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: the index of the truncation point
32 * do_invalidatepage() is called when all or part of the page has become
33 * invalidated by a truncate operation.
35 * do_invalidatepage() does not have to release all buffers, but it must
36 * ensure that no dirty buffer is left outside @offset and that no I/O
37 * is underway against any of the blocks which are outside the truncation
38 * point. Because the caller is about to free (and possibly reuse) those
41 void do_invalidatepage(struct page *page, unsigned long offset)
43 void (*invalidatepage)(struct page *, unsigned long);
44 invalidatepage = page->mapping->a_ops->invalidatepage;
47 invalidatepage = block_invalidatepage;
50 (*invalidatepage)(page, offset);
53 static inline void truncate_partial_page(struct page *page, unsigned partial)
55 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
56 cleancache_flush_page(page->mapping, page);
57 if (page_has_private(page))
58 do_invalidatepage(page, partial);
62 * This cancels just the dirty bit on the kernel page itself, it
63 * does NOT actually remove dirty bits on any mmap's that may be
64 * around. It also leaves the page tagged dirty, so any sync
65 * activity will still find it on the dirty lists, and in particular,
66 * clear_page_dirty_for_io() will still look at the dirty bits in
69 * Doing this should *normally* only ever be done when a page
70 * is truncated, and is not actually mapped anywhere at all. However,
71 * fs/buffer.c does this when it notices that somebody has cleaned
72 * out all the buffers on a page without actually doing it through
73 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
75 void cancel_dirty_page(struct page *page, unsigned int account_size)
77 if (TestClearPageDirty(page)) {
78 struct address_space *mapping = page->mapping;
79 if (mapping && mapping_cap_account_dirty(mapping)) {
80 dec_zone_page_state(page, NR_FILE_DIRTY);
81 dec_bdi_stat(mapping->backing_dev_info,
84 task_io_account_cancelled_write(account_size);
88 EXPORT_SYMBOL(cancel_dirty_page);
91 * If truncate cannot remove the fs-private metadata from the page, the page
92 * becomes orphaned. It will be left on the LRU and may even be mapped into
93 * user pagetables if we're racing with filemap_fault().
95 * We need to bale out if page->mapping is no longer equal to the original
96 * mapping. This happens a) when the VM reclaimed the page while we waited on
97 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
98 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
101 truncate_complete_page(struct address_space *mapping, struct page *page)
103 if (page->mapping != mapping)
106 if (page_has_private(page))
107 do_invalidatepage(page, 0);
109 cancel_dirty_page(page, PAGE_CACHE_SIZE);
111 clear_page_mlock(page);
112 ClearPageMappedToDisk(page);
113 delete_from_page_cache(page);
118 * This is for invalidate_mapping_pages(). That function can be called at
119 * any time, and is not supposed to throw away dirty pages. But pages can
120 * be marked dirty at any time too, so use remove_mapping which safely
121 * discards clean, unused pages.
123 * Returns non-zero if the page was successfully invalidated.
126 invalidate_complete_page(struct address_space *mapping, struct page *page)
130 if (page->mapping != mapping)
133 if (page_has_private(page) && !try_to_release_page(page, 0))
136 clear_page_mlock(page);
137 ret = remove_mapping(mapping, page);
142 int truncate_inode_page(struct address_space *mapping, struct page *page)
144 if (page_mapped(page)) {
145 unmap_mapping_range(mapping,
146 (loff_t)page->index << PAGE_CACHE_SHIFT,
149 return truncate_complete_page(mapping, page);
153 * Used to get rid of pages on hardware memory corruption.
155 int generic_error_remove_page(struct address_space *mapping, struct page *page)
160 * Only punch for normal data pages for now.
161 * Handling other types like directories would need more auditing.
163 if (!S_ISREG(mapping->host->i_mode))
165 return truncate_inode_page(mapping, page);
167 EXPORT_SYMBOL(generic_error_remove_page);
170 * Safely invalidate one page from its pagecache mapping.
171 * It only drops clean, unused pages. The page must be locked.
173 * Returns 1 if the page is successfully invalidated, otherwise 0.
