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: start of the range to invalidate
30 * @length: length of the range to invalidate
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 int offset,
44 void (*invalidatepage)(struct page *, unsigned int, unsigned int);
46 invalidatepage = page->mapping->a_ops->invalidatepage;
49 invalidatepage = block_invalidatepage;
52 (*invalidatepage)(page, offset, length);
55 static inline void truncate_partial_page(struct page *page, unsigned partial)
57 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
58 cleancache_invalidate_page(page->mapping, page);
59 if (page_has_private(page))
60 do_invalidatepage(page, partial, PAGE_CACHE_SIZE - partial);
64 * This cancels just the dirty bit on the kernel page itself, it
65 * does NOT actually remove dirty bits on any mmap's that may be
66 * around. It also leaves the page tagged dirty, so any sync
67 * activity will still find it on the dirty lists, and in particular,
68 * clear_page_dirty_for_io() will still look at the dirty bits in
71 * Doing this should *normally* only ever be done when a page
72 * is truncated, and is not actually mapped anywhere at all. However,
73 * fs/buffer.c does this when it notices that somebody has cleaned
74 * out all the buffers on a page without actually doing it through
75 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
77 void cancel_dirty_page(struct page *page, unsigned int account_size)
79 if (TestClearPageDirty(page)) {
80 struct address_space *mapping = page->mapping;
81 if (mapping && mapping_cap_account_dirty(mapping)) {
82 dec_zone_page_state(page, NR_FILE_DIRTY);
83 dec_bdi_stat(mapping->backing_dev_info,
86 task_io_account_cancelled_write(account_size);
90 EXPORT_SYMBOL(cancel_dirty_page);
93 * If truncate cannot remove the fs-private metadata from the page, the page
94 * becomes orphaned. It will be left on the LRU and may even be mapped into
95 * user pagetables if we're racing with filemap_fault().
97 * We need to bale out if page->mapping is no longer equal to the original
98 * mapping. This happens a) when the VM reclaimed the page while we waited on
99 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
100 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
103 truncate_complete_page(struct address_space *mapping, struct page *page)
105 if (page->mapping != mapping)
108 if (page_has_private(page))
109 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
111 cancel_dirty_page(page, PAGE_CACHE_SIZE);
113 ClearPageMappedToDisk(page);
114 delete_from_page_cache(page);
119 * This is for invalidate_mapping_pages(). That function can be called at
120 * any time, and is not supposed to throw away dirty pages. But pages can
121 * be marked dirty at any time too, so use remove_mapping which safely
122 * discards clean, unused pages.
124 * Returns non-zero if the page was successfully invalidated.
127 invalidate_complete_page(struct address_space *mapping, struct page *page)
131 if (page->mapping != mapping)
134 if (page_has_private(page) && !try_to_release_page(page, 0))
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_range - 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_invalidate_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_invalidate_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 spin_lock_irq(&mapping->tree_lock);
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 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
585 * @lstart: offset of beginning of hole
586 * @lend: offset of last byte of hole
588 * This function should typically be called before the filesystem
589 * releases resources associated with the freed range (eg. deallocates
590 * blocks). This way, pagecache will always stay logically coherent
591 * with on-disk format, and the filesystem would not have to deal with
592 * situations such as writepage being called for a page that has already
593 * had its underlying blocks deallocated.
595 void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
597 struct address_space *mapping = inode->i_mapping;
598 loff_t unmap_start = round_up(lstart, PAGE_SIZE);
599 loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
601 * This rounding is currently just for example: unmap_mapping_range
602 * expands its hole outwards, whereas we want it to contract the hole
603 * inwards. However, existing callers of truncate_pagecache_range are
604 * doing their own page rounding first; and truncate_inode_pages_range
605 * currently BUGs if lend is not pagealigned-1 (it handles partial
606 * page at start of hole, but not partial page at end of hole). Note
607 * unmap_mapping_range allows holelen 0 for all, and we allow lend -1.
611 * Unlike in truncate_pagecache, unmap_mapping_range is called only
612 * once (before truncating pagecache), and without "even_cows" flag:
613 * hole-punching should not remove private COWed pages from the hole.
615 if ((u64)unmap_end > (u64)unmap_start)
616 unmap_mapping_range(mapping, unmap_start,
617 1 + unmap_end - unmap_start, 0);
618 truncate_inode_pages_range(mapping, lstart, lend);
620 EXPORT_SYMBOL(truncate_pagecache_range);