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>
13 #include <linux/swap.h>
14 #include <linux/module.h>
15 #include <linux/pagemap.h>
16 #include <linux/highmem.h>
17 #include <linux/pagevec.h>
18 #include <linux/task_io_accounting_ops.h>
19 #include <linux/buffer_head.h> /* grr. try_to_release_page,
25 * do_invalidatepage - invalidate part or all of a page
26 * @page: the page which is affected
27 * @offset: the index of the truncation point
29 * do_invalidatepage() is called when all or part of the page has become
30 * invalidated by a truncate operation.
32 * do_invalidatepage() does not have to release all buffers, but it must
33 * ensure that no dirty buffer is left outside @offset and that no I/O
34 * is underway against any of the blocks which are outside the truncation
35 * point. Because the caller is about to free (and possibly reuse) those
38 void do_invalidatepage(struct page *page, unsigned long offset)
40 void (*invalidatepage)(struct page *, unsigned long);
41 invalidatepage = page->mapping->a_ops->invalidatepage;
44 invalidatepage = block_invalidatepage;
47 (*invalidatepage)(page, offset);
50 static inline void truncate_partial_page(struct page *page, unsigned partial)
52 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
53 if (page_has_private(page))
54 do_invalidatepage(page, partial);
58 * This cancels just the dirty bit on the kernel page itself, it
59 * does NOT actually remove dirty bits on any mmap's that may be
60 * around. It also leaves the page tagged dirty, so any sync
61 * activity will still find it on the dirty lists, and in particular,
62 * clear_page_dirty_for_io() will still look at the dirty bits in
65 * Doing this should *normally* only ever be done when a page
66 * is truncated, and is not actually mapped anywhere at all. However,
67 * fs/buffer.c does this when it notices that somebody has cleaned
68 * out all the buffers on a page without actually doing it through
69 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
71 void cancel_dirty_page(struct page *page, unsigned int account_size)
73 if (TestClearPageDirty(page)) {
74 struct address_space *mapping = page->mapping;
75 if (mapping && mapping_cap_account_dirty(mapping)) {
76 dec_zone_page_state(page, NR_FILE_DIRTY);
77 dec_bdi_stat(mapping->backing_dev_info,
80 task_io_account_cancelled_write(account_size);
84 EXPORT_SYMBOL(cancel_dirty_page);
87 * If truncate cannot remove the fs-private metadata from the page, the page
88 * becomes orphaned. It will be left on the LRU and may even be mapped into
89 * user pagetables if we're racing with filemap_fault().
91 * We need to bale out if page->mapping is no longer equal to the original
92 * mapping. This happens a) when the VM reclaimed the page while we waited on
93 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
94 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
97 truncate_complete_page(struct address_space *mapping, struct page *page)
99 if (page->mapping != mapping)
102 if (page_has_private(page))
103 do_invalidatepage(page, 0);
105 cancel_dirty_page(page, PAGE_CACHE_SIZE);
107 clear_page_mlock(page);
108 remove_from_page_cache(page);
109 ClearPageMappedToDisk(page);
110 page_cache_release(page); /* pagecache ref */
115 * This is for invalidate_mapping_pages(). That function can be called at
116 * any time, and is not supposed to throw away dirty pages. But pages can
117 * be marked dirty at any time too, so use remove_mapping which safely
118 * discards clean, unused pages.
120 * Returns non-zero if the page was successfully invalidated.
123 invalidate_complete_page(struct address_space *mapping, struct page *page)
127 if (page->mapping != mapping)
130 if (page_has_private(page) && !try_to_release_page(page, 0))
133 clear_page_mlock(page);
134 ret = remove_mapping(mapping, page);
139 int truncate_inode_page(struct address_space *mapping, struct page *page)
141 if (page_mapped(page)) {
142 unmap_mapping_range(mapping,
143 (loff_t)page->index << PAGE_CACHE_SHIFT,
146 return truncate_complete_page(mapping, page);
150 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
151 * @mapping: mapping to truncate
152 * @lstart: offset from which to truncate
153 * @lend: offset to which to truncate
155 * Truncate the page cache, removing the pages that are between
156 * specified offsets (and zeroing out partial page
157 * (if lstart is not page aligned)).
159 * Truncate takes two passes - the first pass is nonblocking. It will not
160 * block on page locks and it will not block on writeback. The second pass
161 * will wait. This is to prevent as much IO as possible in the affected region.
162 * The first pass will remove most pages, so the search cost of the second pass
165 * When looking at page->index outside the page lock we need to be careful to
166 * copy it into a local to avoid races (it could change at any time).
168 * We pass down the cache-hot hint to the page freeing code. Even if the
169 * mapping is large, it is probably the case that the final pages are the most
170 * recently touched, and freeing happens in ascending file offset order.
