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>
27 * do_invalidatepage - invalidate part or all of a page
28 * @page: the page which is affected
29 * @offset: the index of the truncation point
31 * do_invalidatepage() is called when all or part of the page has become
32 * invalidated by a truncate operation.
34 * do_invalidatepage() does not have to release all buffers, but it must
35 * ensure that no dirty buffer is left outside @offset and that no I/O
36 * is underway against any of the blocks which are outside the truncation
37 * point. Because the caller is about to free (and possibly reuse) those
40 void do_invalidatepage(struct page *page, unsigned long offset)
42 void (*invalidatepage)(struct page *, unsigned long);
43 invalidatepage = page->mapping->a_ops->invalidatepage;
46 invalidatepage = block_invalidatepage;
49 (*invalidatepage)(page, offset);
52 static inline void truncate_partial_page(struct page *page, unsigned partial)
54 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
55 cleancache_invalidate_page(page->mapping, page);
56 if (page_has_private(page))
57 do_invalidatepage(page, partial);
61 * This cancels just the dirty bit on the kernel page itself, it
62 * does NOT actually remove dirty bits on any mmap's that may be
63 * around. It also leaves the page tagged dirty, so any sync
64 * activity will still find it on the dirty lists, and in particular,
65 * clear_page_dirty_for_io() will still look at the dirty bits in
68 * Doing this should *normally* only ever be done when a page
69 * is truncated, and is not actually mapped anywhere at all. However,
70 * fs/buffer.c does this when it notices that somebody has cleaned
71 * out all the buffers on a page without actually doing it through
72 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
74 void cancel_dirty_page(struct page *page, unsigned int account_size)
76 if (TestClearPageDirty(page)) {
77 struct address_space *mapping = page->mapping;
78 if (mapping && mapping_cap_account_dirty(mapping)) {
79 dec_zone_page_state(page, NR_FILE_DIRTY);
80 dec_bdi_stat(mapping->backing_dev_info,
83 task_io_account_cancelled_write(account_size);
87 EXPORT_SYMBOL(cancel_dirty_page);
90 * If truncate cannot remove the fs-private metadata from the page, the page
91 * becomes orphaned. It will be left on the LRU and may even be mapped into
92 * user pagetables if we're racing with filemap_fault().
94 * We need to bale out if page->mapping is no longer equal to the original
95 * mapping. This happens a) when the VM reclaimed the page while we waited on
96 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
97 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
100 truncate_complete_page(struct address_space *mapping, struct page *page)
102 if (page->mapping != mapping)
105 if (page_has_private(page))
106 do_invalidatepage(page, 0);
108 cancel_dirty_page(page, PAGE_CACHE_SIZE);
110 clear_page_mlock(page);
111 ClearPageMappedToDisk(page);
112 delete_from_page_cache(page);
117 * This is for invalidate_mapping_pages(). That function can be called at
118 * any time, and is not supposed to throw away dirty pages. But pages can
119 * be marked dirty at any time too, so use remove_mapping which safely
120 * discards clean, unused pages.
122 * Returns non-zero if the page was successfully invalidated.
125 invalidate_complete_page(struct address_space *mapping, struct page *page)
129 if (page->mapping != mapping)
132 if (page_has_private(page) && !try_to_release_page(page, 0))
135 clear_page_mlock(page);
136 ret = remove_mapping(mapping, page);
141 int truncate_inode_page(struct address_space *mapping, struct page *page)
143 if (page_mapped(page)) {
144 unmap_mapping_range(mapping,
145 (loff_t)page->index << PAGE_CACHE_SHIFT,
148 return truncate_complete_page(mapping, page);
152 * Used to get rid of pages on hardware memory corruption.
154 int generic_error_remove_page(struct address_space *mapping, struct page *page)
159 * Only punch for normal data pages for now.
160 * Handling other types like directories would need more auditing.
162 if (!S_ISREG(mapping->host->i_mode))
164 return truncate_inode_page(mapping, page);
166 EXPORT_SYMBOL(generic_error_remove_page);
169 * Safely invalidate one page from its pagecache mapping.
170 * It only drops clean, unused pages. The page must be locked.
172 * Returns 1 if the page is successfully invalidated, otherwise 0.
