is successfully invalidated, otherwise 0.
191 int invalidate_inode_page(struct page *page)
193 struct address_space *mapping = page_mapping(page);
196 if (PageDirty(page) || PageWriteback(page))
198 if (page_mapped(page))
200 return invalidate_complete_page(mapping, page);
204 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
205 * @mapping: mapping to truncate
206 * @lstart: offset from which to truncate
207 * @lend: offset to which to truncate (inclusive)
209 * Truncate the page cache, removing the pages that are between
210 * specified offsets (and zeroing out partial pages
211 * if lstart or lend + 1 is not page aligned).
213 * Truncate takes two passes - the first pass is nonblocking. It will not
214 * block on page locks and it will not block on writeback. The second pass
215 * will wait. This is to prevent as much IO as possible in the affected region.
216 * The first pass will remove most pages, so the search cost of the second pass
219 * We pass down the cache-hot hint to the page freeing code. Even if the
220 * mapping is large, it is probably the case that the final pages are the most
221 * recently touched, and freeing happens in ascending file offset order.
223 * Note that since ->invalidatepage() accepts range to invalidate
224 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
225 * page aligned properly.
227 void truncate_inode_pages_range(struct address_space *mapping,
228 loff_t lstart, loff_t lend)
230 pgoff_t start; /* inclusive */
231 pgoff_t end; /* exclusive */
232 unsigned int partial_start; /* inclusive */
233 unsigned int partial_end; /* exclusive */
235 pgoff_t indices[PAGEVEC_SIZE];
239 cleancache_invalidate_inode(mapping);
240 if (mapping->nrpages == 0 && mapping->nrexceptional == 0)
243 /* Offsets within partial pages */
244 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
245 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
248 * 'start' and 'end' always covers the range of pages to be fully
249 * truncated. Partial pages are covered with 'partial_start' at the
250 * start of the range and 'partial_end' at the end of the range.
251 * Note that 'end' is exclusive while 'lend' is inclusive.
253 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
256 * lend == -1 indicates end-of-file so we have to set 'end'
257 * to the highest possible pgoff_t and since the type is
258 * unsigned we're using -1.
262 end = (lend + 1) >> PAGE_CACHE_SHIFT;
264 pagevec_init(&pvec, 0);
266 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
267 min(end - index, (pgoff_t)PAGEVEC_SIZE),
269 for (i = 0; i < pagevec_count(&pvec); i++) {
270 struct page *page = pvec.pages[i];
272 /* We rely upon deletion not changing page->index */
277 if (radix_tree_exceptional_entry(page)) {
278 clear_exceptional_entry(mapping, index, page);
282 if (!trylock_page(page))
284 WARN_ON(page->index != index);
285 if (PageWriteback(page)) {
289 truncate_inode_page(mapping, page);
292 pagevec_remove_exceptionals(&pvec);
293 pagevec_release(&pvec);
299 struct page *page = find_lock_page(mapping, start - 1);
301 unsigned int top = PAGE_CACHE_SIZE;
303 /* Truncation within a single page */
307 wait_on_page_writeback(page);
308 zero_user_segment(page, partial_start, top);
309 cleancache_invalidate_page(mapping, page);
310 if (page_has_private(page))
311 do_invalidatepage(page, partial_start,
312 top - partial_start);
314 page_cache_release(page);
318 struct page *page = find_lock_page(mapping, end);
320 wait_on_page_writeback(page);
321 zero_user_segment(page, 0, partial_end);
322 cleancache_invalidate_page(mapping, page);
323 if (page_has_private(page))
324 do_invalidatepage(page, 0,
327 page_cache_release(page);
331 * If the truncation happened within a single page no pages
332 * will be released, just zeroed, so we can bail out now.
