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 /* Handled by shmem itself */
29 if (shmem_mapping(mapping))
32 spin_lock_irq(&mapping->tree_lock);
34 * Regular page slots are stabilized by the page lock even
35 * without the tree itself locked. These unlocked entries
36 * need verification under the tree lock.
38 radix_tree_delete_item(&mapping->page_tree, index, entry);
39 spin_unlock_irq(&mapping->tree_lock);
43 * do_invalidatepage - invalidate part or all of a page
44 * @page: the page which is affected
45 * @offset: start of the range to invalidate
46 * @length: length of the range to invalidate
48 * do_invalidatepage() is called when all or part of the page has become
49 * invalidated by a truncate operation.
51 * do_invalidatepage() does not have to release all buffers, but it must
52 * ensure that no dirty buffer is left outside @offset and that no I/O
53 * is underway against any of the blocks which are outside the truncation
54 * point. Because the caller is about to free (and possibly reuse) those
57 void do_invalidatepage(struct page *page, unsigned int offset,
60 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
62 invalidatepage = page->mapping->a_ops->invalidatepage;
65 invalidatepage = block_invalidatepage;
68 (*invalidatepage)(page, offset, length);
72 * This cancels just the dirty bit on the kernel page itself, it
73 * does NOT actually remove dirty bits on any mmap's that may be
74 * around. It also leaves the page tagged dirty, so any sync
75 * activity will still find it on the dirty lists, and in particular,
76 * clear_page_dirty_for_io() will still look at the dirty bits in
79 * Doing this should *normally* only ever be done when a page
80 * is truncated, and is not actually mapped anywhere at all. However,
81 * fs/buffer.c does this when it notices that somebody has cleaned
82 * out all the buffers on a page without actually doing it through
83 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
85 void cancel_dirty_page(struct page *page, unsigned int account_size)
87 if (TestClearPageDirty(page)) {
88 struct address_space *mapping = page->mapping;
89 if (mapping && mapping_cap_account_dirty(mapping)) {
90 dec_zone_page_state(page, NR_FILE_DIRTY);
91 dec_bdi_stat(mapping->backing_dev_info,
94 task_io_account_cancelled_write(account_size);
98 EXPORT_SYMBOL(cancel_dirty_page);
101 * If truncate cannot remove the fs-private metadata from the page, the page
102 * becomes orphaned. It will be left on the LRU and may even be mapped into
103 * user pagetables if we're racing with filemap_fault().
105 * We need to bale out if page->mapping is no longer equal to the original
106 * mapping. This happens a) when the VM reclaimed the page while we waited on
107 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
108 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
111 truncate_complete_page(struct address_space *mapping, struct page *page)
113 if (page->mapping != mapping)
116 if (page_has_private(page))
117 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
119 cancel_dirty_page(page, PAGE_CACHE_SIZE);
121 ClearPageMappedToDisk(page);
122 delete_from_page_cache(page);
127 * This is for invalidate_mapping_pages(). That function can be called at
128 * any time, and is not supposed to throw away dirty pages. But pages can
129 * be marked dirty at any time too, so use remove_mapping which safely
130 * discards clean, unused pages.
132 * Returns non-zero if the page was successfully invalidated.
135 invalidate_complete_page(struct address_space *mapping, struct page *page)
139 if (page->mapping != mapping)
142 if (page_has_private(page) && !try_to_release_page(page, 0))
145 ret = remove_mapping(mapping, page);
150 int truncate_inode_page(struct address_space *mapping, struct page *page)
152 if (page_mapped(page)) {
153 unmap_mapping_range(mapping,
154 (loff_t)page->index << PAGE_CACHE_SHIFT,
157 return truncate_complete_page(mapping, page);
161 * Used to get rid of pages on hardware memory corruption.
163 int generic_error_remove_page(struct address_space *mapping, struct page *page)
168 * Only punch for normal data pages for now.
169 * Handling other types like directories would need more auditing.
171 if (!S_ISREG(mapping->host->i_mode))
173 return truncate_inode_page(mapping, page);
175 EXPORT_SYMBOL(generic_error_remove_page);
178 * Safely invalidate one page from its pagecache mapping.
179 * It only drops clean, unused pages. The page must be locked.
181 * Returns 1 if the page is successfully invalidated, otherwise 0.
