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>
25 static void clear_exceptional_entry(struct address_space *mapping,
26 pgoff_t index, void *entry)
28 struct radix_tree_node *node;
31 /* Handled by shmem itself */
32 if (shmem_mapping(mapping))
35 spin_lock_irq(&mapping->tree_lock);
37 * Regular page slots are stabilized by the page lock even
38 * without the tree itself locked. These unlocked entries
39 * need verification under the tree lock.
41 if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot))
45 radix_tree_replace_slot(slot, NULL);
49 workingset_node_shadows_dec(node);
51 * Don't track node without shadow entries.
53 * Avoid acquiring the list_lru lock if already untracked.
54 * The list_empty() test is safe as node->private_list is
55 * protected by mapping->tree_lock.
57 if (!workingset_node_shadows(node) &&
58 !list_empty(&node->private_list))
59 list_lru_del(&workingset_shadow_nodes, &node->private_list);
60 __radix_tree_delete_node(&mapping->page_tree, node);
62 spin_unlock_irq(&mapping->tree_lock);
66 * do_invalidatepage - invalidate part or all of a page
67 * @page: the page which is affected
68 * @offset: start of the range to invalidate
69 * @length: length of the range to invalidate
71 * do_invalidatepage() is called when all or part of the page has become
72 * invalidated by a truncate operation.
74 * do_invalidatepage() does not have to release all buffers, but it must
75 * ensure that no dirty buffer is left outside @offset and that no I/O
76 * is underway against any of the blocks which are outside the truncation
77 * point. Because the caller is about to free (and possibly reuse) those
80 void do_invalidatepage(struct page *page, unsigned int offset,
83 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
85 invalidatepage = page->mapping->a_ops->invalidatepage;
88 invalidatepage = block_invalidatepage;
91 (*invalidatepage)(page, offset, length);
95 * This cancels just the dirty bit on the kernel page itself, it
96 * does NOT actually remove dirty bits on any mmap's that may be
97 * around. It also leaves the page tagged dirty, so any sync
98 * activity will still find it on the dirty lists, and in particular,
99 * clear_page_dirty_for_io() will still look at the dirty bits in
102 * Doing this should *normally* only ever be done when a page
103 * is truncated, and is not actually mapped anywhere at all. However,
104 * fs/buffer.c does this when it notices that somebody has cleaned
105 * out all the buffers on a page without actually doing it through
106 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
108 void cancel_dirty_page(struct page *page, unsigned int account_size)
110 if (TestClearPageDirty(page)) {
111 struct address_space *mapping = page->mapping;
112 if (mapping && mapping_cap_account_dirty(mapping)) {
113 dec_zone_page_state(page, NR_FILE_DIRTY);
114 dec_bdi_stat(mapping->backing_dev_info,
117 task_io_account_cancelled_write(account_size);
121 EXPORT_SYMBOL(cancel_dirty_page);
124 * If truncate cannot remove the fs-private metadata from the page, the page
125 * becomes orphaned. It will be left on the LRU and may even be mapped into
126 * user pagetables if we're racing with filemap_fault().
128 * We need to bale out if page->mapping is no longer equal to the original
129 * mapping. This happens a) when the VM reclaimed the page while we waited on
130 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
131 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
134 truncate_complete_page(struct address_space *mapping, struct page *page)
136 if (page->mapping != mapping)
139 if (page_has_private(page))
140 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
142 cancel_dirty_page(page, PAGE_CACHE_SIZE);
144 ClearPageMappedToDisk(page);
145 delete_from_page_cache(page);
150 * This is for invalidate_mapping_pages(). That function can be called at
151 * any time, and is not supposed to throw away dirty pages. But pages can
152 * be marked dirty at any time too, so use remove_mapping which safely
153 * discards clean, unused pages.
155 * Returns non-zero if the page was successfully invalidated.