340 if (!pagevec_lookup_entries(&pvec, mapping, index,
341 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
342 /* If all gone from start onwards, we're done */
345 /* Otherwise restart to make sure all gone */
349 if (index == start && indices[0] >= end) {
350 /* All gone out of hole to be punched, we're done */
351 pagevec_remove_exceptionals(&pvec);
352 pagevec_release(&pvec);
355 for (i = 0; i < pagevec_count(&pvec); i++) {
356 struct page *page = pvec.pages[i];
358 /* We rely upon deletion not changing page->index */
361 /* Restart punch to make sure all gone */
366 if (radix_tree_exceptional_entry(page)) {
367 clear_exceptional_entry(mapping, index, page);
372 WARN_ON(page->index != index);
373 wait_on_page_writeback(page);
374 truncate_inode_page(mapping, page);
377 pagevec_remove_exceptionals(&pvec);
378 pagevec_release(&pvec);
381 cleancache_invalidate_inode(mapping);
383 EXPORT_SYMBOL(truncate_inode_pages_range);
386 * truncate_inode_pages - truncate *all* the pages from an offset
387 * @mapping: mapping to truncate
388 * @lstart: offset from which to truncate
390 * Called under (and serialised by) inode->i_mutex.
392 * Note: When this function returns, there can be a page in the process of
393 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
394 * mapping->nrpages can be non-zero when this function returns even after
395 * truncation of the whole mapping.
397 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
399 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
401 EXPORT_SYMBOL(truncate_inode_pages);
404 * truncate_inode_pages_final - truncate *all* pages before inode dies
405 * @mapping: mapping to truncate
407 * Called under (and serialized by) inode->i_mutex.
409 * Filesystems have to use this in the .evict_inode path to inform the
410 * VM that this is the final truncate and the inode is going away.
412 void truncate_inode_pages_final(struct address_space *mapping)
414 unsigned long nrexceptional;
415 unsigned long nrpages;
418 * Page reclaim can not participate in regular inode lifetime
419 * management (can't call iput()) and thus can race with the
420 * inode teardown. Tell it when the address space is exiting,
421 * so that it does not install eviction information after the
422 * final truncate has begun.
424 mapping_set_exiting(mapping);
427 * When reclaim installs eviction entries, it increases
428 * nrexceptional first, then decreases nrpages. Make sure we see
429 * this in the right order or we might miss an entry.
431 nrpages = mapping->nrpages;
433 nrexceptional = mapping->nrexceptional;
435 if (nrpages || nrexceptional) {
437 * As truncation uses a lockless tree lookup, cycle
438 * the tree lock to make sure any ongoing tree
439 * modification that does not see AS_EXITING is
440 * completed before starting the final truncate.
442 spin_lock_irq(&mapping->tree_lock);
443 spin_unlock_irq(&mapping->tree_lock);
445 truncate_inode_pages(mapping, 0);
448 EXPORT_SYMBOL(truncate_inode_pages_final);
451 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
452 * @mapping: the address_space which holds the pages to invalidate
453 * @start: the offset 'from' which to invalidate
454 * @end: the offset 'to' which to invalidate (inclusive)
456 * This function only removes the unlocked pages, if you want to
457 * remove all the pages of one inode, you must call truncate_inode_pages.
459 * invalidate_mapping_pages() will not block on IO activity. It will not
460 * invalidate pages which are dirty, locked, under writeback or mapped into
463 unsigned long invalidate_mapping_pages(struct address_space *mapping,
464 pgoff_t start, pgoff_t end)
466 pgoff_t indices[PAGEVEC_SIZE];
468 pgoff_t index = start;
470 unsigned long count = 0;
473 pagevec_init(&pvec, 0);
474 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
475 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
477 for (i = 0; i < pagevec_count(&pvec); i++) {
478 struct page *page = pvec.pages[i];
480 /* We rely upon deletion not changing page->index */
485 if (radix_tree_exceptional_entry(page)) {
486 clear_exceptional_entry(mapping, index, page);
490 if (!trylock_page(page))
492 WARN_ON(page->index != index);
493 ret = invalidate_inode_page(page);
496 * Invalidation is a hint that the page is no longer
497 * of interest and try to speed up its reclaim.
500 deactivate_file_page(page);
503 pagevec_remove_exceptionals(&pvec);
504 pagevec_release(&pvec);
510 EXPORT_SYMBOL(invalidate_mapping_pages);
513 * This is like invalidate_complete_page(), except it ignores the page's
514 * refcount. We do this because invalidate_inode_pages2() needs stronger
515 * invalidation guarantees, and cannot afford to leave pages behind because
516 * shrink_page_list() has a temp ref on them, or because they're transiently
517 * sitting in the lru_cache_add() pagevecs.
