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/gfp.h>
14 #include <linux/swap.h>
15 #include <linux/export.h>
16 #include <linux/pagemap.h>
17 #include <linux/highmem.h>
18 #include <linux/pagevec.h>
19 #include <linux/task_io_accounting_ops.h>
20 #include <linux/buffer_head.h> /* grr. try_to_release_page,
22 #include <linux/cleancache.h>
23 #include <linux/rmap.h>
26 static void clear_exceptional_entry(struct address_space *mapping,
27 pgoff_t index, void *entry)
29 struct radix_tree_node *node;
32 /* Handled by shmem itself */
33 if (shmem_mapping(mapping))
36 spin_lock_irq(&mapping->tree_lock);
38 * Regular page slots are stabilized by the page lock even
39 * without the tree itself locked. These unlocked entries
40 * need verification under the tree lock.
42 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
46 radix_tree_replace_slot(slot, NULL);
50 workingset_node_shadows_dec(node);
52 * Don't track node without shadow entries.
54 * Avoid acquiring the list_lru lock if already untracked.
55 * The list_empty() test is safe as node->private_list is
56 * protected by mapping->tree_lock.
58 if (!workingset_node_shadows(node) &&
59 !list_empty(&node->private_list))
60 list_lru_del(&workingset_shadow_nodes, &node->private_list);
61 __radix_tree_delete_node(&mapping->page_tree, node);
63 spin_unlock_irq(&mapping->tree_lock);
67 * do_invalidatepage - invalidate part or all of a page
68 * @page: the page which is affected
69 * @offset: start of the range to invalidate
70 * @length: length of the range to invalidate
72 * do_invalidatepage() is called when all or part of the page has become
73 * invalidated by a truncate operation.
75 * do_invalidatepage() does not have to release all buffers, but it must
76 * ensure that no dirty buffer is left outside @offset and that no I/O
77 * is underway against any of the blocks which are outside the truncation
78 * point. Because the caller is about to free (and possibly reuse) those
81 void do_invalidatepage(struct page *page, unsigned int offset,
84 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
86 invalidatepage = page->mapping->a_ops->invalidatepage;
89 invalidatepage = block_invalidatepage;
92 (*invalidatepage)(page, offset, length);
96 * If truncate cannot remove the fs-private metadata from the page, the page
97 * becomes orphaned. It will be left on the LRU and may even be mapped into
98 * user pagetables if we're racing with filemap_fault().
100 * We need to bale out if page->mapping is no longer equal to the original
101 * mapping. This happens a) when the VM reclaimed the page while we waited on
102 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
103 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
106 truncate_complete_page(struct address_space *mapping, struct page *page)
108 if (page->mapping != mapping)
111 if (page_has_private(page))
112 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
115 * Some filesystems seem to re-dirty the page even after
116 * the VM has canceled the dirty bit (eg ext3 journaling).
117 * Hence dirty accounting check is placed after invalidation.
119 if (TestClearPageDirty(page))
120 account_page_cleaned(page, mapping);
122 ClearPageMappedToDisk(page);
123 delete_from_page_cache(page);
128 * This is for invalidate_mapping_pages(). That function can be called at
129 * any time, and is not supposed to throw away dirty pages. But pages can
130 * be marked dirty at any time too, so use remove_mapping which safely
131 * discards clean, unused pages.
133 * Returns non-zero if the page was successfully invalidated.
136 invalidate_complete_page(struct address_space *mapping, struct page *page)
140 if (page->mapping != mapping)
143 if (page_has_private(page) && !try_to_release_page(page, 0))
146 ret = remove_mapping(mapping, page);
151 int truncate_inode_page(struct address_space *mapping, struct page *page)
153 if (page_mapped(page)) {
154 unmap_mapping_range(mapping,
155 (loff_t)page->index << PAGE_CACHE_SHIFT,
158 return truncate_complete_page(mapping, page);
162 * Used to get rid of pages on hardware memory corruption.
164 int generic_error_remove_page(struct address_space *mapping, struct page *page)
169 * Only punch for normal data pages for now.
170 * Handling other types like directories would need more auditing.