158 invalidate_complete_page(struct address_space *mapping, struct page *page)
162 if (page->mapping != mapping)
165 if (page_has_private(page) && !try_to_release_page(page, 0))
168 ret = remove_mapping(mapping, page);
173 int truncate_inode_page(struct address_space *mapping, struct page *page)
175 if (page_mapped(page)) {
176 unmap_mapping_range(mapping,
177 (loff_t)page->index << PAGE_CACHE_SHIFT,
180 return truncate_complete_page(mapping, page);
184 * Used to get rid of pages on hardware memory corruption.
186 int generic_error_remove_page(struct address_space *mapping, struct page *page)
191 * Only punch for normal data pages for now.
192 * Handling other types like directories would need more auditing.
194 if (!S_ISREG(mapping->host->i_mode))
196 return truncate_inode_page(mapping, page);
198 EXPORT_SYMBOL(generic_error_remove_page);
201 * Safely invalidate one page from its pagecache mapping.
202 * It only drops clean, unused pages. The page must be locked.
204 * Returns 1 if the page is successfully invalidated, otherwise 0.
206 int invalidate_inode_page(struct page *page)
208 struct address_space *mapping = page_mapping(page);
211 if (PageDirty(page) || PageWriteback(page))
213 if (page_mapped(page))
215 return invalidate_complete_page(mapping, page);
219 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
220 * @mapping: mapping to truncate
221 * @lstart: offset from which to truncate
222 * @lend: offset to which to truncate (inclusive)
224 * Truncate the page cache, removing the pages that are between
225 * specified offsets (and zeroing out partial pages
226 * if lstart or lend + 1 is not page aligned).
228 * Truncate takes two passes - the first pass is nonblocking. It will not
229 * block on page locks and it will not block on writeback. The second pass
230 * will wait. This is to prevent as much IO as possible in the affected region.
231 * The first pass will remove most pages, so the search cost of the second pass
234 * We pass down the cache-hot hint to the page freeing code. Even if the
235 * mapping is large, it is probably the case that the final pages are the most
236 * recently touched, and freeing happens in ascending file offset order.
238 * Note that since ->invalidatepage() accepts range to invalidate
239 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
240 * page aligned properly.
242 void truncate_inode_pages_range(struct address_space *mapping,
243 loff_t lstart, loff_t lend)
245 pgoff_t start; /* inclusive */
246 pgoff_t end; /* exclusive */
247 unsigned int partial_start; /* inclusive */
248 unsigned int partial_end; /* exclusive */
250 pgoff_t indices[PAGEVEC_SIZE];
254 cleancache_invalidate_inode(mapping);
255 if (mapping->nrpages == 0 && mapping->nrshadows == 0)
258 /* Offsets within partial pages */
259 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
260 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
263 * 'start' and 'end' always covers the range of pages to be fully
264 * truncated. Partial pages are covered with 'partial_start' at the
265 * start of the range and 'partial_end' at the end of the range.
266 * Note that 'end' is exclusive while 'lend' is inclusive.
268 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
271 * lend == -1 indicates end-of-file so we have to set 'end'
272 * to the highest possible pgoff_t and since the type is
273 * unsigned we're using -1.
277 end = (lend + 1) >> PAGE_CACHE_SHIFT;
279 pagevec_init(&pvec, 0);
281 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
282 min(end - index, (pgoff_t)PAGEVEC_SIZE),
284 mem_cgroup_uncharge_start();
285 for (i = 0; i < pagevec_count(&pvec); i++) {
286 struct page *page = pvec.pages[i];
288 /* We rely upon deletion not changing page->index */
293 if (radix_tree_exceptional_entry(page)) {
294 clear_exceptional_entry(mapping, index, page);
298 if (!trylock_page(page))
300 WARN_ON(page->index != index);
301 if (PageWriteback(page)) {
305 truncate_inode_page(mapping, page);
308 pagevec_remove_exceptionals(&pvec);
309 pagevec_release(&pvec);
310 mem_cgroup_uncharge_end();
316 struct page *page = find_lock_page(mapping, start - 1);
318 unsigned int top = PAGE_CACHE_SIZE;
320 /* Truncation within a single page */
324 wait_on_page_writeback(page);
325 zero_user_segment(page, partial_start, top);
326 cleancache_invalidate_page(mapping, page);
327 if (page_has_private(page))
328 do_invalidatepage(page, partial_start,
329 top - partial_start);
331 page_cache_release(page);
335 struct page *page = find_lock_page(mapping, end);
337 wait_on_page_writeback(page);
338 zero_user_segment(page, 0, partial_end);
339 cleancache_invalidate_page(mapping, page);
340 if (page_has_private(page))
341 do_invalidatepage(page, 0,
344 page_cache_release(page);
348 * If the truncation happened within a single page no pages
349 * will be released, just zeroed, so we can bail out now.