183 int invalidate_inode_page(struct page *page)
185 struct address_space *mapping = page_mapping(page);
188 if (PageDirty(page) || PageWriteback(page))
190 if (page_mapped(page))
192 return invalidate_complete_page(mapping, page);
196 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
197 * @mapping: mapping to truncate
198 * @lstart: offset from which to truncate
199 * @lend: offset to which to truncate (inclusive)
201 * Truncate the page cache, removing the pages that are between
202 * specified offsets (and zeroing out partial pages
203 * if lstart or lend + 1 is not page aligned).
205 * Truncate takes two passes - the first pass is nonblocking. It will not
206 * block on page locks and it will not block on writeback. The second pass
207 * will wait. This is to prevent as much IO as possible in the affected region.
208 * The first pass will remove most pages, so the search cost of the second pass
211 * We pass down the cache-hot hint to the page freeing code. Even if the
212 * mapping is large, it is probably the case that the final pages are the most
213 * recently touched, and freeing happens in ascending file offset order.
215 * Note that since ->invalidatepage() accepts range to invalidate
216 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
217 * page aligned properly.
219 void truncate_inode_pages_range(struct address_space *mapping,
220 loff_t lstart, loff_t lend)
222 pgoff_t start; /* inclusive */
223 pgoff_t end; /* exclusive */
224 unsigned int partial_start; /* inclusive */
225 unsigned int partial_end; /* exclusive */
227 pgoff_t indices[PAGEVEC_SIZE];
231 cleancache_invalidate_inode(mapping);
232 if (mapping->nrpages == 0)
235 /* Offsets within partial pages */
236 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
237 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
240 * 'start' and 'end' always covers the range of pages to be fully
241 * truncated. Partial pages are covered with 'partial_start' at the
242 * start of the range and 'partial_end' at the end of the range.
243 * Note that 'end' is exclusive while 'lend' is inclusive.
245 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
248 * lend == -1 indicates end-of-file so we have to set 'end'
249 * to the highest possible pgoff_t and since the type is
250 * unsigned we're using -1.
254 end = (lend + 1) >> PAGE_CACHE_SHIFT;
256 pagevec_init(&pvec, 0);
258 while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
259 min(end - index, (pgoff_t)PAGEVEC_SIZE),
261 mem_cgroup_uncharge_start();
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);
287 mem_cgroup_uncharge_end();
293 struct page *page = find_lock_page(mapping, start - 1);
295 unsigned int top = PAGE_CACHE_SIZE;
297 /* Truncation within a single page */
301 wait_on_page_writeback(page);
302 zero_user_segment(page, partial_start, top);
303 cleancache_invalidate_page(mapping, page);
304 if (page_has_private(page))
305 do_invalidatepage(page, partial_start,
306 top - partial_start);
308 page_cache_release(page);
312 struct page *page = find_lock_page(mapping, end);
314 wait_on_page_writeback(page);
315 zero_user_segment(page, 0, partial_end);
316 cleancache_invalidate_page(mapping, page);
317 if (page_has_private(page))
318 do_invalidatepage(page, 0,
321 page_cache_release(page);
325 * If the truncation happened within a single page no pages
326 * will be released, just zeroed, so we can bail out now.
334 if (!pagevec_lookup_entries(&pvec, mapping, index,
335 min(end - index, (pgoff_t)PAGEVEC_SIZE),
342 if (index == start && indices[0] >= end) {
343 pagevec_remove_exceptionals(&pvec);
344 pagevec_release(&pvec);
347 mem_cgroup_uncharge_start();
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 */
356 if (radix_tree_exceptional_entry(page)) {
357 clear_exceptional_entry(mapping, index, page);
362 WARN_ON(page->index != index);
363 wait_on_page_writeback(page);
364 truncate_inode_page(mapping, page);
367 pagevec_remove_exceptionals(&pvec);
368 pagevec_release(&pvec);
369 mem_cgroup_uncharge_end();
372 cleancache_invalidate_inode(mapping);
374 EXPORT_SYMBOL(truncate_inode_pages_range);
377 * truncate_inode_pages - truncate *all* the pages from an offset
378 * @mapping: mapping to truncate
379 * @lstart: offset from which to truncate
381 * Called under (and serialised by) inode->i_mutex.
383 * Note: When this function returns, there can be a page in the process of
384 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
385 * mapping->nrpages can be non-zero when this function returns even after
386 * truncation of the whole mapping.
