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 /* Handled by shmem itself */
30 if (shmem_mapping(mapping))
33 spin_lock_irq(&mapping->tree_lock);
35 * Regular page slots are stabilized by the page lock even
36 * without the tree itself locked. These unlocked entries
37 * need verification under the tree lock.
39 radix_tree_delete_item(&mapping->page_tree, index, entry);
40 spin_unlock_irq(&mapping->tree_lock);
44 * do_invalidatepage - invalidate part or all of a page
45 * @page: the page which is affected
46 * @offset: start of the range to invalidate
47 * @length: length of the range to invalidate
49 * do_invalidatepage() is called when all or part of the page has become
50 * invalidated by a truncate operation.
52 * do_invalidatepage() does not have to release all buffers, but it must
53 * ensure that no dirty buffer is left outside @offset and that no I/O
54 * is underway against any of the blocks which are outside the truncation
55 * point. Because the caller is about to free (and possibly reuse) those
58 void do_invalidatepage(struct page *page, unsigned int offset,
61 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
63 invalidatepage = page->mapping->a_ops->invalidatepage;
66 invalidatepage = block_invalidatepage;
69 (*invalidatepage)(page, offset, length);
73 * This cancels just the dirty bit on the kernel page itself, it
74 * does NOT actually remove dirty bits on any mmap's that may be
75 * around. It also leaves the page tagged dirty, so any sync
76 * activity will still find it on the dirty lists, and in particular,
77 * clear_page_dirty_for_io() will still look at the dirty bits in
80 * Doing this should *normally* only ever be done when a page
81 * is truncated, and is not actually mapped anywhere at all. However,
82 * fs/buffer.c does this when it notices that somebody has cleaned
83 * out all the buffers on a page without actually doing it through
84 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
86 void cancel_dirty_page(struct page *page, unsigned int account_size)
88 if (TestClearPageDirty(page)) {
89 struct address_space *mapping = page->mapping;
90 if (mapping && mapping_cap_account_dirty(mapping)) {
91 dec_zone_page_state(page, NR_FILE_DIRTY);
92 dec_bdi_stat(mapping->backing_dev_info,
95 task_io_account_cancelled_write(account_size);
99 EXPORT_SYMBOL(cancel_dirty_page);
102 * If truncate cannot remove the fs-private metadata from the page, the page
103 * becomes orphaned. It will be left on the LRU and may even be mapped into
104 * user pagetables if we're racing with filemap_fault().
106 * We need to bale out if page->mapping is no longer equal to the original
107 * mapping. This happens a) when the VM reclaimed the page while we waited on
108 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
109 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
112 truncate_complete_page(struct address_space *mapping, struct page *page)
114 if (page->mapping != mapping)
117 if (page_has_private(page))
118 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
120 cancel_dirty_page(page, PAGE_CACHE_SIZE);
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)
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 mem_cgroup_uncharge_start();
263 for (i = 0; i < pagevec_count(&pvec); i++) {
264 struct page *page = pvec.pages[i];
266 /* We rely upon deletion not changing page->index */
271 if (radix_tree_exceptional_entry(page)) {
272 clear_exceptional_entry(mapping, index, page);
276 if (!trylock_page(page))
278 WARN_ON(page->index != index);
279 if (PageWriteback(page)) {
283 truncate_inode_page(mapping, page);
286 pagevec_remove_exceptionals(&pvec);
287 pagevec_release(&pvec);
288 mem_cgroup_uncharge_end();
294 struct page *page = find_lock_page(mapping, start - 1);
296 unsigned int top = PAGE_CACHE_SIZE;
298 /* Truncation within a single page */
302 wait_on_page_writeback(page);
303 zero_user_segment(page, partial_start, top);
304 cleancache_invalidate_page(mapping, page);
305 if (page_has_private(page))
306 do_invalidatepage(page, partial_start,
307 top - partial_start);
309 page_cache_release(page);
313 struct page *page = find_lock_page(mapping, end);
315 wait_on_page_writeback(page);
316 zero_user_segment(page, 0, partial_end);
317 cleancache_invalidate_page(mapping, page);
318 if (page_has_private(page))
319 do_invalidatepage(page, 0,
322 page_cache_release(page);
326 * If the truncation happened within a single page no pages
327 * will be released, just zeroed, so we can bail out now.
