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 local_lock(workingset_shadow_lock);
61 list_lru_del(&__workingset_shadow_nodes, &node->private_list);
62 local_unlock(workingset_shadow_lock);
64 __radix_tree_delete_node(&mapping->page_tree, node);
66 spin_unlock_irq(&mapping->tree_lock);
70 * do_invalidatepage - invalidate part or all of a page
71 * @page: the page which is affected
72 * @offset: start of the range to invalidate
73 * @length: length of the range to invalidate
75 * do_invalidatepage() is called when all or part of the page has become
76 * invalidated by a truncate operation.
78 * do_invalidatepage() does not have to release all buffers, but it must
79 * ensure that no dirty buffer is left outside @offset and that no I/O
80 * is underway against any of the blocks which are outside the truncation
81 * point. Because the caller is about to free (and possibly reuse) those
84 void do_invalidatepage(struct page *page, unsigned int offset,
87 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
89 invalidatepage = page->mapping->a_ops->invalidatepage;
92 invalidatepage = block_invalidatepage;
95 (*invalidatepage)(page, offset, length);
99 * If truncate cannot remove the fs-private metadata from the page, the page
100 * becomes orphaned. It will be left on the LRU and may even be mapped into
101 * user pagetables if we're racing with filemap_fault().
103 * We need to bale out if page->mapping is no longer equal to the original
104 * mapping. This happens a) when the VM reclaimed the page while we waited on
105 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
106 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
109 truncate_complete_page(struct address_space *mapping, struct page *page)
111 if (page->mapping != mapping)
114 if (page_has_private(page))
115 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
118 * Some filesystems seem to re-dirty the page even after
119 * the VM has canceled the dirty bit (eg ext3 journaling).
120 * Hence dirty accounting check is placed after invalidation.
122 if (TestClearPageDirty(page))
123 account_page_cleaned(page, mapping);
125 ClearPageMappedToDisk(page);
126 delete_from_page_cache(page);
131 * This is for invalidate_mapping_pages(). That function can be called at
132 * any time, and is not supposed to throw away dirty pages. But pages can
133 * be marked dirty at any time too, so use remove_mapping which safely
134 * discards clean, unused pages.
136 * Returns non-zero if the page was successfully invalidated.
139 invalidate_complete_page(struct address_space *mapping, struct page *page)
143 if (page->mapping != mapping)
146 if (page_has_private(page) && !try_to_release_page(page, 0))
149 ret = remove_mapping(mapping, page);
154 int truncate_inode_page(struct address_space *mapping, struct page *page)
156 if (page_mapped(page)) {
157 unmap_mapping_range(mapping,
158 (loff_t)page->index << PAGE_CACHE_SHIFT,
161 return truncate_complete_page(mapping, page);
165 * Used to get rid of pages on hardware memory corruption.
167 int generic_error_remove_page(struct address_space *mapping, struct page *page)
172 * Only punch for normal data pages for now.
173 * Handling other types like directories would need more auditing.
175 if (!S_ISREG(mapping->host->i_mode))
177 return truncate_inode_page(mapping, page);
179 EXPORT_SYMBOL(generic_error_remove_page);
182 * Safely invalidate one page from its pagecache mapping.
183 * It only drops clean, unused pages. The page must be locked.
185 * Returns 1 if the page is successfully invalidated, otherwise 0.
187 int invalidate_inode_page(struct page *page)
189 struct address_space *mapping = page_mapping(page);
192 if (PageDirty(page) || PageWriteback(page))
194 if (page_mapped(page))
196 return invalidate_complete_page(mapping, page);
200 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
201 * @mapping: mapping to truncate
202 * @lstart: offset from which to truncate
203 * @lend: offset to which to truncate (inclusive)
205 * Truncate the page cache, removing the pages that are between
206 * specified offsets (and zeroing out partial pages
207 * if lstart or lend + 1 is not page aligned).