388 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
390 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
392 EXPORT_SYMBOL(truncate_inode_pages);
395 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
396 * @mapping: the address_space which holds the pages to invalidate
397 * @start: the offset 'from' which to invalidate
398 * @end: the offset 'to' which to invalidate (inclusive)
400 * This function only removes the unlocked pages, if you want to
401 * remove all the pages of one inode, you must call truncate_inode_pages.
403 * invalidate_mapping_pages() will not block on IO activity. It will not
404 * invalidate pages which are dirty, locked, under writeback or mapped into
407 unsigned long invalidate_mapping_pages(struct address_space *mapping,
408 pgoff_t start, pgoff_t end)
410 pgoff_t indices[PAGEVEC_SIZE];
412 pgoff_t index = start;
414 unsigned long count = 0;
418 * Note: this function may get called on a shmem/tmpfs mapping:
419 * pagevec_lookup() might then return 0 prematurely (because it
420 * got a gangful of swap entries); but it's hardly worth worrying
421 * about - it can rarely have anything to free from such a mapping
422 * (most pages are dirty), and already skips over any difficulties.
425 pagevec_init(&pvec, 0);
426 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
427 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
429 mem_cgroup_uncharge_start();
430 for (i = 0; i < pagevec_count(&pvec); i++) {
431 struct page *page = pvec.pages[i];
433 /* We rely upon deletion not changing page->index */
438 if (radix_tree_exceptional_entry(page)) {
439 clear_exceptional_entry(mapping, index, page);
443 if (!trylock_page(page))
445 WARN_ON(page->index != index);
446 ret = invalidate_inode_page(page);
449 * Invalidation is a hint that the page is no longer
450 * of interest and try to speed up its reclaim.
453 deactivate_page(page);
456 pagevec_remove_exceptionals(&pvec);
457 pagevec_release(&pvec);
458 mem_cgroup_uncharge_end();
464 EXPORT_SYMBOL(invalidate_mapping_pages);
467 * This is like invalidate_complete_page(), except it ignores the page's
468 * refcount. We do this because invalidate_inode_pages2() needs stronger
469 * invalidation guarantees, and cannot afford to leave pages behind because
470 * shrink_page_list() has a temp ref on them, or because they're transiently
471 * sitting in the lru_cache_add() pagevecs.
474 invalidate_complete_page2(struct address_space *mapping, struct page *page)
476 if (page->mapping != mapping)
479 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
482 spin_lock_irq(&mapping->tree_lock);
486 BUG_ON(page_has_private(page));
487 __delete_from_page_cache(page);
488 spin_unlock_irq(&mapping->tree_lock);
489 mem_cgroup_uncharge_cache_page(page);
491 if (mapping->a_ops->freepage)
492 mapping->a_ops->freepage(page);
494 page_cache_release(page); /* pagecache ref */
497 spin_unlock_irq(&mapping->tree_lock);
501 static int do_launder_page(struct address_space *mapping, struct page *page)
503 if (!PageDirty(page))
505 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
507 return mapping->a_ops->launder_page(page);
511 * invalidate_inode_pages2_range - remove range of pages from an address_space
512 * @mapping: the address_space
513 * @start: the page offset 'from' which to invalidate
514 * @end: the page offset 'to' which to invalidate (inclusive)
516 * Any pages which are found to be mapped into pagetables are unmapped prior to
519 * Returns -EBUSY if any pages could not be invalidated.
521 int invalidate_inode_pages2_range(struct address_space *mapping,
522 pgoff_t start, pgoff_t end)
524 pgoff_t indices[PAGEVEC_SIZE];
530 int did_range_unmap = 0;
532 cleancache_invalidate_inode(mapping);
533 pagevec_init(&pvec, 0);
535 while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
536 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
538 mem_cgroup_uncharge_start();
539 for (i = 0; i < pagevec_count(&pvec); i++) {
540 struct page *page = pvec.pages[i];
542 /* We rely upon deletion not changing page->index */
547 if (radix_tree_exceptional_entry(page)) {
548 clear_exceptional_entry(mapping, index, page);
553 WARN_ON(page->index != index);
554 if (page->mapping != mapping) {
558 wait_on_page_writeback(page);
559 if (page_mapped(page)) {
560 if (!did_range_unmap) {
562 * Zap the rest of the file in one hit.