335 if (!pagevec_lookup_entries(&pvec, mapping, index,
336 min(end - index, (pgoff_t)PAGEVEC_SIZE),
343 if (index == start && indices[0] >= end) {
344 pagevec_remove_exceptionals(&pvec);
345 pagevec_release(&pvec);
348 mem_cgroup_uncharge_start();
349 for (i = 0; i < pagevec_count(&pvec); i++) {
350 struct page *page = pvec.pages[i];
352 /* We rely upon deletion not changing page->index */
357 if (radix_tree_exceptional_entry(page)) {
358 clear_exceptional_entry(mapping, index, page);
363 WARN_ON(page->index != index);
364 wait_on_page_writeback(page);
365 truncate_inode_page(mapping, page);
368 pagevec_remove_exceptionals(&pvec);
369 pagevec_release(&pvec);
370 mem_cgroup_uncharge_end();
373 cleancache_invalidate_inode(mapping);
375 EXPORT_SYMBOL(truncate_inode_pages_range);
378 * truncate_inode_pages - truncate *all* the pages from an offset
379 * @mapping: mapping to truncate
380 * @lstart: offset from which to truncate
382 * Called under (and serialised by) inode->i_mutex.
384 * Note: When this function returns, there can be a page in the process of
385 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
386 * mapping->nrpages can be non-zero when this function returns even after
387 * truncation of the whole mapping.
389 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
391 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
393 EXPORT_SYMBOL(truncate_inode_pages);
396 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
397 * @mapping: the address_space which holds the pages to invalidate
398 * @start: the offset 'from' which to invalidate
399 * @end: the offset 'to' which to invalidate (inclusive)
401 * This function only removes the unlocked pages, if you want to
402 * remove all the pages of one inode, you must call truncate_inode_pages.
404 * invalidate_mapping_pages() will not block on IO activity. It will not
405 * invalidate pages which are dirty, locked, under writeback or mapped into
408 unsigned long invalidate_mapping_pages(struct address_space *mapping,
409 pgoff_t start, pgoff_t end)
411 pgoff_t indices[PAGEVEC_SIZE];
413 pgoff_t index = start;
415 unsigned long count = 0;
419 * Note: this function may get called on a shmem/tmpfs mapping:
420 * pagevec_lookup() might then return 0 prematurely (because it
421 * got a gangful of swap entries); but it's hardly worth worrying
422 * about - it can rarely have anything to free from such a mapping
423 * (most pages are dirty), and already skips over any difficulties.
426 pagevec_init(&pvec, 0);
427 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
428 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
430 mem_cgroup_uncharge_start();
431 for (i = 0; i < pagevec_count(&pvec); i++) {
432 struct page *page = pvec.pages[i];
434 /* We rely upon deletion not changing page->index */
439 if (radix_tree_exceptional_entry(page)) {
440 clear_exceptional_entry(mapping, index, page);
444 if (!trylock_page(page))
446 WARN_ON(page->index != index);
447 ret = invalidate_inode_page(page);
450 * Invalidation is a hint that the page is no longer
451 * of interest and try to speed up its reclaim.
454 deactivate_page(page);
457 pagevec_remove_exceptionals(&pvec);
458 pagevec_release(&pvec);
459 mem_cgroup_uncharge_end();
465 EXPORT_SYMBOL(invalidate_mapping_pages);
468 * This is like invalidate_complete_page(), except it ignores the page's
469 * refcount. We do this because invalidate_inode_pages2() needs stronger
470 * invalidation guarantees, and cannot afford to leave pages behind because
471 * shrink_page_list() has a temp ref on them, or because they're transiently
472 * sitting in the lru_cache_add() pagevecs.