209 * Truncate takes two passes - the first pass is nonblocking. It will not
210 * block on page locks and it will not block on writeback. The second pass
211 * will wait. This is to prevent as much IO as possible in the affected region.
212 * The first pass will remove most pages, so the search cost of the second pass
215 * We pass down the cache-hot hint to the page freeing code. Even if the
216 * mapping is large, it is probably the case that the final pages are the most
217 * recently touched, and freeing happens in ascending file offset order.
219 * Note that since ->invalidatepage() accepts range to invalidate
220 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
221 * page aligned properly.
223 void truncate_inode_pages_range(struct address_space *mapping,
224 loff_t lstart, loff_t lend)
226 pgoff_t start; /* inclusive */
227 pgoff_t end; /* exclusive */
228 unsigned int partial_start; /* inclusive */
229 unsigned int partial_end; /* exclusive */
231 pgoff_t indices[PAGEVEC_SIZE];
235 cleancache_invalidate_inode(mapping);
236 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
239 /* Offsets within partial pages */
240 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
241 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
244 * 'start' and 'end' always covers the range of pages to be fully
245 * truncated. Partial pages are covered with 'partial_start' at the
246 * start of the range and 'partial_end' at the end of the range.
247 * Note that 'end' is exclusive while 'lend' is inclusive.
249 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
252 * lend == -1 indicates end-of-file so we have to set 'end'
253 * to the highest possible pgoff_t and since the type is
254 * unsigned we're using -1.
258 end = (lend + 1) >> PAGE_CACHE_SHIFT;
260 pagevec_init(&pvec, 0);
262 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
263 min(end - index, (pgoff_t)PAGEVEC_SIZE),
265 for (i = 0; i < pagevec_count(&pvec); i++) {
266 struct page *page = pvec.pages[i];
268 /* We rely upon deletion not changing page->index */
273 if (radix_tree_exceptional_entry(page)) {
274 clear_exceptional_entry(mapping, index, page);
278 if (!trylock_page(page))
280 WARN_ON(page->index != index);
281 if (PageWriteback(page)) {
285 truncate_inode_page(mapping, page);
288 pagevec_remove_exceptionals(&pvec);
289 pagevec_release(&pvec);
295 struct page *page = find_lock_page(mapping, start - 1);
297 unsigned int top = PAGE_CACHE_SIZE;
299 /* Truncation within a single page */
303 wait_on_page_writeback(page);
304 zero_user_segment(page, partial_start, top);
305 cleancache_invalidate_page(mapping, page);
306 if (page_has_private(page))
307 do_invalidatepage(page, partial_start,
308 top - partial_start);
310 page_cache_release(page);
314 struct page *page = find_lock_page(mapping, end);
316 wait_on_page_writeback(page);
317 zero_user_segment(page, 0, partial_end);
318 cleancache_invalidate_page(mapping, page);
319 if (page_has_private(page))
320 do_invalidatepage(page, 0,
323 page_cache_release(page);
327 * If the truncation happened within a single page no pages
328 * will be released, just zeroed, so we can bail out now.
336 if (!pagevec_lookup_entries(&pvec, mapping, index,
337 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
338 /* If all gone from start onwards, we're done */
341 /* Otherwise restart to make sure all gone */
345 if (index == start && indices[0] >= end) {
346 /* All gone out of hole to be punched, we're done */
347 pagevec_remove_exceptionals(&pvec);
348 pagevec_release(&pvec);
351 for (i = 0; i < pagevec_count(&pvec); i++) {
352 struct page *page = pvec.pages[i];
354 /* We rely upon deletion not changing page->index */
357 /* Restart punch to make sure all gone */
362 if (radix_tree_exceptional_entry(page)) {
363 clear_exceptional_entry(mapping, index, page);
368 WARN_ON(page->index != index);
369 wait_on_page_writeback(page);
370 truncate_inode_page(mapping, page);
373 pagevec_remove_exceptionals(&pvec);
374 pagevec_release(&pvec);
377 cleancache_invalidate_inode(mapping);
379 EXPORT_SYMBOL(truncate_inode_pages_range);
382 * truncate_inode_pages - truncate *all* the pages from an offset
383 * @mapping: mapping to truncate
384 * @lstart: offset from which to truncate
386 * Called under (and serialised by) inode->i_mutex.
