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 * This cancels just the dirty bit on the kernel page itself, it
97 * does NOT actually remove dirty bits on any mmap's that may be
98 * around. It also leaves the page tagged dirty, so any sync
99 * activity will still find it on the dirty lists, and in particular,
100 * clear_page_dirty_for_io() will still look at the dirty bits in
103 * Doing this should *normally* only ever be done when a page
104 * is truncated, and is not actually mapped anywhere at all. However,
105 * fs/buffer.c does this when it notices that somebody has cleaned
106 * out all the buffers on a page without actually doing it through
107 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
109 void cancel_dirty_page(struct page *page, unsigned int account_size)
111 if (TestClearPageDirty(page)) {
112 struct address_space *mapping = page->mapping;
113 if (mapping && mapping_cap_account_dirty(mapping)) {
114 dec_zone_page_state(page, NR_FILE_DIRTY);
115 dec_bdi_stat(mapping->backing_dev_info,
118 task_io_account_cancelled_write(account_size);
122 EXPORT_SYMBOL(cancel_dirty_page);
125 * If truncate cannot remove the fs-private metadata from the page, the page
126 * becomes orphaned. It will be left on the LRU and may even be mapped into
127 * user pagetables if we're racing with filemap_fault().
129 * We need to bale out if page->mapping is no longer equal to the original
130 * mapping. This happens a) when the VM reclaimed the page while we waited on
131 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
132 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
135 truncate_complete_page(struct address_space *mapping, struct page *page)
137 if (page->mapping != mapping)
140 if (page_has_private(page))
141 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
143 cancel_dirty_page(page, PAGE_CACHE_SIZE);
145 ClearPageMappedToDisk(page);
146 delete_from_page_cache(page);
151 * This is for invalidate_mapping_pages(). That function can be called at
152 * any time, and is not supposed to throw away dirty pages. But pages can
153 * be marked dirty at any time too, so use remove_mapping which safely
154 * discards clean, unused pages.
156 * Returns non-zero if the page was successfully invalidated.
159 invalidate_complete_page(struct address_space *mapping, struct page *page)
163 if (page->mapping != mapping)
166 if (page_has_private(page) && !try_to_release_page(page, 0))
169 ret = remove_mapping(mapping, page);
174 int truncate_inode_page(struct address_space *mapping, struct page *page)
176 if (page_mapped(page)) {
177 unmap_mapping_range(mapping,
178 (loff_t)page->index << PAGE_CACHE_SHIFT,
181 return truncate_complete_page(mapping, page);
185 * Used to get rid of pages on hardware memory corruption.
187 int generic_error_remove_page(struct address_space *mapping, struct page *page)
192 * Only punch for normal data pages for now.
193 * Handling other types like directories would need more auditing.
195 if (!S_ISREG(mapping->host->i_mode))
197 return truncate_inode_page(mapping, page);
199 EXPORT_SYMBOL(generic_error_remove_page);
202 * Safely invalidate one page from its pagecache mapping.
203 * It only drops clean, unused pages. The page must be locked.
205 * Returns 1 if the page is successfully invalidated, otherwise 0.
207 int invalidate_inode_page(struct page *page)
209 struct address_space *mapping = page_mapping(page);
212 if (PageDirty(page) || PageWriteback(page))
214 if (page_mapped(page))
216 return invalidate_complete_page(mapping, page);
220 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
221 * @mapping: mapping to truncate
222 * @lstart: offset from which to truncate
223 * @lend: offset to which to truncate (inclusive)
225 * Truncate the page cache, removing the pages that are between
226 * specified offsets (and zeroing out partial pages
227 * if lstart or lend + 1 is not page aligned).
229 * Truncate takes two passes - the first pass is nonblocking. It will not
230 * block on page locks and it will not block on writeback. The second pass
231 * will wait. This is to prevent as much IO as possible in the affected region.