475 invalidate_complete_page2(struct address_space *mapping, struct page *page)
477 if (page->mapping != mapping)
480 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
483 spin_lock_irq(&mapping->tree_lock);
487 BUG_ON(page_has_private(page));
488 __delete_from_page_cache(page);
489 spin_unlock_irq(&mapping->tree_lock);
490 mem_cgroup_uncharge_cache_page(page);
492 if (mapping->a_ops->freepage)
493 mapping->a_ops->freepage(page);
495 page_cache_release(page); /* pagecache ref */
498 spin_unlock_irq(&mapping->tree_lock);
502 static int do_launder_page(struct address_space *mapping, struct page *page)
504 if (!PageDirty(page))
506 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
508 return mapping->a_ops->launder_page(page);
512 * invalidate_inode_pages2_range - remove range of pages from an address_space
513 * @mapping: the address_space
514 * @start: the page offset 'from' which to invalidate
515 * @end: the page offset 'to' which to invalidate (inclusive)
517 * Any pages which are found to be mapped into pagetables are unmapped prior to
520 * Returns -EBUSY if any pages could not be invalidated.
522 int invalidate_inode_pages2_range(struct address_space *mapping,
523 pgoff_t start, pgoff_t end)
525 pgoff_t indices[PAGEVEC_SIZE];
531 int did_range_unmap = 0;
533 cleancache_invalidate_inode(mapping);
534 pagevec_init(&pvec, 0);
536 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
537 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
539 mem_cgroup_uncharge_start();
540 for (i = 0; i < pagevec_count(&pvec); i++) {
541 struct page *page = pvec.pages[i];
543 /* We rely upon deletion not changing page->index */
548 if (radix_tree_exceptional_entry(page)) {
549 clear_exceptional_entry(mapping, index, page);
554 WARN_ON(page->index != index);
555 if (page->mapping != mapping) {
559 wait_on_page_writeback(page);
560 if (page_mapped(page)) {
561 if (!did_range_unmap) {
563 * Zap the rest of the file in one hit.
565 unmap_mapping_range(mapping,
566 (loff_t)index << PAGE_CACHE_SHIFT,
567 (loff_t)(1 + end - index)
575 unmap_mapping_range(mapping,
576 (loff_t)index << PAGE_CACHE_SHIFT,
580 BUG_ON(page_mapped(page));
581 ret2 = do_launder_page(mapping, page);
583 if (!invalidate_complete_page2(mapping, page))
590 pagevec_remove_exceptionals(&pvec);
591 pagevec_release(&pvec);
592 mem_cgroup_uncharge_end();
596 cleancache_invalidate_inode(mapping);
599 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
602 * invalidate_inode_pages2 - remove all pages from an address_space
603 * @mapping: the address_space
605 * Any pages which are found to be mapped into pagetables are unmapped prior to
608 * Returns -EBUSY if any pages could not be invalidated.
610 int invalidate_inode_pages2(struct address_space *mapping)
612 return invalidate_inode_pages2_range(mapping, 0, -1);
614 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
617 * truncate_pagecache - unmap and remove pagecache that has been truncated
619 * @newsize: new file size
621 * inode's new i_size must already be written before truncate_pagecache
624 * This function should typically be called before the filesystem
625 * releases resources associated with the freed range (eg. deallocates
626 * blocks). This way, pagecache will always stay logically coherent
627 * with on-disk format, and the filesystem would not have to deal with
628 * situations such as writepage being called for a page that has already
629 * had its underlying blocks deallocated.
631 void truncate_pagecache(struct inode *inode, loff_t newsize)
633 struct address_space *mapping = inode->i_mapping;
634 loff_t holebegin = round_up(newsize, PAGE_SIZE);
637 * unmap_mapping_range is called twice, first simply for
638 * efficiency so that truncate_inode_pages does fewer
639 * single-page unmaps. However after this first call, and
640 * before truncate_inode_pages finishes, it is possible for
641 * private pages to be COWed, which remain after
642 * truncate_inode_pages finishes, hence the second
643 * unmap_mapping_range call must be made for correctness.