175 int invalidate_inode_page(struct page *page)
177 struct address_space *mapping = page_mapping(page);
180 if (PageDirty(page) || PageWriteback(page))
182 if (page_mapped(page))
184 return invalidate_complete_page(mapping, page);
188 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
189 * @mapping: mapping to truncate
190 * @lstart: offset from which to truncate
191 * @lend: offset to which to truncate
193 * Truncate the page cache, removing the pages that are between
194 * specified offsets (and zeroing out partial page
195 * (if lstart is not page aligned)).
197 * Truncate takes two passes - the first pass is nonblocking. It will not
198 * block on page locks and it will not block on writeback. The second pass
199 * will wait. This is to prevent as much IO as possible in the affected region.
200 * The first pass will remove most pages, so the search cost of the second pass
203 * We pass down the cache-hot hint to the page freeing code. Even if the
204 * mapping is large, it is probably the case that the final pages are the most
205 * recently touched, and freeing happens in ascending file offset order.
207 void truncate_inode_pages_range(struct address_space *mapping,
208 loff_t lstart, loff_t lend)
210 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
211 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
217 cleancache_flush_inode(mapping);
218 if (mapping->nrpages == 0)
221 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
222 end = (lend >> PAGE_CACHE_SHIFT);
224 pagevec_init(&pvec, 0);
226 while (index <= end && pagevec_lookup(&pvec, mapping, index,
227 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
228 mem_cgroup_uncharge_start();
229 for (i = 0; i < pagevec_count(&pvec); i++) {
230 struct page *page = pvec.pages[i];
232 /* We rely upon deletion not changing page->index */
237 if (!trylock_page(page))
239 WARN_ON(page->index != index);
240 if (PageWriteback(page)) {
244 truncate_inode_page(mapping, page);
247 pagevec_release(&pvec);
248 mem_cgroup_uncharge_end();
254 struct page *page = find_lock_page(mapping, start - 1);
256 wait_on_page_writeback(page);
257 truncate_partial_page(page, partial);
259 page_cache_release(page);
266 if (!pagevec_lookup(&pvec, mapping, index,
267 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
273 if (index == start && pvec.pages[0]->index > end) {
274 pagevec_release(&pvec);
277 mem_cgroup_uncharge_start();
278 for (i = 0; i < pagevec_count(&pvec); i++) {
279 struct page *page = pvec.pages[i];
281 /* We rely upon deletion not changing page->index */
287 WARN_ON(page->index != index);
288 wait_on_page_writeback(page);
289 truncate_inode_page(mapping, page);
292 pagevec_release(&pvec);
293 mem_cgroup_uncharge_end();
296 cleancache_flush_inode(mapping);
298 EXPORT_SYMBOL(truncate_inode_pages_range);
301 * truncate_inode_pages - truncate *all* the pages from an offset
302 * @mapping: mapping to truncate
303 * @lstart: offset from which to truncate
305 * Called under (and serialised by) inode->i_mutex.
307 * Note: When this function returns, there can be a page in the process of
308 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
309 * mapping->nrpages can be non-zero when this function returns even after
310 * truncation of the whole mapping.
312 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
314 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
316 EXPORT_SYMBOL(truncate_inode_pages);
319 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
320 * @mapping: the address_space which holds the pages to invalidate
321 * @start: the offset 'from' which to invalidate
322 * @end: the offset 'to' which to invalidate (inclusive)
324 * This function only removes the unlocked pages, if you want to
325 * remove all the pages of one inode, you must call truncate_inode_pages.
327 * invalidate_mapping_pages() will not block on IO activity. It will not
328 * invalidate pages which are dirty, locked, under writeback or mapped into
331 unsigned long invalidate_mapping_pages(struct address_space *mapping,
332 pgoff_t start, pgoff_t end)
335 pgoff_t index = start;
337 unsigned long count = 0;
341 * Note: this function may get called on a shmem/tmpfs mapping:
342 * pagevec_lookup() might then return 0 prematurely (because it
343 * got a gangful of swap entries); but it's hardly worth worrying
344 * about - it can rarely have anything to free from such a mapping
345 * (most pages are dirty), and already skips over any difficulties.