388 * Note: When this function returns, there can be a page in the process of
389 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
390 * mapping->nrpages can be non-zero when this function returns even after
391 * truncation of the whole mapping.
393 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
395 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
397 EXPORT_SYMBOL(truncate_inode_pages);
400 * truncate_inode_pages_final - truncate *all* pages before inode dies
401 * @mapping: mapping to truncate
403 * Called under (and serialized by) inode->i_mutex.
405 * Filesystems have to use this in the .evict_inode path to inform the
406 * VM that this is the final truncate and the inode is going away.
408 void truncate_inode_pages_final(struct address_space *mapping)
410 unsigned long nrshadows;
411 unsigned long nrpages;
414 * Page reclaim can not participate in regular inode lifetime
415 * management (can't call iput()) and thus can race with the
416 * inode teardown. Tell it when the address space is exiting,
417 * so that it does not install eviction information after the
418 * final truncate has begun.
420 mapping_set_exiting(mapping);
423 * When reclaim installs eviction entries, it increases
424 * nrshadows first, then decreases nrpages. Make sure we see
425 * this in the right order or we might miss an entry.
427 nrpages = mapping->nrpages;
429 nrshadows = mapping->nrshadows;
431 if (nrpages || nrshadows) {
433 * As truncation uses a lockless tree lookup, cycle
434 * the tree lock to make sure any ongoing tree
435 * modification that does not see AS_EXITING is
436 * completed before starting the final truncate.
438 spin_lock_irq(&mapping->tree_lock);
439 spin_unlock_irq(&mapping->tree_lock);
441 truncate_inode_pages(mapping, 0);
444 EXPORT_SYMBOL(truncate_inode_pages_final);
447 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
448 * @mapping: the address_space which holds the pages to invalidate
449 * @start: the offset 'from' which to invalidate
450 * @end: the offset 'to' which to invalidate (inclusive)
452 * This function only removes the unlocked pages, if you want to
453 * remove all the pages of one inode, you must call truncate_inode_pages.
455 * invalidate_mapping_pages() will not block on IO activity. It will not
456 * invalidate pages which are dirty, locked, under writeback or mapped into
459 unsigned long invalidate_mapping_pages(struct address_space *mapping,
460 pgoff_t start, pgoff_t end)
462 pgoff_t indices[PAGEVEC_SIZE];
464 pgoff_t index = start;
466 unsigned long count = 0;
469 pagevec_init(&pvec, 0);
470 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
471 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
473 for (i = 0; i < pagevec_count(&pvec); i++) {
474 struct page *page = pvec.pages[i];
476 /* We rely upon deletion not changing page->index */
481 if (radix_tree_exceptional_entry(page)) {
482 clear_exceptional_entry(mapping, index, page);
486 if (!trylock_page(page))
488 WARN_ON(page->index != index);
489 ret = invalidate_inode_page(page);
492 * Invalidation is a hint that the page is no longer
493 * of interest and try to speed up its reclaim.
496 deactivate_file_page(page);
499 pagevec_remove_exceptionals(&pvec);
500 pagevec_release(&pvec);
506 EXPORT_SYMBOL(invalidate_mapping_pages);
509 * This is like invalidate_complete_page(), except it ignores the page's
510 * refcount. We do this because invalidate_inode_pages2() needs stronger
511 * invalidation guarantees, and cannot afford to leave pages behind because
512 * shrink_page_list() has a temp ref on them, or because they're transiently
513 * sitting in the lru_cache_add() pagevecs.
