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>
28 * do_invalidatepage - invalidate part or all of a page
29 * @page: the page which is affected
30 * @offset: start of the range to invalidate
31 * @length: length of the range to invalidate
33 * do_invalidatepage() is called when all or part of the page has become
34 * invalidated by a truncate operation.
36 * do_invalidatepage() does not have to release all buffers, but it must
37 * ensure that no dirty buffer is left outside @offset and that no I/O
38 * is underway against any of the blocks which are outside the truncation
39 * point. Because the caller is about to free (and possibly reuse) those
42 void do_invalidatepage(struct page *page, unsigned int offset,
45 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
47 invalidatepage = page->mapping->a_ops->invalidatepage;
50 invalidatepage = block_invalidatepage;
53 (*invalidatepage)(page, offset, length);
57 * This cancels just the dirty bit on the kernel page itself, it
58 * does NOT actually remove dirty bits on any mmap's that may be
59 * around. It also leaves the page tagged dirty, so any sync
60 * activity will still find it on the dirty lists, and in particular,
61 * clear_page_dirty_for_io() will still look at the dirty bits in
64 * Doing this should *normally* only ever be done when a page
65 * is truncated, and is not actually mapped anywhere at all. However,
66 * fs/buffer.c does this when it notices that somebody has cleaned
67 * out all the buffers on a page without actually doing it through
68 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
70 void cancel_dirty_page(struct page *page, unsigned int account_size)
72 if (TestClearPageDirty(page)) {
73 struct address_space *mapping = page->mapping;
74 if (mapping && mapping_cap_account_dirty(mapping)) {
75 dec_zone_page_state(page, NR_FILE_DIRTY);
76 dec_bdi_stat(mapping->backing_dev_info,
79 task_io_account_cancelled_write(account_size);
83 EXPORT_SYMBOL(cancel_dirty_page);
86 * If truncate cannot remove the fs-private metadata from the page, the page
87 * becomes orphaned. It will be left on the LRU and may even be mapped into
88 * user pagetables if we're racing with filemap_fault().
90 * We need to bale out if page->mapping is no longer equal to the original
91 * mapping. This happens a) when the VM reclaimed the page while we waited on
92 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
93 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
96 truncate_complete_page(struct address_space *mapping, struct page *page)
98 if (page->mapping != mapping)
101 if (page_has_private(page))
102 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
104 cancel_dirty_page(page, PAGE_CACHE_SIZE);
106 ClearPageMappedToDisk(page);
107 delete_from_page_cache(page);
112 * This is for invalidate_mapping_pages(). That function can be called at
113 * any time, and is not supposed to throw away dirty pages. But pages can
114 * be marked dirty at any time too, so use remove_mapping which safely
115 * discards clean, unused pages.
117 * Returns non-zero if the page was successfully invalidated.
120 invalidate_complete_page(struct address_space *mapping, struct page *page)
124 if (page->mapping != mapping)
127 if (page_has_private(page) && !try_to_release_page(page, 0))
130 ret = remove_mapping(mapping, page);
135 int truncate_inode_page(struct address_space *mapping, struct page *page)
137 if (page_mapped(page)) {
138 unmap_mapping_range(mapping,
139 (loff_t)page->index << PAGE_CACHE_SHIFT,
142 return truncate_complete_page(mapping, page);
146 * Used to get rid of pages on hardware memory corruption.
148 int generic_error_remove_page(struct address_space *mapping, struct page *page)
153 * Only punch for normal data pages for now.
154 * Handling other types like directories would need more auditing.
156 if (!S_ISREG(mapping->host->i_mode))
158 return truncate_inode_page(mapping, page);
160 EXPORT_SYMBOL(generic_error_remove_page);
163 * Safely invalidate one page from its pagecache mapping.
164 * It only drops clean, unused pages. The page must be locked.
166 * Returns 1 if the page is successfully invalidated, otherwise 0.
168 int invalidate_inode_page(struct page *page)
170 struct address_space *mapping = page_mapping(page);
173 if (PageDirty(page) || PageWriteback(page))
175 if (page_mapped(page))
177 return invalidate_complete_page(mapping, page);
181 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
182 * @mapping: mapping to truncate
183 * @lstart: offset from which to truncate
184 * @lend: offset to which to truncate (inclusive)
186 * Truncate the page cache, removing the pages that are between
187 * specified offsets (and zeroing out partial pages
188 * if lstart or lend + 1 is not page aligned).
190 * Truncate takes two passes - the first pass is nonblocking. It will not
191 * block on page locks and it will not block on writeback. The second pass
192 * will wait. This is to prevent as much IO as possible in the affected region.