348 pagevec_init(&pvec, 0);
349 while (index <= end && pagevec_lookup(&pvec, mapping, index,
350 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
351 mem_cgroup_uncharge_start();
352 for (i = 0; i < pagevec_count(&pvec); i++) {
353 struct page *page = pvec.pages[i];
355 /* We rely upon deletion not changing page->index */
360 if (!trylock_page(page))
362 WARN_ON(page->index != index);
363 ret = invalidate_inode_page(page);
366 * Invalidation is a hint that the page is no longer
367 * of interest and try to speed up its reclaim.
370 deactivate_page(page);
373 pagevec_release(&pvec);
374 mem_cgroup_uncharge_end();
380 EXPORT_SYMBOL(invalidate_mapping_pages);
383 * This is like invalidate_complete_page(), except it ignores the page's
384 * refcount. We do this because invalidate_inode_pages2() needs stronger
385 * invalidation guarantees, and cannot afford to leave pages behind because
386 * shrink_page_list() has a temp ref on them, or because they're transiently
387 * sitting in the lru_cache_add() pagevecs.
390 invalidate_complete_page2(struct address_space *mapping, struct page *page)
392 if (page->mapping != mapping)
395 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
398 clear_page_mlock(page);
400 spin_lock_irq(&mapping->tree_lock);
404 BUG_ON(page_has_private(page));
405 __delete_from_page_cache(page);
406 spin_unlock_irq(&mapping->tree_lock);
407 mem_cgroup_uncharge_cache_page(page);
409 if (mapping->a_ops->freepage)
410 mapping->a_ops->freepage(page);
412 page_cache_release(page); /* pagecache ref */
415 spin_unlock_irq(&mapping->tree_lock);
419 static int do_launder_page(struct address_space *mapping, struct page *page)
421 if (!PageDirty(page))
423 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
425 return mapping->a_ops->launder_page(page);
429 * invalidate_inode_pages2_range - remove range of pages from an address_space
430 * @mapping: the address_space
431 * @start: the page offset 'from' which to invalidate
432 * @end: the page offset 'to' which to invalidate (inclusive)
434 * Any pages which are found to be mapped into pagetables are unmapped prior to
437 * Returns -EBUSY if any pages could not be invalidated.
439 int invalidate_inode_pages2_range(struct address_space *mapping,
440 pgoff_t start, pgoff_t end)
447 int did_range_unmap = 0;
449 cleancache_flush_inode(mapping);
450 pagevec_init(&pvec, 0);
452 while (index <= end && pagevec_lookup(&pvec, mapping, index,
453 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
454 mem_cgroup_uncharge_start();
455 for (i = 0; i < pagevec_count(&pvec); i++) {
456 struct page *page = pvec.pages[i];
458 /* We rely upon deletion not changing page->index */
464 WARN_ON(page->index != index);
465 if (page->mapping != mapping) {
469 wait_on_page_writeback(page);
470 if (page_mapped(page)) {
471 if (!did_range_unmap) {
473 * Zap the rest of the file in one hit.
475 unmap_mapping_range(mapping,
476 (loff_t)index << PAGE_CACHE_SHIFT,
477 (loff_t)(1 + end - index)
485 unmap_mapping_range(mapping,
486 (loff_t)index << PAGE_CACHE_SHIFT,
490 BUG_ON(page_mapped(page));
491 ret2 = do_launder_page(mapping, page);
493 if (!invalidate_complete_page2(mapping, page))
500 pagevec_release(&pvec);
501 mem_cgroup_uncharge_end();
505 cleancache_flush_inode(mapping);
508 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
511 * invalidate_inode_pages2 - remove all pages from an address_space
512 * @mapping: the address_space
514 * Any pages which are found to be mapped into pagetables are unmapped prior to
517 * Returns -EBUSY if any pages could not be invalidated.