357 if (!pagevec_lookup_entries(&pvec, mapping, index,
358 min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
359 /* If all gone from start onwards, we're done */
362 /* Otherwise restart to make sure all gone */
366 if (index == start && indices[0] >= end) {
367 /* All gone out of hole to be punched, we're done */
368 pagevec_remove_exceptionals(&pvec);
369 pagevec_release(&pvec);
372 mem_cgroup_uncharge_start();
373 for (i = 0; i < pagevec_count(&pvec); i++) {
374 struct page *page = pvec.pages[i];
376 /* We rely upon deletion not changing page->index */
379 /* Restart punch to make sure all gone */
384 if (radix_tree_exceptional_entry(page)) {
385 clear_exceptional_entry(mapping, index, page);
390 WARN_ON(page->index != index);
391 wait_on_page_writeback(page);
392 truncate_inode_page(mapping, page);
395 pagevec_remove_exceptionals(&pvec);
396 pagevec_release(&pvec);
397 mem_cgroup_uncharge_end();
400 cleancache_invalidate_inode(mapping);
402 EXPORT_SYMBOL(truncate_inode_pages_range);
405 * truncate_inode_pages - truncate *all* the pages from an offset
406 * @mapping: mapping to truncate
407 * @lstart: offset from which to truncate
409 * Called under (and serialised by) inode->i_mutex.
411 * Note: When this function returns, there can be a page in the process of
412 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
413 * mapping->nrpages can be non-zero when this function returns even after
414 * truncation of the whole mapping.
416 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
418 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
420 EXPORT_SYMBOL(truncate_inode_pages);
423 * truncate_inode_pages_final - truncate *all* pages before inode dies
424 * @mapping: mapping to truncate
426 * Called under (and serialized by) inode->i_mutex.
428 * Filesystems have to use this in the .evict_inode path to inform the
429 * VM that this is the final truncate and the inode is going away.
431 void truncate_inode_pages_final(struct address_space *mapping)
433 unsigned long nrshadows;
434 unsigned long nrpages;
437 * Page reclaim can not participate in regular inode lifetime
438 * management (can't call iput()) and thus can race with the
439 * inode teardown. Tell it when the address space is exiting,
440 * so that it does not install eviction information after the
441 * final truncate has begun.
443 mapping_set_exiting(mapping);
446 * When reclaim installs eviction entries, it increases
447 * nrshadows first, then decreases nrpages. Make sure we see
448 * this in the right order or we might miss an entry.
450 nrpages = mapping->nrpages;
452 nrshadows = mapping->nrshadows;
454 if (nrpages || nrshadows) {
456 * As truncation uses a lockless tree lookup, cycle
457 * the tree lock to make sure any ongoing tree
458 * modification that does not see AS_EXITING is
459 * completed before starting the final truncate.
461 spin_lock_irq(&mapping->tree_lock);
462 spin_unlock_irq(&mapping->tree_lock);
464 truncate_inode_pages(mapping, 0);
467 EXPORT_SYMBOL(truncate_inode_pages_final);
470 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
471 * @mapping: the address_space which holds the pages to invalidate
472 * @start: the offset 'from' which to invalidate
473 * @end: the offset 'to' which to invalidate (inclusive)
475 * This function only removes the unlocked pages, if you want to
476 * remove all the pages of one inode, you must call truncate_inode_pages.