193 * The first pass will remove most pages, so the search cost of the second pass
196 * We pass down the cache-hot hint to the page freeing code. Even if the
197 * mapping is large, it is probably the case that the final pages are the most
198 * recently touched, and freeing happens in ascending file offset order.
200 * Note that since ->invalidatepage() accepts range to invalidate
201 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
202 * page aligned properly.
204 void truncate_inode_pages_range(struct address_space *mapping,
205 loff_t lstart, loff_t lend)
207 pgoff_t start; /* inclusive */
208 pgoff_t end; /* exclusive */
209 unsigned int partial_start; /* inclusive */
210 unsigned int partial_end; /* exclusive */
215 cleancache_invalidate_inode(mapping);
216 if (mapping->nrpages == 0)
219 /* Offsets within partial pages */
220 partial_start = lstart & (PAGE_CACHE_SIZE - 1);
221 partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
224 * 'start' and 'end' always covers the range of pages to be fully
225 * truncated. Partial pages are covered with 'partial_start' at the
226 * start of the range and 'partial_end' at the end of the range.
227 * Note that 'end' is exclusive while 'lend' is inclusive.
229 start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
232 * lend == -1 indicates end-of-file so we have to set 'end'
233 * to the highest possible pgoff_t and since the type is
234 * unsigned we're using -1.
238 end = (lend + 1) >> PAGE_CACHE_SHIFT;
240 pagevec_init(&pvec, 0);
242 while (index < end && pagevec_lookup(&pvec, mapping, index,
243 min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
244 mem_cgroup_uncharge_start();
245 for (i = 0; i < pagevec_count(&pvec); i++) {
246 struct page *page = pvec.pages[i];
248 /* We rely upon deletion not changing page->index */
253 if (!trylock_page(page))
255 WARN_ON(page->index != index);
256 if (PageWriteback(page)) {
260 truncate_inode_page(mapping, page);
263 pagevec_release(&pvec);
264 mem_cgroup_uncharge_end();
270 struct page *page = find_lock_page(mapping, start - 1);
272 unsigned int top = PAGE_CACHE_SIZE;
274 /* Truncation within a single page */
278 wait_on_page_writeback(page);
279 zero_user_segment(page, partial_start, top);
280 cleancache_invalidate_page(mapping, page);
281 if (page_has_private(page))
282 do_invalidatepage(page, partial_start,
283 top - partial_start);
285 page_cache_release(page);
289 struct page *page = find_lock_page(mapping, end);
291 wait_on_page_writeback(page);
292 zero_user_segment(page, 0, partial_end);
293 cleancache_invalidate_page(mapping, page);
294 if (page_has_private(page))
295 do_invalidatepage(page, 0,
298 page_cache_release(page);
302 * If the truncation happened within a single page no pages
303 * will be released, just zeroed, so we can bail out now.
311 if (!pagevec_lookup(&pvec, mapping, index,
312 min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
318 if (index == start && pvec.pages[0]->index >= end) {
319 pagevec_release(&pvec);
322 mem_cgroup_uncharge_start();
323 for (i = 0; i < pagevec_count(&pvec); i++) {
324 struct page *page = pvec.pages[i];
326 /* We rely upon deletion not changing page->index */
332 WARN_ON(page->index != index);
333 wait_on_page_writeback(page);
334 truncate_inode_page(mapping, page);
337 pagevec_release(&pvec);
338 mem_cgroup_uncharge_end();
341 cleancache_invalidate_inode(mapping);
343 EXPORT_SYMBOL(truncate_inode_pages_range);
346 * truncate_inode_pages - truncate *all* the pages from an offset
347 * @mapping: mapping to truncate
348 * @lstart: offset from which to truncate
350 * Called under (and serialised by) inode->i_mutex.
352 * Note: When this function returns, there can be a page in the process of
353 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
354 * mapping->nrpages can be non-zero when this function returns even after
355 * truncation of the whole mapping.
357 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
359 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
361 EXPORT_SYMBOL(truncate_inode_pages);
364 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
365 * @mapping: the address_space which holds the pages to invalidate
366 * @start: the offset 'from' which to invalidate
367 * @end: the offset 'to' which to invalidate (inclusive)
369 * This function only removes the unlocked pages, if you want to
370 * remove all the pages of one inode, you must call truncate_inode_pages.