519 int invalidate_inode_pages2(struct address_space *mapping)
521 return invalidate_inode_pages2_range(mapping, 0, -1);
523 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
526 * truncate_pagecache - unmap and remove pagecache that has been truncated
528 * @oldsize: old file size
529 * @newsize: new file size
531 * inode's new i_size must already be written before truncate_pagecache
534 * This function should typically be called before the filesystem
535 * releases resources associated with the freed range (eg. deallocates
536 * blocks). This way, pagecache will always stay logically coherent
537 * with on-disk format, and the filesystem would not have to deal with
538 * situations such as writepage being called for a page that has already
539 * had its underlying blocks deallocated.
541 void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize)
543 struct address_space *mapping = inode->i_mapping;
544 loff_t holebegin = round_up(newsize, PAGE_SIZE);
547 * unmap_mapping_range is called twice, first simply for
548 * efficiency so that truncate_inode_pages does fewer
549 * single-page unmaps. However after this first call, and
550 * before truncate_inode_pages finishes, it is possible for
551 * private pages to be COWed, which remain after
552 * truncate_inode_pages finishes, hence the second
553 * unmap_mapping_range call must be made for correctness.
555 unmap_mapping_range(mapping, holebegin, 0, 1);
556 truncate_inode_pages(mapping, newsize);
557 unmap_mapping_range(mapping, holebegin, 0, 1);
559 EXPORT_SYMBOL(truncate_pagecache);
562 * truncate_setsize - update inode and pagecache for a new file size
564 * @newsize: new file size
566 * truncate_setsize updates i_size and performs pagecache truncation (if
567 * necessary) to @newsize. It will be typically be called from the filesystem's
568 * setattr function when ATTR_SIZE is passed in.
570 * Must be called with inode_mutex held and before all filesystem specific
571 * block truncation has been performed.
573 void truncate_setsize(struct inode *inode, loff_t newsize)
577 oldsize = inode->i_size;
578 i_size_write(inode, newsize);
579 if (newsize > oldsize)
580 pagecache_isize_extended(inode, oldsize, newsize);
581 truncate_pagecache(inode, oldsize, newsize);
583 EXPORT_SYMBOL(truncate_setsize);
586 * pagecache_isize_extended - update pagecache after extension of i_size
587 * @inode: inode for which i_size was extended
588 * @from: original inode size
589 * @to: new inode size
591 * Handle extension of inode size either caused by extending truncate or by
592 * write starting after current i_size. We mark the page straddling current
593 * i_size RO so that page_mkwrite() is called on the nearest write access to
594 * the page. This way filesystem can be sure that page_mkwrite() is called on
595 * the page before user writes to the page via mmap after the i_size has been
598 * The function must be called after i_size is updated so that page fault
599 * coming after we unlock the page will already see the new i_size.
600 * The function must be called while we still hold i_mutex - this not only
601 * makes sure i_size is stable but also that userspace cannot observe new
602 * i_size value before we are prepared to store mmap writes at new inode size.
604 void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
606 int bsize = 1 << inode->i_blkbits;
611 WARN_ON(to > inode->i_size);
613 if (from >= to || bsize == PAGE_CACHE_SIZE)
615 /* Page straddling @from will not have any hole block created? */
616 rounded_from = round_up(from, bsize);
617 if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
620 index = from >> PAGE_CACHE_SHIFT;
621 page = find_lock_page(inode->i_mapping, index);
622 /* Page not cached? Nothing to do */
626 * See clear_page_dirty_for_io() for details why set_page_dirty()
629 if (page_mkclean(page))
630 set_page_dirty(page);
632 page_cache_release(page);
634 EXPORT_SYMBOL(pagecache_isize_extended);
637 * vmtruncate - unmap mappings "freed" by truncate() syscall
638 * @inode: inode of the file used
639 * @newsize: file offset to start truncating
641 * This function is deprecated and truncate_setsize or truncate_pagecache
642 * should be used instead, together with filesystem specific block truncation.
644 int vmtruncate(struct inode *inode, loff_t newsize)
648 error = inode_newsize_ok(inode, newsize);
652 truncate_setsize(inode, newsize);
653 if (inode->i_op->truncate)
654 inode->i_op->truncate(inode);
657 EXPORT_SYMBOL(vmtruncate);