174 int invalidate_inode_page(struct page *page)
176 struct address_space *mapping = page_mapping(page);
179 if (PageDirty(page) || PageWriteback(page))
181 if (page_mapped(page))
183 return invalidate_complete_page(mapping, page);
187 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
188 * @mapping: mapping to truncate
189 * @lstart: offset from which to truncate
190 * @lend: offset to which to truncate
192 * Truncate the page cache, removing the pages that are between
193 * specified offsets (and zeroing out partial page
194 * (if lstart is not page aligned)).
196 * Truncate takes two passes - the first pass is nonblocking. It will not
197 * block on page locks and it will not block on writeback. The second pass
198 * will wait. This is to prevent as much IO as possible in the affected region.
199 * The first pass will remove most pages, so the search cost of the second pass
202 * We pass down the cache-hot hint to the page freeing code. Even if the
203 * mapping is large, it is probably the case that the final pages are the most
204 * recently touched, and freeing happens in ascending file offset order.
206 void truncate_inode_pages_range(struct address_space *mapping,
207 loff_t lstart, loff_t lend)
209 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
210 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
216 cleancache_invalidate_inode(mapping);
217 if (mapping->nrpages == 0)
220 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
221 end = (lend >> PAGE_CACHE_SHIFT);
223 pagevec_init(&pvec, 0);
225 while (index <= end && pagevec_lookup(&pvec, mapping, index,
226 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
227 mem_cgroup_uncharge_start();
228 for (i = 0; i < pagevec_count(&pvec); i++) {
229 struct page *page = pvec.pages[i];
231 /* We rely upon deletion not changing page->index */
236 if (!trylock_page(page))
238 WARN_ON(page->index != index);
239 if (PageWriteback(page)) {
243 truncate_inode_page(mapping, page);
246 pagevec_release(&pvec);
247 mem_cgroup_uncharge_end();
253 struct page *page = find_lock_page(mapping, start - 1);
255 wait_on_page_writeback(page);
256 truncate_partial_page(page, partial);
258 page_cache_release(page);
265 if (!pagevec_lookup(&pvec, mapping, index,
266 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
272 if (index == start && pvec.pages[0]->index > end) {
273 pagevec_release(&pvec);
276 mem_cgroup_uncharge_start();
277 for (i = 0; i < pagevec_count(&pvec); i++) {
278 struct page *page = pvec.pages[i];
280 /* We rely upon deletion not changing page->index */
286 WARN_ON(page->index != index);
287 wait_on_page_writeback(page);
288 truncate_inode_page(mapping, page);
291 pagevec_release(&pvec);
292 mem_cgroup_uncharge_end();
295 cleancache_invalidate_inode(mapping);
297 EXPORT_SYMBOL(truncate_inode_pages_range);
300 * truncate_inode_pages - truncate *all* the pages from an offset
301 * @mapping: mapping to truncate
302 * @lstart: offset from which to truncate
304 * Called under (and serialised by) inode->i_mutex.
306 * Note: When this function returns, there can be a page in the process of
307 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
308 * mapping->nrpages can be non-zero when this function returns even after
309 * truncation of the whole mapping.
311 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
313 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
315 EXPORT_SYMBOL(truncate_inode_pages);
318 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
319 * @mapping: the address_space which holds the pages to invalidate
320 * @start: the offset 'from' which to invalidate
321 * @end: the offset 'to' which to invalidate (inclusive)
323 * This function only removes the unlocked pages, if you want to
324 * remove all the pages of one inode, you must call truncate_inode_pages.
326 * invalidate_mapping_pages() will not block on IO activity. It will not
327 * invalidate pages which are dirty, locked, under writeback or mapped into
330 unsigned long invalidate_mapping_pages(struct address_space *mapping,
331 pgoff_t start, pgoff_t end)
334 pgoff_t index = start;
336 unsigned long count = 0;
340 * Note: this function may get called on a shmem/tmpfs mapping:
341 * pagevec_lookup() might then return 0 prematurely (because it
342 * got a gangful of swap entries); but it's hardly worth worrying
343 * about - it can rarely have anything to free from such a mapping
344 * (most pages are dirty), and already skips over any difficulties.