478 * invalidate_mapping_pages() will not block on IO activity. It will not
479 * invalidate pages which are dirty, locked, under writeback or mapped into
482 unsigned long invalidate_mapping_pages(struct address_space *mapping,
483 pgoff_t start, pgoff_t end)
485 pgoff_t indices[PAGEVEC_SIZE];
487 pgoff_t index = start;
489 unsigned long count = 0;
492 pagevec_init(&pvec, 0);
493 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
494 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
496 mem_cgroup_uncharge_start();
497 for (i = 0; i < pagevec_count(&pvec); i++) {
498 struct page *page = pvec.pages[i];
500 /* We rely upon deletion not changing page->index */
505 if (radix_tree_exceptional_entry(page)) {
506 clear_exceptional_entry(mapping, index, page);
510 if (!trylock_page(page))
512 WARN_ON(page->index != index);
513 ret = invalidate_inode_page(page);
516 * Invalidation is a hint that the page is no longer
517 * of interest and try to speed up its reclaim.
520 deactivate_page(page);
523 pagevec_remove_exceptionals(&pvec);
524 pagevec_release(&pvec);
525 mem_cgroup_uncharge_end();
531 EXPORT_SYMBOL(invalidate_mapping_pages);
534 * This is like invalidate_complete_page(), except it ignores the page's
535 * refcount. We do this because invalidate_inode_pages2() needs stronger
536 * invalidation guarantees, and cannot afford to leave pages behind because
537 * shrink_page_list() has a temp ref on them, or because they're transiently
538 * sitting in the lru_cache_add() pagevecs.
541 invalidate_complete_page2(struct address_space *mapping, struct page *page)
543 if (page->mapping != mapping)
546 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
549 spin_lock_irq(&mapping->tree_lock);
553 BUG_ON(page_has_private(page));
554 __delete_from_page_cache(page, NULL);
555 spin_unlock_irq(&mapping->tree_lock);
556 mem_cgroup_uncharge_cache_page(page);
558 if (mapping->a_ops->freepage)
559 mapping->a_ops->freepage(page);
561 page_cache_release(page); /* pagecache ref */
564 spin_unlock_irq(&mapping->tree_lock);
568 static int do_launder_page(struct address_space *mapping, struct page *page)
570 if (!PageDirty(page))
572 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
574 return mapping->a_ops->launder_page(page);
578 * invalidate_inode_pages2_range - remove range of pages from an address_space
579 * @mapping: the address_space
580 * @start: the page offset 'from' which to invalidate
581 * @end: the page offset 'to' which to invalidate (inclusive)
583 * Any pages which are found to be mapped into pagetables are unmapped prior to
586 * Returns -EBUSY if any pages could not be invalidated.
588 int invalidate_inode_pages2_range(struct address_space *mapping,
589 pgoff_t start, pgoff_t end)
591 pgoff_t indices[PAGEVEC_SIZE];
597 int did_range_unmap = 0;
599 cleancache_invalidate_inode(mapping);
600 pagevec_init(&pvec, 0);
602 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
603 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
605 mem_cgroup_uncharge_start();
606 for (i = 0; i < pagevec_count(&pvec); i++) {
607 struct page *page = pvec.pages[i];
609 /* We rely upon deletion not changing page->index */
614 if (radix_tree_exceptional_entry(page)) {
615 clear_exceptional_entry(mapping, index, page);
620 WARN_ON(page->index != index);
621 if (page->mapping != mapping) {
625 wait_on_page_writeback(page);
626 if (page_mapped(page)) {
627 if (!did_range_unmap) {
629 * Zap the rest of the file in one hit.