172 if (!S_ISREG(mapping->host->i_mode))
174 return truncate_inode_page(mapping, page);
176 EXPORT_SYMBOL(generic_error_remove_page);
179 * Safely invalidate one page from its pagecache mapping.
180 * It only drops clean, unused pages. The page must be locked.
182 * Returns 1 if the page is successfully invalidated, otherwise 0.
184 int invalidate_inode_page(struct page *page)
186 struct address_space *mapping = page_mapping(page);
189 if (PageDirty(page) || PageWriteback(page))
191 if (page_mapped(page))
193 return invalidate_complete_page(mapping, page);
197 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
198 * @mapping: mapping to truncate
199 * @lstart: offset from which to truncate
200 * @lend: offset to which to truncate (inclusive)
202 * Truncate the page cache, removing the pages that are between
203 * specified offsets (and zeroing out partial pages
204 * if lstart or lend + 1 is not page aligned).
206 * Truncate takes two passes - the first pass is nonblocking. It will not
207 * block on page locks and it will not block on writeback. The second pass
208 * will wait. This is to prevent as much IO as possible in the affected region.
209 * The first pass will remove most pages, so the search cost of the second pass
212 * We pass down the cache-hot hint to the page freeing code. Even if the
213 * mapping is large, it is probably the case that the final pages are the most
214 * recently touched, and freeing happens in ascending file offset order.
216 * Note that since ->invalidatepage() accepts range to invalidate
217 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
218 * page aligned properly.
220 void truncate_inode_pages_range(struct address_space *mapping,
221 loff_t lstart, loff_t lend)
223 pgoff_t start; /* inclusive */
224 pgoff_t end; /* exclusive */
225 unsigned int partial_start; /* inclusive */
226 unsigned int partial_end; /* exclusive */
228 pgoff_t indices[PAGEVEC_SIZE];
232 cleancache_invalidate_inode(mapping);
233 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
236 /* Offsets within partial pages */
237 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
238 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
241 * 'start' and 'end' always covers the range of pages to be fully
242 * truncated. Partial pages are covered with 'partial_start' at the
243 * start of the range and 'partial_end' at the end of the range.
244 * Note that 'end' is exclusive while 'lend' is inclusive.
246 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
249 * lend == -1 indicates end-of-file so we have to set 'end'
250 * to the highest possible pgoff_t and since the type is
251 * unsigned we're using -1.
255 end = (lend + 1) >> PAGE_CACHE_SHIFT;
257 pagevec_init(&pvec, 0);
259 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
260 min(end - index, (pgoff_t)PAGEVEC_SIZE),
262 for (i = 0; i < pagevec_count(&pvec); i++) {
263 struct page *page = pvec.pages[i];
265 /* We rely upon deletion not changing page->index */
270 if (radix_tree_exceptional_entry(page)) {
271 clear_exceptional_entry(mapping, index, page);
275 if (!trylock_page(page))
277 WARN_ON(page->index != index);
278 if (PageWriteback(page)) {
282 truncate_inode_page(mapping, page);
285 pagevec_remove_exceptionals(&pvec);
286 pagevec_release(&pvec);
292 struct page *page = find_lock_page(mapping, start - 1);
294 unsigned int top = PAGE_CACHE_SIZE;
296 /* Truncation within a single page */
300 wait_on_page_writeback(page);
301 zero_user_segment(page, partial_start, top);
302 cleancache_invalidate_page(mapping, page);
303 if (page_has_private(page))
304 do_invalidatepage(page, partial_start,
305 top - partial_start);
307 page_cache_release(page);
311 struct page *page = find_lock_page(mapping, end);
313 wait_on_page_writeback(page);
314 zero_user_segment(page, 0, partial_end);
315 cleancache_invalidate_page(mapping, page);
316 if (page_has_private(page))
317 do_invalidatepage(page, 0,
320 page_cache_release(page);
324 * If the truncation happened within a single page no pages
325 * will be released, just zeroed, so we can bail out now.