516 invalidate_complete_page2(struct address_space *mapping, struct page *page)
518 if (page->mapping != mapping)
521 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
524 spin_lock_irq(&mapping->tree_lock);
528 BUG_ON(page_has_private(page));
529 __delete_from_page_cache(page, NULL);
530 spin_unlock_irq(&mapping->tree_lock);
532 if (mapping->a_ops->freepage)
533 mapping->a_ops->freepage(page);
535 page_cache_release(page); /* pagecache ref */
538 spin_unlock_irq(&mapping->tree_lock);
542 static int do_launder_page(struct address_space *mapping, struct page *page)
544 if (!PageDirty(page))
546 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
548 return mapping->a_ops->launder_page(page);
552 * invalidate_inode_pages2_range - remove range of pages from an address_space
553 * @mapping: the address_space
554 * @start: the page offset 'from' which to invalidate
555 * @end: the page offset 'to' which to invalidate (inclusive)
557 * Any pages which are found to be mapped into pagetables are unmapped prior to
560 * Returns -EBUSY if any pages could not be invalidated.
562 int invalidate_inode_pages2_range(struct address_space *mapping,
563 pgoff_t start, pgoff_t end)
565 pgoff_t indices[PAGEVEC_SIZE];
571 int did_range_unmap = 0;
573 cleancache_invalidate_inode(mapping);
574 pagevec_init(&pvec, 0);
576 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
577 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
579 for (i = 0; i < pagevec_count(&pvec); i++) {
580 struct page *page = pvec.pages[i];
582 /* We rely upon deletion not changing page->index */
587 if (radix_tree_exceptional_entry(page)) {
588 clear_exceptional_entry(mapping, index, page);
593 WARN_ON(page->index != index);
594 if (page->mapping != mapping) {
598 wait_on_page_writeback(page);
599 if (page_mapped(page)) {
600 if (!did_range_unmap) {
602 * Zap the rest of the file in one hit.
604 unmap_mapping_range(mapping,
605 (loff_t)index << PAGE_CACHE_SHIFT,
606 (loff_t)(1 + end - index)
614 unmap_mapping_range(mapping,
615 (loff_t)index << PAGE_CACHE_SHIFT,
619 BUG_ON(page_mapped(page));
620 ret2 = do_launder_page(mapping, page);
622 if (!invalidate_complete_page2(mapping, page))
629 pagevec_remove_exceptionals(&pvec);
630 pagevec_release(&pvec);
634 cleancache_invalidate_inode(mapping);
637 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
640 * invalidate_inode_pages2 - remove all pages from an address_space
641 * @mapping: the address_space
643 * Any pages which are found to be mapped into pagetables are unmapped prior to
646 * Returns -EBUSY if any pages could not be invalidated.
648 int invalidate_inode_pages2(struct address_space *mapping)
650 return invalidate_inode_pages2_range(mapping, 0, -1);
652 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
655 * truncate_pagecache - unmap and remove pagecache that has been truncated
657 * @newsize: new file size
659 * inode's new i_size must already be written before truncate_pagecache
662 * This function should typically be called before the filesystem
663 * releases resources associated with the freed range (eg. deallocates
664 * blocks). This way, pagecache will always stay logically coherent
665 * with on-disk format, and the filesystem would not have to deal with
666 * situations such as writepage being called for a page that has already
667 * had its underlying blocks deallocated.
669 void truncate_pagecache(struct inode *inode, loff_t newsize)
671 struct address_space *mapping = inode->i_mapping;
672 loff_t holebegin = round_up(newsize, PAGE_SIZE);
675 * unmap_mapping_range is called twice, first simply for
676 * efficiency so that truncate_inode_pages does fewer
677 * single-page unmaps. However after this first call, and
678 * before truncate_inode_pages finishes, it is possible for
679 * private pages to be COWed, which remain after
680 * truncate_inode_pages finishes, hence the second
681 * unmap_mapping_range call must be made for correctness.