631 unmap_mapping_range(mapping,
632 (loff_t)index << PAGE_CACHE_SHIFT,
633 (loff_t)(1 + end - index)
641 unmap_mapping_range(mapping,
642 (loff_t)index << PAGE_CACHE_SHIFT,
646 BUG_ON(page_mapped(page));
647 ret2 = do_launder_page(mapping, page);
649 if (!invalidate_complete_page2(mapping, page))
656 pagevec_remove_exceptionals(&pvec);
657 pagevec_release(&pvec);
658 mem_cgroup_uncharge_end();
662 cleancache_invalidate_inode(mapping);
665 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
668 * invalidate_inode_pages2 - remove all pages from an address_space
669 * @mapping: the address_space
671 * Any pages which are found to be mapped into pagetables are unmapped prior to
674 * Returns -EBUSY if any pages could not be invalidated.
676 int invalidate_inode_pages2(struct address_space *mapping)
678 return invalidate_inode_pages2_range(mapping, 0, -1);
680 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
683 * truncate_pagecache - unmap and remove pagecache that has been truncated
685 * @newsize: new file size
687 * inode's new i_size must already be written before truncate_pagecache
690 * This function should typically be called before the filesystem
691 * releases resources associated with the freed range (eg. deallocates
692 * blocks). This way, pagecache will always stay logically coherent
693 * with on-disk format, and the filesystem would not have to deal with
694 * situations such as writepage being called for a page that has already
695 * had its underlying blocks deallocated.
697 void truncate_pagecache(struct inode *inode, loff_t newsize)
699 struct address_space *mapping = inode->i_mapping;
700 loff_t holebegin = round_up(newsize, PAGE_SIZE);
703 * unmap_mapping_range is called twice, first simply for
704 * efficiency so that truncate_inode_pages does fewer
705 * single-page unmaps. However after this first call, and
706 * before truncate_inode_pages finishes, it is possible for
707 * private pages to be COWed, which remain after
708 * truncate_inode_pages finishes, hence the second
709 * unmap_mapping_range call must be made for correctness.
711 unmap_mapping_range(mapping, holebegin, 0, 1);
712 truncate_inode_pages(mapping, newsize);
713 unmap_mapping_range(mapping, holebegin, 0, 1);
715 EXPORT_SYMBOL(truncate_pagecache);
718 * truncate_setsize - update inode and pagecache for a new file size
720 * @newsize: new file size
722 * truncate_setsize updates i_size and performs pagecache truncation (if
723 * necessary) to @newsize. It will be typically be called from the filesystem's
724 * setattr function when ATTR_SIZE is passed in.
726 * Must be called with inode_mutex held and before all filesystem specific
727 * block truncation has been performed.
729 void truncate_setsize(struct inode *inode, loff_t newsize)
731 i_size_write(inode, newsize);
732 truncate_pagecache(inode, newsize);
734 EXPORT_SYMBOL(truncate_setsize);
737 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
739 * @lstart: offset of beginning of hole
740 * @lend: offset of last byte of hole
742 * This function should typically be called before the filesystem
743 * releases resources associated with the freed range (eg. deallocates
744 * blocks). This way, pagecache will always stay logically coherent
745 * with on-disk format, and the filesystem would not have to deal with
746 * situations such as writepage being called for a page that has already
747 * had its underlying blocks deallocated.
749 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
751 struct address_space *mapping = inode->i_mapping;
752 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
753 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
755 * This rounding is currently just for example: unmap_mapping_range
756 * expands its hole outwards, whereas we want it to contract the hole
757 * inwards. However, existing callers of truncate_pagecache_range are
758 * doing their own page rounding first. Note that unmap_mapping_range
759 * allows holelen 0 for all, and we allow lend -1 for end of file.
763 * Unlike in truncate_pagecache, unmap_mapping_range is called only
764 * once (before truncating pagecache), and without "even_cows" flag:
765 * hole-punching should not remove private COWed pages from the hole.
767 if ((u64)unmap_end > (u64)unmap_start)
768 unmap_mapping_range(mapping, unmap_start,
769 1 + unmap_end - unmap_start, 0);
770 truncate_inode_pages_range(mapping, lstart, lend);
772 EXPORT_SYMBOL(truncate_pagecache_range);