333 if (!pagevec_lookup_entries(&pvec, mapping, index,
334 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
335 /* If all gone from start onwards, we're done */
338 /* Otherwise restart to make sure all gone */
342 if (index == start && indices[0] >= end) {
343 /* All gone out of hole to be punched, we're done */
344 pagevec_remove_exceptionals(&pvec);
345 pagevec_release(&pvec);
348 for (i = 0; i < pagevec_count(&pvec); i++) {
349 struct page *page = pvec.pages[i];
351 /* We rely upon deletion not changing page->index */
354 /* Restart punch to make sure all gone */
359 if (radix_tree_exceptional_entry(page)) {
360 clear_exceptional_entry(mapping, index, page);
365 WARN_ON(page->index != index);
366 wait_on_page_writeback(page);
367 truncate_inode_page(mapping, page);
370 pagevec_remove_exceptionals(&pvec);
371 pagevec_release(&pvec);
374 cleancache_invalidate_inode(mapping);
376 EXPORT_SYMBOL(truncate_inode_pages_range);
379 * truncate_inode_pages - truncate *all* the pages from an offset
380 * @mapping: mapping to truncate
381 * @lstart: offset from which to truncate
383 * Called under (and serialised by) inode->i_mutex.
385 * Note: When this function returns, there can be a page in the process of
386 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
387 * mapping->nrpages can be non-zero when this function returns even after
388 * truncation of the whole mapping.
390 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
392 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
394 EXPORT_SYMBOL(truncate_inode_pages);
397 * truncate_inode_pages_final - truncate *all* pages before inode dies
398 * @mapping: mapping to truncate
400 * Called under (and serialized by) inode->i_mutex.
402 * Filesystems have to use this in the .evict_inode path to inform the
403 * VM that this is the final truncate and the inode is going away.
405 void truncate_inode_pages_final(struct address_space *mapping)
407 unsigned long nrshadows;
408 unsigned long nrpages;
411 * Page reclaim can not participate in regular inode lifetime
412 * management (can't call iput()) and thus can race with the
413 * inode teardown. Tell it when the address space is exiting,
414 * so that it does not install eviction information after the
415 * final truncate has begun.
417 mapping_set_exiting(mapping);
420 * When reclaim installs eviction entries, it increases
421 * nrshadows first, then decreases nrpages. Make sure we see
422 * this in the right order or we might miss an entry.
424 nrpages = mapping->nrpages;
426 nrshadows = mapping->nrshadows;
428 if (nrpages || nrshadows) {
430 * As truncation uses a lockless tree lookup, cycle
431 * the tree lock to make sure any ongoing tree
432 * modification that does not see AS_EXITING is
433 * completed before starting the final truncate.
435 spin_lock_irq(&mapping->tree_lock);
436 spin_unlock_irq(&mapping->tree_lock);
438 truncate_inode_pages(mapping, 0);
441 EXPORT_SYMBOL(truncate_inode_pages_final);
444 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
445 * @mapping: the address_space which holds the pages to invalidate
446 * @start: the offset 'from' which to invalidate
447 * @end: the offset 'to' which to invalidate (inclusive)
449 * This function only removes the unlocked pages, if you want to
450 * remove all the pages of one inode, you must call truncate_inode_pages.
452 * invalidate_mapping_pages() will not block on IO activity. It will not
453 * invalidate pages which are dirty, locked, under writeback or mapped into
456 unsigned long invalidate_mapping_pages(struct address_space *mapping,
457 pgoff_t start, pgoff_t end)
459 pgoff_t indices[PAGEVEC_SIZE];
461 pgoff_t index = start;
463 unsigned long count = 0;
466 pagevec_init(&pvec, 0);
467 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
468 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
470 for (i = 0; i < pagevec_count(&pvec); i++) {
471 struct page *page = pvec.pages[i];
473 /* We rely upon deletion not changing page->index */
478 if (radix_tree_exceptional_entry(page)) {
479 clear_exceptional_entry(mapping, index, page);
483 if (!trylock_page(page))
485 WARN_ON(page->index != index);
486 ret = invalidate_inode_page(page);
489 * Invalidation is a hint that the page is no longer
490 * of interest and try to speed up its reclaim.
