2 * mm/truncate.c - code for taking down pages from address_spaces
4 * Copyright (C) 2002, Linus Torvalds
10 #include <linux/kernel.h>
12 #include <linux/swap.h>
13 #include <linux/module.h>
14 #include <linux/pagemap.h>
15 #include <linux/pagevec.h>
16 #include <linux/task_io_accounting_ops.h>
17 #include <linux/buffer_head.h> /* grr. try_to_release_page,
22 * do_invalidatepage - invalidate part of all of a page
23 * @page: the page which is affected
24 * @offset: the index of the truncation point
26 * do_invalidatepage() is called when all or part of the page has become
27 * invalidated by a truncate operation.
29 * do_invalidatepage() does not have to release all buffers, but it must
30 * ensure that no dirty buffer is left outside @offset and that no I/O
31 * is underway against any of the blocks which are outside the truncation
32 * point. Because the caller is about to free (and possibly reuse) those
35 void do_invalidatepage(struct page *page, unsigned long offset)
37 void (*invalidatepage)(struct page *, unsigned long);
38 invalidatepage = page->mapping->a_ops->invalidatepage;
41 invalidatepage = block_invalidatepage;
44 (*invalidatepage)(page, offset);
47 static inline void truncate_partial_page(struct page *page, unsigned partial)
49 memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
50 if (PagePrivate(page))
51 do_invalidatepage(page, partial);
54 void cancel_dirty_page(struct page *page, unsigned int account_size)
56 /* If we're cancelling the page, it had better not be mapped any more */
57 if (page_mapped(page)) {
58 static unsigned int warncount;
60 WARN_ON(++warncount < 5);
63 if (TestClearPageDirty(page) && account_size) {
64 dec_zone_page_state(page, NR_FILE_DIRTY);
65 task_io_account_cancelled_write(account_size);
70 * If truncate cannot remove the fs-private metadata from the page, the page
71 * becomes anonymous. It will be left on the LRU and may even be mapped into
72 * user pagetables if we're racing with filemap_nopage().
74 * We need to bale out if page->mapping is no longer equal to the original
75 * mapping. This happens a) when the VM reclaimed the page while we waited on
76 * its lock, b) when a concurrent invalidate_inode_pages got there first and
77 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
80 truncate_complete_page(struct address_space *mapping, struct page *page)
82 if (page->mapping != mapping)
85 cancel_dirty_page(page, PAGE_CACHE_SIZE);
87 if (PagePrivate(page))
88 do_invalidatepage(page, 0);
90 ClearPageUptodate(page);
91 ClearPageMappedToDisk(page);
92 remove_from_page_cache(page);
93 page_cache_release(page); /* pagecache ref */
97 * This is for invalidate_inode_pages(). That function can be called at
98 * any time, and is not supposed to throw away dirty pages. But pages can
99 * be marked dirty at any time too, so use remove_mapping which safely
100 * discards clean, unused pages.
102 * Returns non-zero if the page was successfully invalidated.
105 invalidate_complete_page(struct address_space *mapping, struct page *page)
109 if (page->mapping != mapping)
112 if (PagePrivate(page) && !try_to_release_page(page, 0))
115 ret = remove_mapping(mapping, page);
121 * truncate_inode_pages - truncate range of pages specified by start and
123 * @mapping: mapping to truncate
124 * @lstart: offset from which to truncate
125 * @lend: offset to which to truncate
127 * Truncate the page cache, removing the pages that are between
128 * specified offsets (and zeroing out partial page
129 * (if lstart is not page aligned)).
131 * Truncate takes two passes - the first pass is nonblocking. It will not
132 * block on page locks and it will not block on writeback. The second pass
133 * will wait. This is to prevent as much IO as possible in the affected region.
134 * The first pass will remove most pages, so the search cost of the second pass
137 * When looking at page->index outside the page lock we need to be careful to
138 * copy it into a local to avoid races (it could change at any time).
140 * We pass down the cache-hot hint to the page freeing code. Even if the
141 * mapping is large, it is probably the case that the final pages are the most
142 * recently touched, and freeing happens in ascending file offset order.
144 void truncate_inode_pages_range(struct address_space *mapping,
145 loff_t lstart, loff_t lend)
147 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
149 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
154 if (mapping->nrpages == 0)
157 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
158 end = (lend >> PAGE_CACHE_SHIFT);
160 pagevec_init(&pvec, 0);
162 while (next <= end &&
163 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
164 for (i = 0; i < pagevec_count(&pvec); i++) {
165 struct page *page = pvec.pages[i];
166 pgoff_t page_index = page->index;
168 if (page_index > end) {
173 if (page_index > next)
176 if (TestSetPageLocked(page))
178 if (PageWriteback(page)) {
182 truncate_complete_page(mapping, page);
185 pagevec_release(&pvec);
190 struct page *page = find_lock_page(mapping, start - 1);
192 wait_on_page_writeback(page);
193 truncate_partial_page(page, partial);
195 page_cache_release(page);
202 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
208 if (pvec.pages[0]->index > end) {
209 pagevec_release(&pvec);
212 for (i = 0; i < pagevec_count(&pvec); i++) {
213 struct page *page = pvec.pages[i];
215 if (page->index > end)
218 wait_on_page_writeback(page);
219 if (page->index > next)
222 truncate_complete_page(mapping, page);
225 pagevec_release(&pvec);
228 EXPORT_SYMBOL(truncate_inode_pages_range);
231 * truncate_inode_pages - truncate *all* the pages from an offset
232 * @mapping: mapping to truncate
233 * @lstart: offset from which to truncate
235 * Called under (and serialised by) inode->i_mutex.