564 unmap_mapping_range(mapping,
565 (loff_t)index << PAGE_CACHE_SHIFT,
566 (loff_t)(1 + end - index)
574 unmap_mapping_range(mapping,
575 (loff_t)index << PAGE_CACHE_SHIFT,
579 BUG_ON(page_mapped(page));
580 ret2 = do_launder_page(mapping, page);
582 if (!invalidate_complete_page2(mapping, page))
589 pagevec_remove_exceptionals(&pvec);
590 pagevec_release(&pvec);
591 mem_cgroup_uncharge_end();
595 cleancache_invalidate_inode(mapping);
598 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
601 * invalidate_inode_pages2 - remove all pages from an address_space
602 * @mapping: the address_space
604 * Any pages which are found to be mapped into pagetables are unmapped prior to
607 * Returns -EBUSY if any pages could not be invalidated.
609 int invalidate_inode_pages2(struct address_space *mapping)
611 return invalidate_inode_pages2_range(mapping, 0, -1);
613 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
616 * truncate_pagecache - unmap and remove pagecache that has been truncated
618 * @newsize: new file size
620 * inode's new i_size must already be written before truncate_pagecache
623 * This function should typically be called before the filesystem
624 * releases resources associated with the freed range (eg. deallocates
625 * blocks). This way, pagecache will always stay logically coherent
626 * with on-disk format, and the filesystem would not have to deal with
627 * situations such as writepage being called for a page that has already
628 * had its underlying blocks deallocated.
630 void truncate_pagecache(struct inode *inode, loff_t newsize)
632 struct address_space *mapping = inode->i_mapping;
633 loff_t holebegin = round_up(newsize, PAGE_SIZE);
636 * unmap_mapping_range is called twice, first simply for
637 * efficiency so that truncate_inode_pages does fewer
638 * single-page unmaps. However after this first call, and
639 * before truncate_inode_pages finishes, it is possible for
640 * private pages to be COWed, which remain after
641 * truncate_inode_pages finishes, hence the second
642 * unmap_mapping_range call must be made for correctness.
644 unmap_mapping_range(mapping, holebegin, 0, 1);
645 truncate_inode_pages(mapping, newsize);
646 unmap_mapping_range(mapping, holebegin, 0, 1);
648 EXPORT_SYMBOL(truncate_pagecache);
651 * truncate_setsize - update inode and pagecache for a new file size
653 * @newsize: new file size
655 * truncate_setsize updates i_size and performs pagecache truncation (if
656 * necessary) to @newsize. It will be typically be called from the filesystem's
657 * setattr function when ATTR_SIZE is passed in.
659 * Must be called with inode_mutex held and before all filesystem specific
660 * block truncation has been performed.
662 void truncate_setsize(struct inode *inode, loff_t newsize)
664 i_size_write(inode, newsize);
665 truncate_pagecache(inode, newsize);
667 EXPORT_SYMBOL(truncate_setsize);
670 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
672 * @lstart: offset of beginning of hole
673 * @lend: offset of last byte of hole
675 * This function should typically be called before the filesystem
676 * releases resources associated with the freed range (eg. deallocates
677 * blocks). This way, pagecache will always stay logically coherent
678 * with on-disk format, and the filesystem would not have to deal with
679 * situations such as writepage being called for a page that has already
680 * had its underlying blocks deallocated.
682 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
684 struct address_space *mapping = inode->i_mapping;
685 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
686 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
688 * This rounding is currently just for example: unmap_mapping_range
689 * expands its hole outwards, whereas we want it to contract the hole
690 * inwards. However, existing callers of truncate_pagecache_range are
691 * doing their own page rounding first. Note that unmap_mapping_range
692 * allows holelen 0 for all, and we allow lend -1 for end of file.
696 * Unlike in truncate_pagecache, unmap_mapping_range is called only
697 * once (before truncating pagecache), and without "even_cows" flag:
698 * hole-punching should not remove private COWed pages from the hole.
700 if ((u64)unmap_end > (u64)unmap_start)
701 unmap_mapping_range(mapping, unmap_start,
702 1 + unmap_end - unmap_start, 0);
703 truncate_inode_pages_range(mapping, lstart, lend);
705 EXPORT_SYMBOL(truncate_pagecache_range);