683 unmap_mapping_range(mapping, holebegin, 0, 1);
684 truncate_inode_pages(mapping, newsize);
685 unmap_mapping_range(mapping, holebegin, 0, 1);
687 EXPORT_SYMBOL(truncate_pagecache);
690 * truncate_setsize - update inode and pagecache for a new file size
692 * @newsize: new file size
694 * truncate_setsize updates i_size and performs pagecache truncation (if
695 * necessary) to @newsize. It will be typically be called from the filesystem's
696 * setattr function when ATTR_SIZE is passed in.
698 * Must be called with a lock serializing truncates and writes (generally
699 * i_mutex but e.g. xfs uses a different lock) and before all filesystem
700 * specific block truncation has been performed.
702 void truncate_setsize(struct inode *inode, loff_t newsize)
704 loff_t oldsize = inode->i_size;
706 i_size_write(inode, newsize);
707 if (newsize > oldsize)
708 pagecache_isize_extended(inode, oldsize, newsize);
709 truncate_pagecache(inode, newsize);
711 EXPORT_SYMBOL(truncate_setsize);
714 * pagecache_isize_extended - update pagecache after extension of i_size
715 * @inode: inode for which i_size was extended
716 * @from: original inode size
717 * @to: new inode size
719 * Handle extension of inode size either caused by extending truncate or by
720 * write starting after current i_size. We mark the page straddling current
721 * i_size RO so that page_mkwrite() is called on the nearest write access to
722 * the page. This way filesystem can be sure that page_mkwrite() is called on
723 * the page before user writes to the page via mmap after the i_size has been
726 * The function must be called after i_size is updated so that page fault
727 * coming after we unlock the page will already see the new i_size.
728 * The function must be called while we still hold i_mutex - this not only
729 * makes sure i_size is stable but also that userspace cannot observe new
730 * i_size value before we are prepared to store mmap writes at new inode size.
732 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
734 int bsize = 1 << inode->i_blkbits;
739 WARN_ON(to > inode->i_size);
741 if (from >= to || bsize == PAGE_CACHE_SIZE)
743 /* Page straddling @from will not have any hole block created? */
744 rounded_from = round_up(from, bsize);
745 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
748 index = from >> PAGE_CACHE_SHIFT;
749 page = find_lock_page(inode->i_mapping, index);
750 /* Page not cached? Nothing to do */
754 * See clear_page_dirty_for_io() for details why set_page_dirty()
757 if (page_mkclean(page))
758 set_page_dirty(page);
760 page_cache_release(page);
762 EXPORT_SYMBOL(pagecache_isize_extended);
765 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
767 * @lstart: offset of beginning of hole
768 * @lend: offset of last byte of hole
770 * This function should typically be called before the filesystem
771 * releases resources associated with the freed range (eg. deallocates
772 * blocks). This way, pagecache will always stay logically coherent
773 * with on-disk format, and the filesystem would not have to deal with
774 * situations such as writepage being called for a page that has already
775 * had its underlying blocks deallocated.
777 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
779 struct address_space *mapping = inode->i_mapping;
780 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
781 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
783 * This rounding is currently just for example: unmap_mapping_range
784 * expands its hole outwards, whereas we want it to contract the hole
785 * inwards. However, existing callers of truncate_pagecache_range are
786 * doing their own page rounding first. Note that unmap_mapping_range
787 * allows holelen 0 for all, and we allow lend -1 for end of file.
791 * Unlike in truncate_pagecache, unmap_mapping_range is called only
792 * once (before truncating pagecache), and without "even_cows" flag:
793 * hole-punching should not remove private COWed pages from the hole.
795 if ((u64)unmap_end > (u64)unmap_start)
796 unmap_mapping_range(mapping, unmap_start,
797 1 + unmap_end - unmap_start, 0);
798 truncate_inode_pages_range(mapping, lstart, lend);
800 EXPORT_SYMBOL(truncate_pagecache_range);