520 invalidate_complete_page2(struct address_space *mapping, struct page *page)
522 struct mem_cgroup *memcg;
525 if (page->mapping != mapping)
528 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
531 memcg = mem_cgroup_begin_page_stat(page);
532 spin_lock_irqsave(&mapping->tree_lock, flags);
536 BUG_ON(page_has_private(page));
537 __delete_from_page_cache(page, NULL, memcg);
538 spin_unlock_irqrestore(&mapping->tree_lock, flags);
539 mem_cgroup_end_page_stat(memcg);
541 if (mapping->a_ops->freepage)
542 mapping->a_ops->freepage(page);
544 page_cache_release(page); /* pagecache ref */
547 spin_unlock_irqrestore(&mapping->tree_lock, flags);
548 mem_cgroup_end_page_stat(memcg);
552 static int do_launder_page(struct address_space *mapping, struct page *page)
554 if (!PageDirty(page))
556 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
558 return mapping->a_ops->launder_page(page);
562 * invalidate_inode_pages2_range - remove range of pages from an address_space
563 * @mapping: the address_space
564 * @start: the page offset 'from' which to invalidate
565 * @end: the page offset 'to' which to invalidate (inclusive)
567 * Any pages which are found to be mapped into pagetables are unmapped prior to
570 * Returns -EBUSY if any pages could not be invalidated.
572 int invalidate_inode_pages2_range(struct address_space *mapping,
573 pgoff_t start, pgoff_t end)
575 pgoff_t indices[PAGEVEC_SIZE];
581 int did_range_unmap = 0;
583 cleancache_invalidate_inode(mapping);
584 pagevec_init(&pvec, 0);
586 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
587 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
589 for (i = 0; i < pagevec_count(&pvec); i++) {
590 struct page *page = pvec.pages[i];
592 /* We rely upon deletion not changing page->index */
597 if (radix_tree_exceptional_entry(page)) {
598 clear_exceptional_entry(mapping, index, page);
603 WARN_ON(page->index != index);
604 if (page->mapping != mapping) {
608 wait_on_page_writeback(page);
609 if (page_mapped(page)) {
610 if (!did_range_unmap) {
612 * Zap the rest of the file in one hit.
614 unmap_mapping_range(mapping,
615 (loff_t)index << PAGE_CACHE_SHIFT,
616 (loff_t)(1 + end - index)
624 unmap_mapping_range(mapping,
625 (loff_t)index << PAGE_CACHE_SHIFT,
629 BUG_ON(page_mapped(page));
630 ret2 = do_launder_page(mapping, page);
632 if (!invalidate_complete_page2(mapping, page))
639 pagevec_remove_exceptionals(&pvec);
640 pagevec_release(&pvec);
644 cleancache_invalidate_inode(mapping);
647 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
650 * invalidate_inode_pages2 - remove all pages from an address_space
651 * @mapping: the address_space
653 * Any pages which are found to be mapped into pagetables are unmapped prior to
656 * Returns -EBUSY if any pages could not be invalidated.
658 int invalidate_inode_pages2(struct address_space *mapping)
660 return invalidate_inode_pages2_range(mapping, 0, -1);
662 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
665 * truncate_pagecache - unmap and remove pagecache that has been truncated
667 * @newsize: new file size
669 * inode's new i_size must already be written before truncate_pagecache
672 * This function should typically be called before the filesystem
673 * releases resources associated with the freed range (eg. deallocates
674 * blocks). This way, pagecache will always stay logically coherent
675 * with on-disk format, and the filesystem would not have to deal with
676 * situations such as writepage being called for a page that has already
677 * had its underlying blocks deallocated.
679 void truncate_pagecache(struct inode *inode, loff_t newsize)
681 struct address_space *mapping = inode->i_mapping;
682 loff_t holebegin = round_up(newsize, PAGE_SIZE);
685 * unmap_mapping_range is called twice, first simply for
686 * efficiency so that truncate_inode_pages does fewer
687 * single-page unmaps. However after this first call, and
688 * before truncate_inode_pages finishes, it is possible for
689 * private pages to be COWed, which remain after
690 * truncate_inode_pages finishes, hence the second
691 * unmap_mapping_range call must be made for correctness.