232 * The first pass will remove most pages, so the search cost of the second pass
235 * We pass down the cache-hot hint to the page freeing code. Even if the
236 * mapping is large, it is probably the case that the final pages are the most
237 * recently touched, and freeing happens in ascending file offset order.
239 * Note that since ->invalidatepage() accepts range to invalidate
240 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
241 * page aligned properly.
243 void truncate_inode_pages_range(struct address_space *mapping,
244 loff_t lstart, loff_t lend)
246 pgoff_t start; /* inclusive */
247 pgoff_t end; /* exclusive */
248 unsigned int partial_start; /* inclusive */
249 unsigned int partial_end; /* exclusive */
251 pgoff_t indices[PAGEVEC_SIZE];
255 cleancache_invalidate_inode(mapping);
256 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
259 /* Offsets within partial pages */
260 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
261 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
264 * 'start' and 'end' always covers the range of pages to be fully
265 * truncated. Partial pages are covered with 'partial_start' at the
266 * start of the range and 'partial_end' at the end of the range.
267 * Note that 'end' is exclusive while 'lend' is inclusive.
269 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
272 * lend == -1 indicates end-of-file so we have to set 'end'
273 * to the highest possible pgoff_t and since the type is
274 * unsigned we're using -1.
278 end = (lend + 1) >> PAGE_CACHE_SHIFT;
280 pagevec_init(&pvec, 0);
282 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
283 min(end - index, (pgoff_t)PAGEVEC_SIZE),
285 mem_cgroup_uncharge_start();
286 for (i = 0; i < pagevec_count(&pvec); i++) {
287 struct page *page = pvec.pages[i];
289 /* We rely upon deletion not changing page->index */
294 if (radix_tree_exceptional_entry(page)) {
295 clear_exceptional_entry(mapping, index, page);
299 if (!trylock_page(page))
301 WARN_ON(page->index != index);
302 if (PageWriteback(page)) {
306 truncate_inode_page(mapping, page);
309 pagevec_remove_exceptionals(&pvec);
310 pagevec_release(&pvec);
311 mem_cgroup_uncharge_end();
317 struct page *page = find_lock_page(mapping, start - 1);
319 unsigned int top = PAGE_CACHE_SIZE;
321 /* Truncation within a single page */
325 wait_on_page_writeback(page);
326 zero_user_segment(page, partial_start, top);
327 cleancache_invalidate_page(mapping, page);
328 if (page_has_private(page))
329 do_invalidatepage(page, partial_start,
330 top - partial_start);
332 page_cache_release(page);
336 struct page *page = find_lock_page(mapping, end);
338 wait_on_page_writeback(page);
339 zero_user_segment(page, 0, partial_end);
340 cleancache_invalidate_page(mapping, page);
341 if (page_has_private(page))
342 do_invalidatepage(page, 0,
345 page_cache_release(page);
349 * If the truncation happened within a single page no pages
350 * will be released, just zeroed, so we can bail out now.
358 if (!pagevec_lookup_entries(&pvec, mapping, index,
359 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
360 /* If all gone from start onwards, we're done */
363 /* Otherwise restart to make sure all gone */
367 if (index == start && indices[0] >= end) {
368 /* All gone out of hole to be punched, we're done */
369 pagevec_remove_exceptionals(&pvec);
370 pagevec_release(&pvec);
373 mem_cgroup_uncharge_start();
374 for (i = 0; i < pagevec_count(&pvec); i++) {
375 struct page *page = pvec.pages[i];
377 /* We rely upon deletion not changing page->index */
380 /* Restart punch to make sure all gone */
385 if (radix_tree_exceptional_entry(page)) {
386 clear_exceptional_entry(mapping, index, page);
391 WARN_ON(page->index != index);
392 wait_on_page_writeback(page);
393 truncate_inode_page(mapping, page);
396 pagevec_remove_exceptionals(&pvec);
397 pagevec_release(&pvec);
398 mem_cgroup_uncharge_end();
401 cleancache_invalidate_inode(mapping);
403 EXPORT_SYMBOL(truncate_inode_pages_range);
406 * truncate_inode_pages - truncate *all* the pages from an offset
407 * @mapping: mapping to truncate
408 * @lstart: offset from which to truncate
410 * Called under (and serialised by) inode->i_mutex.