172 void truncate_inode_pages_range(struct address_space *mapping,
173 loff_t lstart, loff_t lend)
175 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
177 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
182 if (mapping->nrpages == 0)
185 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
186 end = (lend >> PAGE_CACHE_SHIFT);
188 pagevec_init(&pvec, 0);
190 while (next <= end &&
191 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
192 for (i = 0; i < pagevec_count(&pvec); i++) {
193 struct page *page = pvec.pages[i];
194 pgoff_t page_index = page->index;
196 if (page_index > end) {
201 if (page_index > next)
204 if (!trylock_page(page))
206 if (PageWriteback(page)) {
210 truncate_inode_page(mapping, page);
213 pagevec_release(&pvec);
218 struct page *page = find_lock_page(mapping, start - 1);
220 wait_on_page_writeback(page);
221 truncate_partial_page(page, partial);
223 page_cache_release(page);
230 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
236 if (pvec.pages[0]->index > end) {
237 pagevec_release(&pvec);
240 for (i = 0; i < pagevec_count(&pvec); i++) {
241 struct page *page = pvec.pages[i];
243 if (page->index > end)
246 wait_on_page_writeback(page);
247 truncate_inode_page(mapping, page);
248 if (page->index > next)
253 pagevec_release(&pvec);
256 EXPORT_SYMBOL(truncate_inode_pages_range);
259 * truncate_inode_pages - truncate *all* the pages from an offset
260 * @mapping: mapping to truncate
261 * @lstart: offset from which to truncate
263 * Called under (and serialised by) inode->i_mutex.
265 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
267 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
269 EXPORT_SYMBOL(truncate_inode_pages);
272 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
273 * @mapping: the address_space which holds the pages to invalidate
274 * @start: the offset 'from' which to invalidate
275 * @end: the offset 'to' which to invalidate (inclusive)
277 * This function only removes the unlocked pages, if you want to
278 * remove all the pages of one inode, you must call truncate_inode_pages.
280 * invalidate_mapping_pages() will not block on IO activity. It will not
281 * invalidate pages which are dirty, locked, under writeback or mapped into
284 unsigned long invalidate_mapping_pages(struct address_space *mapping,
285 pgoff_t start, pgoff_t end)
288 pgoff_t next = start;
289 unsigned long ret = 0;
292 pagevec_init(&pvec, 0);
293 while (next <= end &&
294 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
295 for (i = 0; i < pagevec_count(&pvec); i++) {
296 struct page *page = pvec.pages[i];
300 lock_failed = !trylock_page(page);
303 * We really shouldn't be looking at the ->index of an
304 * unlocked page. But we're not allowed to lock these
305 * pages. So we rely upon nobody altering the ->index
306 * of this (pinned-by-us) page.
315 if (PageDirty(page) || PageWriteback(page))
317 if (page_mapped(page))
319 ret += invalidate_complete_page(mapping, page);
325 pagevec_release(&pvec);
330 EXPORT_SYMBOL(invalidate_mapping_pages);
333 * This is like invalidate_complete_page(), except it ignores the page's
334 * refcount. We do this because invalidate_inode_pages2() needs stronger
335 * invalidation guarantees, and cannot afford to leave pages behind because
336 * shrink_page_list() has a temp ref on them, or because they're transiently
337 * sitting in the lru_cache_add() pagevecs.
340 invalidate_complete_page2(struct address_space *mapping, struct page *page)
342 if (page->mapping != mapping)
345 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
348 spin_lock_irq(&mapping->tree_lock);
352 clear_page_mlock(page);
353 BUG_ON(page_has_private(page));
354 __remove_from_page_cache(page);
355 spin_unlock_irq(&mapping->tree_lock);
356 mem_cgroup_uncharge_cache_page(page);
357 page_cache_release(page); /* pagecache ref */
360 spin_unlock_irq(&mapping->tree_lock);
364 static int do_launder_page(struct address_space *mapping, struct page *page)
366 if (!PageDirty(page))
368 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
370 return mapping->a_ops->launder_page(page);
374 * invalidate_inode_pages2_range - remove range of pages from an address_space
375 * @mapping: the address_space
376 * @start: the page offset 'from' which to invalidate
377 * @end: the page offset 'to' which to invalidate (inclusive)
379 * Any pages which are found to be mapped into pagetables are unmapped prior to
382 * Returns -EBUSY if any pages could not be invalidated.
384 int invalidate_inode_pages2_range(struct address_space *mapping,
385 pgoff_t start, pgoff_t end)
392 int did_range_unmap = 0;
395 pagevec_init(&pvec, 0);
397 while (next <= end && !wrapped &&
398 pagevec_lookup(&pvec, mapping, next,
399 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
400 for (i = 0; i < pagevec_count(&pvec); i++) {
401 struct page *page = pvec.pages[i];
405 if (page->mapping != mapping) {
409 page_index = page->index;
410 next = page_index + 1;
413 if (page_index > end) {
417 wait_on_page_writeback(page);
418 if (page_mapped(page)) {
419 if (!did_range_unmap) {
421 * Zap the rest of the file in one hit.
423 unmap_mapping_range(mapping,
424 (loff_t)page_index<<PAGE_CACHE_SHIFT,
425 (loff_t)(end - page_index + 1)
433 unmap_mapping_range(mapping,
434 (loff_t)page_index<<PAGE_CACHE_SHIFT,
438 BUG_ON(page_mapped(page));
439 ret2 = do_launder_page(mapping, page);
441 if (!invalidate_complete_page2(mapping, page))
448 pagevec_release(&pvec);
453 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
456 * invalidate_inode_pages2 - remove all pages from an address_space
457 * @mapping: the address_space
459 * Any pages which are found to be mapped into pagetables are unmapped prior to
462 * Returns -EIO if any pages could not be invalidated.
464 int invalidate_inode_pages2(struct address_space *mapping)
466 return invalidate_inode_pages2_range(mapping, 0, -1);
468 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);