372 * invalidate_mapping_pages() will not block on IO activity. It will not
373 * invalidate pages which are dirty, locked, under writeback or mapped into
376 unsigned long invalidate_mapping_pages(struct address_space *mapping,
377 pgoff_t start, pgoff_t end)
380 pgoff_t index = start;
382 unsigned long count = 0;
386 * Note: this function may get called on a shmem/tmpfs mapping:
387 * pagevec_lookup() might then return 0 prematurely (because it
388 * got a gangful of swap entries); but it's hardly worth worrying
389 * about - it can rarely have anything to free from such a mapping
390 * (most pages are dirty), and already skips over any difficulties.
393 pagevec_init(&pvec, 0);
394 while (index <= end && pagevec_lookup(&pvec, mapping, index,
395 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
396 mem_cgroup_uncharge_start();
397 for (i = 0; i < pagevec_count(&pvec); i++) {
398 struct page *page = pvec.pages[i];
400 /* We rely upon deletion not changing page->index */
405 if (!trylock_page(page))
407 WARN_ON(page->index != index);
408 ret = invalidate_inode_page(page);
411 * Invalidation is a hint that the page is no longer
412 * of interest and try to speed up its reclaim.
415 deactivate_page(page);
418 pagevec_release(&pvec);
419 mem_cgroup_uncharge_end();
425 EXPORT_SYMBOL(invalidate_mapping_pages);
428 * This is like invalidate_complete_page(), except it ignores the page's
429 * refcount. We do this because invalidate_inode_pages2() needs stronger
430 * invalidation guarantees, and cannot afford to leave pages behind because
431 * shrink_page_list() has a temp ref on them, or because they're transiently
432 * sitting in the lru_cache_add() pagevecs.
435 invalidate_complete_page2(struct address_space *mapping, struct page *page)
437 if (page->mapping != mapping)
440 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
443 spin_lock_irq(&mapping->tree_lock);
447 BUG_ON(page_has_private(page));
448 __delete_from_page_cache(page);
449 spin_unlock_irq(&mapping->tree_lock);
450 mem_cgroup_uncharge_cache_page(page);
452 if (mapping->a_ops->freepage)
453 mapping->a_ops->freepage(page);
455 page_cache_release(page); /* pagecache ref */
458 spin_unlock_irq(&mapping->tree_lock);
462 static int do_launder_page(struct address_space *mapping, struct page *page)
464 if (!PageDirty(page))
466 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
468 return mapping->a_ops->launder_page(page);
472 * invalidate_inode_pages2_range - remove range of pages from an address_space
473 * @mapping: the address_space
474 * @start: the page offset 'from' which to invalidate
475 * @end: the page offset 'to' which to invalidate (inclusive)
477 * Any pages which are found to be mapped into pagetables are unmapped prior to
480 * Returns -EBUSY if any pages could not be invalidated.
482 int invalidate_inode_pages2_range(struct address_space *mapping,
483 pgoff_t start, pgoff_t end)
490 int did_range_unmap = 0;
492 cleancache_invalidate_inode(mapping);
493 pagevec_init(&pvec, 0);
495 while (index <= end && pagevec_lookup(&pvec, mapping, index,
496 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
497 mem_cgroup_uncharge_start();
498 for (i = 0; i < pagevec_count(&pvec); i++) {
499 struct page *page = pvec.pages[i];
501 /* We rely upon deletion not changing page->index */
507 WARN_ON(page->index != index);
508 if (page->mapping != mapping) {
512 wait_on_page_writeback(page);
513 if (page_mapped(page)) {
514 if (!did_range_unmap) {
516 * Zap the rest of the file in one hit.
518 unmap_mapping_range(mapping,
519 (loff_t)index << PAGE_CACHE_SHIFT,
520 (loff_t)(1 + end - index)
528 unmap_mapping_range(mapping,
529 (loff_t)index << PAGE_CACHE_SHIFT,
533 BUG_ON(page_mapped(page));
534 ret2 = do_launder_page(mapping, page);
536 if (!invalidate_complete_page2(mapping, page))
543 pagevec_release(&pvec);
544 mem_cgroup_uncharge_end();
548 cleancache_invalidate_inode(mapping);
551 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
554 * invalidate_inode_pages2 - remove all pages from an address_space
555 * @mapping: the address_space
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(struct address_space *mapping)
564 return invalidate_inode_pages2_range(mapping, 0, -1);
566 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
569 * truncate_pagecache - unmap and remove pagecache that has been truncated
571 * @newsize: new file size
573 * inode's new i_size must already be written before truncate_pagecache
576 * This function should typically be called before the filesystem
577 * releases resources associated with the freed range (eg. deallocates
578 * blocks). This way, pagecache will always stay logically coherent
579 * with on-disk format, and the filesystem would not have to deal with
580 * situations such as writepage being called for a page that has already
581 * had its underlying blocks deallocated.