347 pagevec_init(&pvec, 0);
348 while (index <= end && pagevec_lookup(&pvec, mapping, index,
349 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
350 mem_cgroup_uncharge_start();
351 for (i = 0; i < pagevec_count(&pvec); i++) {
352 struct page *page = pvec.pages[i];
354 /* We rely upon deletion not changing page->index */
359 if (!trylock_page(page))
361 WARN_ON(page->index != index);
362 ret = invalidate_inode_page(page);
365 * Invalidation is a hint that the page is no longer
366 * of interest and try to speed up its reclaim.
369 deactivate_page(page);
372 pagevec_release(&pvec);
373 mem_cgroup_uncharge_end();
379 EXPORT_SYMBOL(invalidate_mapping_pages);
382 * This is like invalidate_complete_page(), except it ignores the page's
383 * refcount. We do this because invalidate_inode_pages2() needs stronger
384 * invalidation guarantees, and cannot afford to leave pages behind because
385 * shrink_page_list() has a temp ref on them, or because they're transiently
386 * sitting in the lru_cache_add() pagevecs.
389 invalidate_complete_page2(struct address_space *mapping, struct page *page)
391 if (page->mapping != mapping)
394 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
397 spin_lock_irq(&mapping->tree_lock);
401 clear_page_mlock(page);
402 BUG_ON(page_has_private(page));
403 __delete_from_page_cache(page);
404 spin_unlock_irq(&mapping->tree_lock);
405 mem_cgroup_uncharge_cache_page(page);
407 if (mapping->a_ops->freepage)
408 mapping->a_ops->freepage(page);
410 page_cache_release(page); /* pagecache ref */
413 spin_unlock_irq(&mapping->tree_lock);
417 static int do_launder_page(struct address_space *mapping, struct page *page)
419 if (!PageDirty(page))
421 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
423 return mapping->a_ops->launder_page(page);
427 * invalidate_inode_pages2_range - remove range of pages from an address_space
428 * @mapping: the address_space
429 * @start: the page offset 'from' which to invalidate
430 * @end: the page offset 'to' which to invalidate (inclusive)
432 * Any pages which are found to be mapped into pagetables are unmapped prior to
435 * Returns -EBUSY if any pages could not be invalidated.
437 int invalidate_inode_pages2_range(struct address_space *mapping,
438 pgoff_t start, pgoff_t end)
445 int did_range_unmap = 0;
447 cleancache_invalidate_inode(mapping);
448 pagevec_init(&pvec, 0);
450 while (index <= end && pagevec_lookup(&pvec, mapping, index,
451 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
452 mem_cgroup_uncharge_start();
453 for (i = 0; i < pagevec_count(&pvec); i++) {
454 struct page *page = pvec.pages[i];
456 /* We rely upon deletion not changing page->index */
462 WARN_ON(page->index != index);
463 if (page->mapping != mapping) {
467 wait_on_page_writeback(page);
468 if (page_mapped(page)) {
469 if (!did_range_unmap) {
471 * Zap the rest of the file in one hit.
473 unmap_mapping_range(mapping,
474 (loff_t)index << PAGE_CACHE_SHIFT,
475 (loff_t)(1 + end - index)
483 unmap_mapping_range(mapping,
484 (loff_t)index << PAGE_CACHE_SHIFT,
488 BUG_ON(page_mapped(page));
489 ret2 = do_launder_page(mapping, page);
491 if (!invalidate_complete_page2(mapping, page))
498 pagevec_release(&pvec);
499 mem_cgroup_uncharge_end();
503 cleancache_invalidate_inode(mapping);
506 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
509 * invalidate_inode_pages2 - remove all pages from an address_space
510 * @mapping: the address_space
512 * Any pages which are found to be mapped into pagetables are unmapped prior to
515 * Returns -EBUSY if any pages could not be invalidated.
517 int invalidate_inode_pages2(struct address_space *mapping)
519 return invalidate_inode_pages2_range(mapping, 0, -1);
521 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
524 * truncate_pagecache - unmap and remove pagecache that has been truncated
526 * @oldsize: old file size
527 * @newsize: new file size
529 * inode's new i_size must already be written before truncate_pagecache
532 * This function should typically be called before the filesystem
533 * releases resources associated with the freed range (eg. deallocates
534 * blocks). This way, pagecache will always stay logically coherent
535 * with on-disk format, and the filesystem would not have to deal with
536 * situations such as writepage being called for a page that has already
537 * had its underlying blocks deallocated.