412 * Note: When this function returns, there can be a page in the process of
413 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
414 * mapping->nrpages can be non-zero when this function returns even after
415 * truncation of the whole mapping.
417 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
419 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
421 EXPORT_SYMBOL(truncate_inode_pages);
424 * truncate_inode_pages_final - truncate *all* pages before inode dies
425 * @mapping: mapping to truncate
427 * Called under (and serialized by) inode->i_mutex.
429 * Filesystems have to use this in the .evict_inode path to inform the
430 * VM that this is the final truncate and the inode is going away.
432 void truncate_inode_pages_final(struct address_space *mapping)
434 unsigned long nrshadows;
435 unsigned long nrpages;
438 * Page reclaim can not participate in regular inode lifetime
439 * management (can't call iput()) and thus can race with the
440 * inode teardown. Tell it when the address space is exiting,
441 * so that it does not install eviction information after the
442 * final truncate has begun.
444 mapping_set_exiting(mapping);
447 * When reclaim installs eviction entries, it increases
448 * nrshadows first, then decreases nrpages. Make sure we see
449 * this in the right order or we might miss an entry.
451 nrpages = mapping->nrpages;
453 nrshadows = mapping->nrshadows;
455 if (nrpages || nrshadows) {
457 * As truncation uses a lockless tree lookup, cycle
458 * the tree lock to make sure any ongoing tree
459 * modification that does not see AS_EXITING is
460 * completed before starting the final truncate.
462 spin_lock_irq(&mapping->tree_lock);
463 spin_unlock_irq(&mapping->tree_lock);
467 * Cleancache needs notification even if there are no pages or shadow
470 truncate_inode_pages(mapping, 0);
472 EXPORT_SYMBOL(truncate_inode_pages_final);
475 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
476 * @mapping: the address_space which holds the pages to invalidate
477 * @start: the offset 'from' which to invalidate
478 * @end: the offset 'to' which to invalidate (inclusive)
480 * This function only removes the unlocked pages, if you want to
481 * remove all the pages of one inode, you must call truncate_inode_pages.
483 * invalidate_mapping_pages() will not block on IO activity. It will not
484 * invalidate pages which are dirty, locked, under writeback or mapped into
487 unsigned long invalidate_mapping_pages(struct address_space *mapping,
488 pgoff_t start, pgoff_t end)
490 pgoff_t indices[PAGEVEC_SIZE];
492 pgoff_t index = start;
494 unsigned long count = 0;
497 pagevec_init(&pvec, 0);
498 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
499 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
501 mem_cgroup_uncharge_start();
502 for (i = 0; i < pagevec_count(&pvec); i++) {
503 struct page *page = pvec.pages[i];
505 /* We rely upon deletion not changing page->index */
510 if (radix_tree_exceptional_entry(page)) {
511 clear_exceptional_entry(mapping, index, page);
515 if (!trylock_page(page))
517 WARN_ON(page->index != index);
518 ret = invalidate_inode_page(page);
521 * Invalidation is a hint that the page is no longer
522 * of interest and try to speed up its reclaim.
525 deactivate_page(page);
528 pagevec_remove_exceptionals(&pvec);
529 pagevec_release(&pvec);
530 mem_cgroup_uncharge_end();
536 EXPORT_SYMBOL(invalidate_mapping_pages);
539 * This is like invalidate_complete_page(), except it ignores the page's
540 * refcount. We do this because invalidate_inode_pages2() needs stronger
541 * invalidation guarantees, and cannot afford to leave pages behind because
542 * shrink_page_list() has a temp ref on them, or because they're transiently
543 * sitting in the lru_cache_add() pagevecs.