645 unmap_mapping_range(mapping, holebegin, 0, 1);
646 truncate_inode_pages(mapping, newsize);
647 unmap_mapping_range(mapping, holebegin, 0, 1);
649 EXPORT_SYMBOL(truncate_pagecache);
652 * truncate_setsize - update inode and pagecache for a new file size
654 * @newsize: new file size
656 * truncate_setsize updates i_size and performs pagecache truncation (if
657 * necessary) to @newsize. It will be typically be called from the filesystem's
658 * setattr function when ATTR_SIZE is passed in.
660 * Must be called with inode_mutex held and before all filesystem specific
661 * block truncation has been performed.
663 void truncate_setsize(struct inode *inode, loff_t newsize)
665 loff_t oldsize = inode->i_size;
667 i_size_write(inode, newsize);
668 if (newsize > oldsize)
669 pagecache_isize_extended(inode, oldsize, newsize);
670 truncate_pagecache(inode, newsize);
672 EXPORT_SYMBOL(truncate_setsize);
675 * pagecache_isize_extended - update pagecache after extension of i_size
676 * @inode: inode for which i_size was extended
677 * @from: original inode size
678 * @to: new inode size
680 * Handle extension of inode size either caused by extending truncate or by
681 * write starting after current i_size. We mark the page straddling current
682 * i_size RO so that page_mkwrite() is called on the nearest write access to
683 * the page. This way filesystem can be sure that page_mkwrite() is called on
684 * the page before user writes to the page via mmap after the i_size has been
687 * The function must be called after i_size is updated so that page fault
688 * coming after we unlock the page will already see the new i_size.
689 * The function must be called while we still hold i_mutex - this not only
690 * makes sure i_size is stable but also that userspace cannot observe new
691 * i_size value before we are prepared to store mmap writes at new inode size.
693 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
695 int bsize = 1 << inode->i_blkbits;
700 WARN_ON(to > inode->i_size);
702 if (from >= to || bsize == PAGE_CACHE_SIZE)
704 /* Page straddling @from will not have any hole block created? */
705 rounded_from = round_up(from, bsize);
706 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
709 index = from >> PAGE_CACHE_SHIFT;
710 page = find_lock_page(inode->i_mapping, index);
711 /* Page not cached? Nothing to do */
715 * See clear_page_dirty_for_io() for details why set_page_dirty()
718 if (page_mkclean(page))
719 set_page_dirty(page);
721 page_cache_release(page);
723 EXPORT_SYMBOL(pagecache_isize_extended);
726 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
728 * @lstart: offset of beginning of hole
729 * @lend: offset of last byte of hole
731 * This function should typically be called before the filesystem
732 * releases resources associated with the freed range (eg. deallocates
733 * blocks). This way, pagecache will always stay logically coherent
734 * with on-disk format, and the filesystem would not have to deal with
735 * situations such as writepage being called for a page that has already
736 * had its underlying blocks deallocated.
738 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
740 struct address_space *mapping = inode->i_mapping;
741 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
742 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
744 * This rounding is currently just for example: unmap_mapping_range
745 * expands its hole outwards, whereas we want it to contract the hole
746 * inwards. However, existing callers of truncate_pagecache_range are
747 * doing their own page rounding first. Note that unmap_mapping_range
748 * allows holelen 0 for all, and we allow lend -1 for end of file.
752 * Unlike in truncate_pagecache, unmap_mapping_range is called only
753 * once (before truncating pagecache), and without "even_cows" flag:
754 * hole-punching should not remove private COWed pages from the hole.
756 if ((u64)unmap_end > (u64)unmap_start)
757 unmap_mapping_range(mapping, unmap_start,
758 1 + unmap_end - unmap_start, 0);
759 truncate_inode_pages_range(mapping, lstart, lend);
761 EXPORT_SYMBOL(truncate_pagecache_range);