493 deactivate_file_page(page);
496 pagevec_remove_exceptionals(&pvec);
497 pagevec_release(&pvec);
503 EXPORT_SYMBOL(invalidate_mapping_pages);
506 * This is like invalidate_complete_page(), except it ignores the page's
507 * refcount. We do this because invalidate_inode_pages2() needs stronger
508 * invalidation guarantees, and cannot afford to leave pages behind because
509 * shrink_page_list() has a temp ref on them, or because they're transiently
510 * sitting in the lru_cache_add() pagevecs.
513 invalidate_complete_page2(struct address_space *mapping, struct page *page)
515 if (page->mapping != mapping)
518 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
521 spin_lock_irq(&mapping->tree_lock);
525 BUG_ON(page_has_private(page));
526 __delete_from_page_cache(page, NULL);
527 spin_unlock_irq(&mapping->tree_lock);
529 if (mapping->a_ops->freepage)
530 mapping->a_ops->freepage(page);
532 page_cache_release(page); /* pagecache ref */
535 spin_unlock_irq(&mapping->tree_lock);
539 static int do_launder_page(struct address_space *mapping, struct page *page)
541 if (!PageDirty(page))
543 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
545 return mapping->a_ops->launder_page(page);
549 * invalidate_inode_pages2_range - remove range of pages from an address_space
550 * @mapping: the address_space
551 * @start: the page offset 'from' which to invalidate
552 * @end: the page offset 'to' which to invalidate (inclusive)
554 * Any pages which are found to be mapped into pagetables are unmapped prior to
557 * Returns -EBUSY if any pages could not be invalidated.
559 int invalidate_inode_pages2_range(struct address_space *mapping,
560 pgoff_t start, pgoff_t end)
562 pgoff_t indices[PAGEVEC_SIZE];
568 int did_range_unmap = 0;
570 cleancache_invalidate_inode(mapping);
571 pagevec_init(&pvec, 0);
573 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
574 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
576 for (i = 0; i < pagevec_count(&pvec); i++) {
577 struct page *page = pvec.pages[i];
579 /* We rely upon deletion not changing page->index */
584 if (radix_tree_exceptional_entry(page)) {
585 clear_exceptional_entry(mapping, index, page);
590 WARN_ON(page->index != index);
591 if (page->mapping != mapping) {
595 wait_on_page_writeback(page);
596 if (page_mapped(page)) {
597 if (!did_range_unmap) {
599 * Zap the rest of the file in one hit.
601 unmap_mapping_range(mapping,
602 (loff_t)index << PAGE_CACHE_SHIFT,
603 (loff_t)(1 + end - index)
611 unmap_mapping_range(mapping,
612 (loff_t)index << PAGE_CACHE_SHIFT,
616 BUG_ON(page_mapped(page));
617 ret2 = do_launder_page(mapping, page);
619 if (!invalidate_complete_page2(mapping, page))
626 pagevec_remove_exceptionals(&pvec);
627 pagevec_release(&pvec);
631 cleancache_invalidate_inode(mapping);
634 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
637 * invalidate_inode_pages2 - remove all pages from an address_space
638 * @mapping: the address_space
640 * Any pages which are found to be mapped into pagetables are unmapped prior to
643 * Returns -EBUSY if any pages could not be invalidated.
645 int invalidate_inode_pages2(struct address_space *mapping)
647 return invalidate_inode_pages2_range(mapping, 0, -1);
649 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
652 * truncate_pagecache - unmap and remove pagecache that has been truncated
654 * @newsize: new file size
656 * inode's new i_size must already be written before truncate_pagecache
659 * This function should typically be called before the filesystem
660 * releases resources associated with the freed range (eg. deallocates
661 * blocks). This way, pagecache will always stay logically coherent
662 * with on-disk format, and the filesystem would not have to deal with
663 * situations such as writepage being called for a page that has already
664 * had its underlying blocks deallocated.
666 void truncate_pagecache(struct inode *inode, loff_t newsize)
668 struct address_space *mapping = inode->i_mapping;
669 loff_t holebegin = round_up(newsize, PAGE_SIZE);
672 * unmap_mapping_range is called twice, first simply for
673 * efficiency so that truncate_inode_pages does fewer
674 * single-page unmaps. However after this first call, and
675 * before truncate_inode_pages finishes, it is possible for
676 * private pages to be COWed, which remain after
677 * truncate_inode_pages finishes, hence the second
678 * unmap_mapping_range call must be made for correctness.