546 invalidate_complete_page2(struct address_space *mapping, struct page *page)
548 if (page->mapping != mapping)
551 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
554 spin_lock_irq(&mapping->tree_lock);
558 BUG_ON(page_has_private(page));
559 __delete_from_page_cache(page, NULL);
560 spin_unlock_irq(&mapping->tree_lock);
561 mem_cgroup_uncharge_cache_page(page);
563 if (mapping->a_ops->freepage)
564 mapping->a_ops->freepage(page);
566 page_cache_release(page); /* pagecache ref */
569 spin_unlock_irq(&mapping->tree_lock);
573 static int do_launder_page(struct address_space *mapping, struct page *page)
575 if (!PageDirty(page))
577 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
579 return mapping->a_ops->launder_page(page);
583 * invalidate_inode_pages2_range - remove range of pages from an address_space
584 * @mapping: the address_space
585 * @start: the page offset 'from' which to invalidate
586 * @end: the page offset 'to' which to invalidate (inclusive)
588 * Any pages which are found to be mapped into pagetables are unmapped prior to
591 * Returns -EBUSY if any pages could not be invalidated.
593 int invalidate_inode_pages2_range(struct address_space *mapping,
594 pgoff_t start, pgoff_t end)
596 pgoff_t indices[PAGEVEC_SIZE];
602 int did_range_unmap = 0;
604 cleancache_invalidate_inode(mapping);
605 pagevec_init(&pvec, 0);
607 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
608 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
610 mem_cgroup_uncharge_start();
611 for (i = 0; i < pagevec_count(&pvec); i++) {
612 struct page *page = pvec.pages[i];
614 /* We rely upon deletion not changing page->index */
619 if (radix_tree_exceptional_entry(page)) {
620 clear_exceptional_entry(mapping, index, page);
625 WARN_ON(page->index != index);
626 if (page->mapping != mapping) {
630 wait_on_page_writeback(page);
631 if (page_mapped(page)) {
632 if (!did_range_unmap) {
634 * Zap the rest of the file in one hit.
636 unmap_mapping_range(mapping,
637 (loff_t)index << PAGE_CACHE_SHIFT,
638 (loff_t)(1 + end - index)
646 unmap_mapping_range(mapping,
647 (loff_t)index << PAGE_CACHE_SHIFT,
651 BUG_ON(page_mapped(page));
652 ret2 = do_launder_page(mapping, page);
654 if (!invalidate_complete_page2(mapping, page))
661 pagevec_remove_exceptionals(&pvec);
662 pagevec_release(&pvec);
663 mem_cgroup_uncharge_end();
667 cleancache_invalidate_inode(mapping);
670 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
673 * invalidate_inode_pages2 - remove all pages from an address_space
674 * @mapping: the address_space
676 * Any pages which are found to be mapped into pagetables are unmapped prior to
679 * Returns -EBUSY if any pages could not be invalidated.
681 int invalidate_inode_pages2(struct address_space *mapping)
683 return invalidate_inode_pages2_range(mapping, 0, -1);
685 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
688 * truncate_pagecache - unmap and remove pagecache that has been truncated
690 * @newsize: new file size
692 * inode's new i_size must already be written before truncate_pagecache
695 * This function should typically be called before the filesystem
696 * releases resources associated with the freed range (eg. deallocates
697 * blocks). This way, pagecache will always stay logically coherent
698 * with on-disk format, and the filesystem would not have to deal with
699 * situations such as writepage being called for a page that has already
700 * had its underlying blocks deallocated.
702 void truncate_pagecache(struct inode *inode, loff_t newsize)
704 struct address_space *mapping = inode->i_mapping;
705 loff_t holebegin = round_up(newsize, PAGE_SIZE);
708 * unmap_mapping_range is called twice, first simply for
709 * efficiency so that truncate_inode_pages does fewer
710 * single-page unmaps. However after this first call, and
711 * before truncate_inode_pages finishes, it is possible for
712 * private pages to be COWed, which remain after
713 * truncate_inode_pages finishes, hence the second
714 * unmap_mapping_range call must be made for correctness.