539 void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize)
541 struct address_space *mapping = inode->i_mapping;
542 loff_t holebegin = round_up(newsize, PAGE_SIZE);
545 * unmap_mapping_range is called twice, first simply for
546 * efficiency so that truncate_inode_pages does fewer
547 * single-page unmaps. However after this first call, and
548 * before truncate_inode_pages finishes, it is possible for
549 * private pages to be COWed, which remain after
550 * truncate_inode_pages finishes, hence the second
551 * unmap_mapping_range call must be made for correctness.
553 unmap_mapping_range(mapping, holebegin, 0, 1);
554 truncate_inode_pages(mapping, newsize);
555 unmap_mapping_range(mapping, holebegin, 0, 1);
557 EXPORT_SYMBOL(truncate_pagecache);
560 * truncate_setsize - update inode and pagecache for a new file size
562 * @newsize: new file size
564 * truncate_setsize updates i_size and performs pagecache truncation (if
565 * necessary) to @newsize. It will be typically be called from the filesystem's
566 * setattr function when ATTR_SIZE is passed in.
568 * Must be called with inode_mutex held and before all filesystem specific
569 * block truncation has been performed.
571 void truncate_setsize(struct inode *inode, loff_t newsize)
575 oldsize = inode->i_size;
576 i_size_write(inode, newsize);
578 truncate_pagecache(inode, oldsize, newsize);
580 EXPORT_SYMBOL(truncate_setsize);
583 * vmtruncate - unmap mappings "freed" by truncate() syscall
584 * @inode: inode of the file used
585 * @newsize: file offset to start truncating
587 * This function is deprecated and truncate_setsize or truncate_pagecache
588 * should be used instead, together with filesystem specific block truncation.
590 int vmtruncate(struct inode *inode, loff_t newsize)
594 error = inode_newsize_ok(inode, newsize);
598 truncate_setsize(inode, newsize);
599 if (inode->i_op->truncate)
600 inode->i_op->truncate(inode);
603 EXPORT_SYMBOL(vmtruncate);
605 int vmtruncate_range(struct inode *inode, loff_t lstart, loff_t lend)
607 struct address_space *mapping = inode->i_mapping;
608 loff_t holebegin = round_up(lstart, PAGE_SIZE);
609 loff_t holelen = 1 + lend - holebegin;
612 * If the underlying filesystem is not going to provide
613 * a way to truncate a range of blocks (punch a hole) -
614 * we should return failure right now.
616 if (!inode->i_op->truncate_range)
619 mutex_lock(&inode->i_mutex);
620 inode_dio_wait(inode);
621 unmap_mapping_range(mapping, holebegin, holelen, 1);
622 inode->i_op->truncate_range(inode, lstart, lend);
623 /* unmap again to remove racily COWed private pages */
624 unmap_mapping_range(mapping, holebegin, holelen, 1);
625 mutex_unlock(&inode->i_mutex);
631 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
633 * @lstart: offset of beginning of hole
634 * @lend: offset of last byte of hole
636 * This function should typically be called before the filesystem
637 * releases resources associated with the freed range (eg. deallocates
638 * blocks). This way, pagecache will always stay logically coherent
639 * with on-disk format, and the filesystem would not have to deal with
640 * situations such as writepage being called for a page that has already
641 * had its underlying blocks deallocated.
643 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
645 struct address_space *mapping = inode->i_mapping;
646 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
647 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
649 * This rounding is currently just for example: unmap_mapping_range
650 * expands its hole outwards, whereas we want it to contract the hole
651 * inwards. However, existing callers of truncate_pagecache_range are
652 * doing their own page rounding first; and truncate_inode_pages_range
653 * currently BUGs if lend is not pagealigned-1 (it handles partial
654 * page at start of hole, but not partial page at end of hole). Note
655 * unmap_mapping_range allows holelen 0 for all, and we allow lend -1.
659 * Unlike in truncate_pagecache, unmap_mapping_range is called only
660 * once (before truncating pagecache), and without "even_cows" flag:
661 * hole-punching should not remove private COWed pages from the hole.
663 if ((u64)unmap_end > (u64)unmap_start)
664 unmap_mapping_range(mapping, unmap_start,
665 1 + unmap_end - unmap_start, 0);
666 truncate_inode_pages_range(mapping, lstart, lend);
668 EXPORT_SYMBOL(truncate_pagecache_range);