693 unmap_mapping_range(mapping, holebegin, 0, 1);
694 truncate_inode_pages(mapping, newsize);
695 unmap_mapping_range(mapping, holebegin, 0, 1);
697 EXPORT_SYMBOL(truncate_pagecache);
700 * truncate_setsize - update inode and pagecache for a new file size
702 * @newsize: new file size
704 * truncate_setsize updates i_size and performs pagecache truncation (if
705 * necessary) to @newsize. It will be typically be called from the filesystem's
706 * setattr function when ATTR_SIZE is passed in.
708 * Must be called with a lock serializing truncates and writes (generally
709 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
710 * specific block truncation has been performed.
712 void truncate_setsize(struct inode *inode, loff_t newsize)
714 loff_t oldsize = inode->i_size;
716 i_size_write(inode, newsize);
717 if (newsize > oldsize)
718 pagecache_isize_extended(inode, oldsize, newsize);
719 truncate_pagecache(inode, newsize);
721 EXPORT_SYMBOL(truncate_setsize);
724 * pagecache_isize_extended - update pagecache after extension of i_size
725 * @inode: inode for which i_size was extended
726 * @from: original inode size
727 * @to: new inode size
729 * Handle extension of inode size either caused by extending truncate or by
730 * write starting after current i_size. We mark the page straddling current
731 * i_size RO so that page_mkwrite() is called on the nearest write access to
732 * the page. This way filesystem can be sure that page_mkwrite() is called on
733 * the page before user writes to the page via mmap after the i_size has been
736 * The function must be called after i_size is updated so that page fault
737 * coming after we unlock the page will already see the new i_size.
738 * The function must be called while we still hold i_mutex - this not only
739 * makes sure i_size is stable but also that userspace cannot observe new
740 * i_size value before we are prepared to store mmap writes at new inode size.
742 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
744 int bsize = 1 << inode->i_blkbits;
749 WARN_ON(to > inode->i_size);
751 if (from >= to || bsize == PAGE_CACHE_SIZE)
753 /* Page straddling @from will not have any hole block created? */
754 rounded_from = round_up(from, bsize);
755 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
758 index = from >> PAGE_CACHE_SHIFT;
759 page = find_lock_page(inode->i_mapping, index);
760 /* Page not cached? Nothing to do */
764 * See clear_page_dirty_for_io() for details why set_page_dirty()
767 if (page_mkclean(page))
768 set_page_dirty(page);
770 page_cache_release(page);
772 EXPORT_SYMBOL(pagecache_isize_extended);
775 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
777 * @lstart: offset of beginning of hole
778 * @lend: offset of last byte of hole
780 * This function should typically be called before the filesystem
781 * releases resources associated with the freed range (eg. deallocates
782 * blocks). This way, pagecache will always stay logically coherent
783 * with on-disk format, and the filesystem would not have to deal with
784 * situations such as writepage being called for a page that has already
785 * had its underlying blocks deallocated.
787 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
789 struct address_space *mapping = inode->i_mapping;
790 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
791 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
793 * This rounding is currently just for example: unmap_mapping_range
794 * expands its hole outwards, whereas we want it to contract the hole
795 * inwards. However, existing callers of truncate_pagecache_range are
796 * doing their own page rounding first. Note that unmap_mapping_range
797 * allows holelen 0 for all, and we allow lend -1 for end of file.
801 * Unlike in truncate_pagecache, unmap_mapping_range is called only
802 * once (before truncating pagecache), and without "even_cows" flag:
803 * hole-punching should not remove private COWed pages from the hole.
805 if ((u64)unmap_end > (u64)unmap_start)
806 unmap_mapping_range(mapping, unmap_start,
807 1 + unmap_end - unmap_start, 0);
808 truncate_inode_pages_range(mapping, lstart, lend);
810 EXPORT_SYMBOL(truncate_pagecache_range);