237 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
239 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
241 EXPORT_SYMBOL(truncate_inode_pages);
244 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
245 * @mapping: the address_space which holds the pages to invalidate
246 * @start: the offset 'from' which to invalidate
247 * @end: the offset 'to' which to invalidate (inclusive)
249 * This function only removes the unlocked pages, if you want to
250 * remove all the pages of one inode, you must call truncate_inode_pages.
252 * invalidate_mapping_pages() will not block on IO activity. It will not
253 * invalidate pages which are dirty, locked, under writeback or mapped into
256 unsigned long invalidate_mapping_pages(struct address_space *mapping,
257 pgoff_t start, pgoff_t end)
260 pgoff_t next = start;
261 unsigned long ret = 0;
264 pagevec_init(&pvec, 0);
265 while (next <= end &&
266 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
267 for (i = 0; i < pagevec_count(&pvec); i++) {
268 struct page *page = pvec.pages[i];
272 lock_failed = TestSetPageLocked(page);
275 * We really shouldn't be looking at the ->index of an
276 * unlocked page. But we're not allowed to lock these
277 * pages. So we rely upon nobody altering the ->index
278 * of this (pinned-by-us) page.
287 if (PageDirty(page) || PageWriteback(page))
289 if (page_mapped(page))
291 ret += invalidate_complete_page(mapping, page);
297 pagevec_release(&pvec);
302 unsigned long invalidate_inode_pages(struct address_space *mapping)
304 return invalidate_mapping_pages(mapping, 0, ~0UL);
306 EXPORT_SYMBOL(invalidate_inode_pages);
309 * This is like invalidate_complete_page(), except it ignores the page's
310 * refcount. We do this because invalidate_inode_pages2() needs stronger
311 * invalidation guarantees, and cannot afford to leave pages behind because
312 * shrink_list() has a temp ref on them, or because they're transiently sitting
313 * in the lru_cache_add() pagevecs.
316 invalidate_complete_page2(struct address_space *mapping, struct page *page)
318 if (page->mapping != mapping)
321 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
324 write_lock_irq(&mapping->tree_lock);
328 BUG_ON(PagePrivate(page));
329 __remove_from_page_cache(page);
330 write_unlock_irq(&mapping->tree_lock);
331 ClearPageUptodate(page);
332 page_cache_release(page); /* pagecache ref */
335 write_unlock_irq(&mapping->tree_lock);
340 * invalidate_inode_pages2_range - remove range of pages from an address_space
341 * @mapping: the address_space
342 * @start: the page offset 'from' which to invalidate
343 * @end: the page offset 'to' which to invalidate (inclusive)
345 * Any pages which are found to be mapped into pagetables are unmapped prior to
348 * Returns -EIO if any pages could not be invalidated.
350 int invalidate_inode_pages2_range(struct address_space *mapping,
351 pgoff_t start, pgoff_t end)
357 int did_range_unmap = 0;
360 pagevec_init(&pvec, 0);
362 while (next <= end && !ret && !wrapped &&
363 pagevec_lookup(&pvec, mapping, next,
364 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
365 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
366 struct page *page = pvec.pages[i];
370 if (page->mapping != mapping) {
374 page_index = page->index;
375 next = page_index + 1;
378 if (page_index > end) {
382 wait_on_page_writeback(page);
383 while (page_mapped(page)) {
384 if (!did_range_unmap) {
386 * Zap the rest of the file in one hit.
388 unmap_mapping_range(mapping,
389 (loff_t)page_index<<PAGE_CACHE_SHIFT,
390 (loff_t)(end - page_index + 1)
398 unmap_mapping_range(mapping,
399 (loff_t)page_index<<PAGE_CACHE_SHIFT,
403 if (!invalidate_complete_page2(mapping, page))
407 pagevec_release(&pvec);
413 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
416 * invalidate_inode_pages2 - remove all pages from an address_space
417 * @mapping: the address_space
419 * Any pages which are found to be mapped into pagetables are unmapped prior to
422 * Returns -EIO if any pages could not be invalidated.
424 int invalidate_inode_pages2(struct address_space *mapping)
426 return invalidate_inode_pages2_range(mapping, 0, -1);
428 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);