583 void truncate_pagecache(struct inode *inode, loff_t newsize)
585 struct address_space *mapping = inode->i_mapping;
586 loff_t holebegin = round_up(newsize, PAGE_SIZE);
589 * unmap_mapping_range is called twice, first simply for
590 * efficiency so that truncate_inode_pages does fewer
591 * single-page unmaps. However after this first call, and
592 * before truncate_inode_pages finishes, it is possible for
593 * private pages to be COWed, which remain after
594 * truncate_inode_pages finishes, hence the second
595 * unmap_mapping_range call must be made for correctness.
597 unmap_mapping_range(mapping, holebegin, 0, 1);
598 truncate_inode_pages(mapping, newsize);
599 unmap_mapping_range(mapping, holebegin, 0, 1);
601 EXPORT_SYMBOL(truncate_pagecache);
604 * truncate_setsize - update inode and pagecache for a new file size
606 * @newsize: new file size
608 * truncate_setsize updates i_size and performs pagecache truncation (if
609 * necessary) to @newsize. It will be typically be called from the filesystem's
610 * setattr function when ATTR_SIZE is passed in.
612 * Must be called with inode_mutex held and before all filesystem specific
613 * block truncation has been performed.
615 void truncate_setsize(struct inode *inode, loff_t newsize)
617 loff_t oldsize = inode->i_size;
619 i_size_write(inode, newsize);
620 if (newsize > oldsize)
621 pagecache_isize_extended(inode, oldsize, newsize);
622 truncate_pagecache(inode, newsize);
624 EXPORT_SYMBOL(truncate_setsize);
627 * pagecache_isize_extended - update pagecache after extension of i_size
628 * @inode: inode for which i_size was extended
629 * @from: original inode size
630 * @to: new inode size
632 * Handle extension of inode size either caused by extending truncate or by
633 * write starting after current i_size. We mark the page straddling current
634 * i_size RO so that page_mkwrite() is called on the nearest write access to
635 * the page. This way filesystem can be sure that page_mkwrite() is called on
636 * the page before user writes to the page via mmap after the i_size has been
639 * The function must be called after i_size is updated so that page fault
640 * coming after we unlock the page will already see the new i_size.
641 * The function must be called while we still hold i_mutex - this not only
642 * makes sure i_size is stable but also that userspace cannot observe new
643 * i_size value before we are prepared to store mmap writes at new inode size.
645 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
647 int bsize = 1 << inode->i_blkbits;
652 WARN_ON(!mutex_is_locked(&inode->i_mutex));
653 WARN_ON(to > inode->i_size);
655 if (from >= to || bsize == PAGE_CACHE_SIZE)
657 /* Page straddling @from will not have any hole block created? */
658 rounded_from = round_up(from, bsize);
659 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
662 index = from >> PAGE_CACHE_SHIFT;
663 page = find_lock_page(inode->i_mapping, index);
664 /* Page not cached? Nothing to do */
668 * See clear_page_dirty_for_io() for details why set_page_dirty()
671 if (page_mkclean(page))
672 set_page_dirty(page);
674 page_cache_release(page);
676 EXPORT_SYMBOL(pagecache_isize_extended);
679 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
681 * @lstart: offset of beginning of hole
682 * @lend: offset of last byte of hole
684 * This function should typically be called before the filesystem
685 * releases resources associated with the freed range (eg. deallocates
686 * blocks). This way, pagecache will always stay logically coherent
687 * with on-disk format, and the filesystem would not have to deal with
688 * situations such as writepage being called for a page that has already
689 * had its underlying blocks deallocated.
691 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
693 struct address_space *mapping = inode->i_mapping;
694 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
695 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
697 * This rounding is currently just for example: unmap_mapping_range
698 * expands its hole outwards, whereas we want it to contract the hole
699 * inwards. However, existing callers of truncate_pagecache_range are
700 * doing their own page rounding first. Note that unmap_mapping_range
701 * allows holelen 0 for all, and we allow lend -1 for end of file.
705 * Unlike in truncate_pagecache, unmap_mapping_range is called only
706 * once (before truncating pagecache), and without "even_cows" flag:
707 * hole-punching should not remove private COWed pages from the hole.
709 if ((u64)unmap_end > (u64)unmap_start)
710 unmap_mapping_range(mapping, unmap_start,
711 1 + unmap_end - unmap_start, 0);
712 truncate_inode_pages_range(mapping, lstart, lend);
714 EXPORT_SYMBOL(truncate_pagecache_range);