2 * mm/truncate.c - code for taking down pages from address_spaces
4 * Copyright (C) 2002, Linus Torvalds
6 * 10Sep2002 akpm@zip.com.au
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/buffer_head.h> /* grr. try_to_release_page,
21 * do_invalidatepage - invalidate part of all of a page
22 * @page: the page which is affected
23 * @offset: the index of the truncation point
25 * do_invalidatepage() is called when all or part of the page has become
26 * invalidated by a truncate operation.
28 * do_invalidatepage() does not have to release all buffers, but it must
29 * ensure that no dirty buffer is left outside @offset and that no I/O
30 * is underway against any of the blocks which are outside the truncation
31 * point. Because the caller is about to free (and possibly reuse) those
34 void do_invalidatepage(struct page *page, unsigned long offset)
36 void (*invalidatepage)(struct page *, unsigned long);
37 invalidatepage = page->mapping->a_ops->invalidatepage;
40 invalidatepage = block_invalidatepage;
43 (*invalidatepage)(page, offset);
46 static inline void truncate_partial_page(struct page *page, unsigned partial)
48 memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
49 if (PagePrivate(page))
50 do_invalidatepage(page, partial);
54 * If truncate cannot remove the fs-private metadata from the page, the page
55 * becomes anonymous. It will be left on the LRU and may even be mapped into
56 * user pagetables if we're racing with filemap_nopage().
58 * We need to bale out if page->mapping is no longer equal to the original
59 * mapping. This happens a) when the VM reclaimed the page while we waited on
60 * its lock, b) when a concurrent invalidate_inode_pages got there first and
61 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
64 truncate_complete_page(struct address_space *mapping, struct page *page)
66 if (page->mapping != mapping)
69 if (PagePrivate(page))
70 do_invalidatepage(page, 0);
72 clear_page_dirty(page);
73 ClearPageUptodate(page);
74 ClearPageMappedToDisk(page);
75 remove_from_page_cache(page);
76 page_cache_release(page); /* pagecache ref */
80 * This is for invalidate_inode_pages(). That function can be called at
81 * any time, and is not supposed to throw away dirty pages. But pages can
82 * be marked dirty at any time too, so use remove_mapping which safely
83 * discards clean, unused pages.
85 * Returns non-zero if the page was successfully invalidated.
88 invalidate_complete_page(struct address_space *mapping, struct page *page)
92 if (page->mapping != mapping)
95 if (PagePrivate(page) && !try_to_release_page(page, 0))
98 ret = remove_mapping(mapping, page);
99 ClearPageUptodate(page);
105 * truncate_inode_pages - truncate range of pages specified by start and
107 * @mapping: mapping to truncate
108 * @lstart: offset from which to truncate
109 * @lend: offset to which to truncate
111 * Truncate the page cache, removing the pages that are between
112 * specified offsets (and zeroing out partial page
113 * (if lstart is not page aligned)).
115 * Truncate takes two passes - the first pass is nonblocking. It will not
116 * block on page locks and it will not block on writeback. The second pass
117 * will wait. This is to prevent as much IO as possible in the affected region.
118 * The first pass will remove most pages, so the search cost of the second pass
121 * When looking at page->index outside the page lock we need to be careful to
122 * copy it into a local to avoid races (it could change at any time).
124 * We pass down the cache-hot hint to the page freeing code. Even if the
125 * mapping is large, it is probably the case that the final pages are the most
126 * recently touched, and freeing happens in ascending file offset order.
128 void truncate_inode_pages_range(struct address_space *mapping,
129 loff_t lstart, loff_t lend)
131 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
133 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
138 if (mapping->nrpages == 0)
141 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
142 end = (lend >> PAGE_CACHE_SHIFT);
144 pagevec_init(&pvec, 0);
146 while (next <= end &&
147 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
148 for (i = 0; i < pagevec_count(&pvec); i++) {
149 struct page *page = pvec.pages[i];
150 pgoff_t page_index = page->index;
152 if (page_index > end) {
157 if (page_index > next)
160 if (TestSetPageLocked(page))
162 if (PageWriteback(page)) {
166 truncate_complete_page(mapping, page);
169 pagevec_release(&pvec);
174 struct page *page = find_lock_page(mapping, start - 1);
176 wait_on_page_writeback(page);
177 truncate_partial_page(page, partial);
179 page_cache_release(page);
186 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
192 if (pvec.pages[0]->index > end) {
193 pagevec_release(&pvec);
196 for (i = 0; i < pagevec_count(&pvec); i++) {
197 struct page *page = pvec.pages[i];
199 if (page->index > end)
202 wait_on_page_writeback(page);
203 if (page->index > next)
206 truncate_complete_page(mapping, page);
209 pagevec_release(&pvec);
212 EXPORT_SYMBOL(truncate_inode_pages_range);
215 * truncate_inode_pages - truncate *all* the pages from an offset
216 * @mapping: mapping to truncate
217 * @lstart: offset from which to truncate
219 * Called under (and serialised by) inode->i_mutex.