716 unmap_mapping_range(mapping, holebegin, 0, 1);
717 truncate_inode_pages(mapping, newsize);
718 unmap_mapping_range(mapping, holebegin, 0, 1);
720 EXPORT_SYMBOL(truncate_pagecache);
723 * truncate_setsize - update inode and pagecache for a new file size
725 * @newsize: new file size
727 * truncate_setsize updates i_size and performs pagecache truncation (if
728 * necessary) to @newsize. It will be typically be called from the filesystem's
729 * setattr function when ATTR_SIZE is passed in.
731 * Must be called with inode_mutex held and before all filesystem specific
732 * block truncation has been performed.
734 void truncate_setsize(struct inode *inode, loff_t newsize)
736 loff_t oldsize = inode->i_size;
738 i_size_write(inode, newsize);
739 if (newsize > oldsize)
740 pagecache_isize_extended(inode, oldsize, newsize);
741 truncate_pagecache(inode, newsize);
743 EXPORT_SYMBOL(truncate_setsize);
746 * pagecache_isize_extended - update pagecache after extension of i_size
747 * @inode: inode for which i_size was extended
748 * @from: original inode size
749 * @to: new inode size
751 * Handle extension of inode size either caused by extending truncate or by
752 * write starting after current i_size. We mark the page straddling current
753 * i_size RO so that page_mkwrite() is called on the nearest write access to
754 * the page. This way filesystem can be sure that page_mkwrite() is called on
755 * the page before user writes to the page via mmap after the i_size has been
758 * The function must be called after i_size is updated so that page fault
759 * coming after we unlock the page will already see the new i_size.
760 * The function must be called while we still hold i_mutex - this not only
761 * makes sure i_size is stable but also that userspace cannot observe new
762 * i_size value before we are prepared to store mmap writes at new inode size.
764 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
766 int bsize = 1 << inode->i_blkbits;
771 WARN_ON(to > inode->i_size);
773 if (from >= to || bsize == PAGE_CACHE_SIZE)
775 /* Page straddling @from will not have any hole block created? */
776 rounded_from = round_up(from, bsize);
777 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
780 index = from >> PAGE_CACHE_SHIFT;
781 page = find_lock_page(inode->i_mapping, index);
782 /* Page not cached? Nothing to do */
786 * See clear_page_dirty_for_io() for details why set_page_dirty()
789 if (page_mkclean(page))
790 set_page_dirty(page);
792 page_cache_release(page);
794 EXPORT_SYMBOL(pagecache_isize_extended);
797 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
799 * @lstart: offset of beginning of hole
800 * @lend: offset of last byte of hole
802 * This function should typically be called before the filesystem
803 * releases resources associated with the freed range (eg. deallocates
804 * blocks). This way, pagecache will always stay logically coherent
805 * with on-disk format, and the filesystem would not have to deal with
806 * situations such as writepage being called for a page that has already
807 * had its underlying blocks deallocated.
809 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
811 struct address_space *mapping = inode->i_mapping;
812 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
813 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
815 * This rounding is currently just for example: unmap_mapping_range
816 * expands its hole outwards, whereas we want it to contract the hole
817 * inwards. However, existing callers of truncate_pagecache_range are
818 * doing their own page rounding first. Note that unmap_mapping_range
819 * allows holelen 0 for all, and we allow lend -1 for end of file.
823 * Unlike in truncate_pagecache, unmap_mapping_range is called only
824 * once (before truncating pagecache), and without "even_cows" flag:
825 * hole-punching should not remove private COWed pages from the hole.
827 if ((u64)unmap_end > (u64)unmap_start)
828 unmap_mapping_range(mapping, unmap_start,
829 1 + unmap_end - unmap_start, 0);
830 truncate_inode_pages_range(mapping, lstart, lend);
832 EXPORT_SYMBOL(truncate_pagecache_range);