680 unmap_mapping_range(mapping, holebegin, 0, 1);
681 truncate_inode_pages(mapping, newsize);
682 unmap_mapping_range(mapping, holebegin, 0, 1);
684 EXPORT_SYMBOL(truncate_pagecache);
687 * truncate_setsize - update inode and pagecache for a new file size
689 * @newsize: new file size
691 * truncate_setsize updates i_size and performs pagecache truncation (if
692 * necessary) to @newsize. It will be typically be called from the filesystem's
693 * setattr function when ATTR_SIZE is passed in.
695 * Must be called with a lock serializing truncates and writes (generally
696 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
697 * specific block truncation has been performed.
699 void truncate_setsize(struct inode *inode, loff_t newsize)
701 loff_t oldsize = inode->i_size;
703 i_size_write(inode, newsize);
704 if (newsize > oldsize)
705 pagecache_isize_extended(inode, oldsize, newsize);
706 truncate_pagecache(inode, newsize);
708 EXPORT_SYMBOL(truncate_setsize);
711 * pagecache_isize_extended - update pagecache after extension of i_size
712 * @inode: inode for which i_size was extended
713 * @from: original inode size
714 * @to: new inode size
716 * Handle extension of inode size either caused by extending truncate or by
717 * write starting after current i_size. We mark the page straddling current
718 * i_size RO so that page_mkwrite() is called on the nearest write access to
719 * the page. This way filesystem can be sure that page_mkwrite() is called on
720 * the page before user writes to the page via mmap after the i_size has been
723 * The function must be called after i_size is updated so that page fault
724 * coming after we unlock the page will already see the new i_size.
725 * The function must be called while we still hold i_mutex - this not only
726 * makes sure i_size is stable but also that userspace cannot observe new
727 * i_size value before we are prepared to store mmap writes at new inode size.
729 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
731 int bsize = 1 << inode->i_blkbits;
736 WARN_ON(to > inode->i_size);
738 if (from >= to || bsize == PAGE_CACHE_SIZE)
740 /* Page straddling @from will not have any hole block created? */
741 rounded_from = round_up(from, bsize);
742 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
745 index = from >> PAGE_CACHE_SHIFT;
746 page = find_lock_page(inode->i_mapping, index);
747 /* Page not cached? Nothing to do */
751 * See clear_page_dirty_for_io() for details why set_page_dirty()
754 if (page_mkclean(page))
755 set_page_dirty(page);
757 page_cache_release(page);
759 EXPORT_SYMBOL(pagecache_isize_extended);
762 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
764 * @lstart: offset of beginning of hole
765 * @lend: offset of last byte of hole
767 * This function should typically be called before the filesystem
768 * releases resources associated with the freed range (eg. deallocates
769 * blocks). This way, pagecache will always stay logically coherent
770 * with on-disk format, and the filesystem would not have to deal with
771 * situations such as writepage being called for a page that has already
772 * had its underlying blocks deallocated.
774 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
776 struct address_space *mapping = inode->i_mapping;
777 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
778 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
780 * This rounding is currently just for example: unmap_mapping_range
781 * expands its hole outwards, whereas we want it to contract the hole
782 * inwards. However, existing callers of truncate_pagecache_range are
783 * doing their own page rounding first. Note that unmap_mapping_range
784 * allows holelen 0 for all, and we allow lend -1 for end of file.
788 * Unlike in truncate_pagecache, unmap_mapping_range is called only
789 * once (before truncating pagecache), and without "even_cows" flag:
790 * hole-punching should not remove private COWed pages from the hole.
792 if ((u64)unmap_end > (u64)unmap_start)
793 unmap_mapping_range(mapping, unmap_start,
794 1 + unmap_end - unmap_start, 0);
795 truncate_inode_pages_range(mapping, lstart, lend);
797 EXPORT_SYMBOL(truncate_pagecache_range);