221 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
223 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
225 EXPORT_SYMBOL(truncate_inode_pages);
228 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
229 * @mapping: the address_space which holds the pages to invalidate
230 * @start: the offset 'from' which to invalidate
231 * @end: the offset 'to' which to invalidate (inclusive)
233 * This function only removes the unlocked pages, if you want to
234 * remove all the pages of one inode, you must call truncate_inode_pages.
236 * invalidate_mapping_pages() will not block on IO activity. It will not
237 * invalidate pages which are dirty, locked, under writeback or mapped into
240 unsigned long invalidate_mapping_pages(struct address_space *mapping,
241 pgoff_t start, pgoff_t end)
244 pgoff_t next = start;
245 unsigned long ret = 0;
248 pagevec_init(&pvec, 0);
249 while (next <= end &&
250 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
251 for (i = 0; i < pagevec_count(&pvec); i++) {
252 struct page *page = pvec.pages[i];
256 lock_failed = TestSetPageLocked(page);
259 * We really shouldn't be looking at the ->index of an
260 * unlocked page. But we're not allowed to lock these
261 * pages. So we rely upon nobody altering the ->index
262 * of this (pinned-by-us) page.
271 if (PageDirty(page) || PageWriteback(page))
273 if (page_mapped(page))
275 ret += invalidate_complete_page(mapping, page);
281 pagevec_release(&pvec);
286 unsigned long invalidate_inode_pages(struct address_space *mapping)
288 return invalidate_mapping_pages(mapping, 0, ~0UL);
290 EXPORT_SYMBOL(invalidate_inode_pages);
293 * This is like invalidate_complete_page(), except it ignores the page's
294 * refcount. We do this because invalidate_inode_pages2() needs stronger
295 * invalidation guarantees, and cannot afford to leave pages behind because
296 * shrink_list() has a temp ref on them, or because they're transiently sitting
297 * in the lru_cache_add() pagevecs.
300 invalidate_complete_page2(struct address_space *mapping, struct page *page)
302 if (page->mapping != mapping)
305 if (PagePrivate(page) && !try_to_release_page(page, 0))
308 write_lock_irq(&mapping->tree_lock);
312 BUG_ON(PagePrivate(page));
313 __remove_from_page_cache(page);
314 write_unlock_irq(&mapping->tree_lock);
315 ClearPageUptodate(page);
316 page_cache_release(page); /* pagecache ref */
319 write_unlock_irq(&mapping->tree_lock);
324 * invalidate_inode_pages2_range - remove range of pages from an address_space
325 * @mapping: the address_space
326 * @start: the page offset 'from' which to invalidate
327 * @end: the page offset 'to' which to invalidate (inclusive)
329 * Any pages which are found to be mapped into pagetables are unmapped prior to
332 * Returns -EIO if any pages could not be invalidated.
334 int invalidate_inode_pages2_range(struct address_space *mapping,
335 pgoff_t start, pgoff_t end)
341 int did_range_unmap = 0;
344 pagevec_init(&pvec, 0);
346 while (next <= end && !ret && !wrapped &&
347 pagevec_lookup(&pvec, mapping, next,
348 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
349 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
350 struct page *page = pvec.pages[i];
355 if (page->mapping != mapping) {
359 page_index = page->index;
360 next = page_index + 1;
363 if (page_index > end) {
367 wait_on_page_writeback(page);
368 while (page_mapped(page)) {
369 if (!did_range_unmap) {
371 * Zap the rest of the file in one hit.
373 unmap_mapping_range(mapping,
374 (loff_t)page_index<<PAGE_CACHE_SHIFT,
375 (loff_t)(end - page_index + 1)
383 unmap_mapping_range(mapping,
384 (loff_t)page_index<<PAGE_CACHE_SHIFT,
388 was_dirty = test_clear_page_dirty(page);
389 if (!invalidate_complete_page2(mapping, page)) {
391 set_page_dirty(page);
396 pagevec_release(&pvec);
402 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
405 * invalidate_inode_pages2 - remove all pages from an address_space
406 * @mapping: the address_space
408 * Any pages which are found to be mapped into pagetables are unmapped prior to
411 * Returns -EIO if any pages could not be invalidated.
413 int invalidate_inode_pages2(struct address_space *mapping)
415 return invalidate_inode_pages2_range(mapping, 0, -1);
417 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);