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);
104 * truncate_inode_pages - truncate range of pages specified by start and
106 * @mapping: mapping to truncate
107 * @lstart: offset from which to truncate
108 * @lend: offset to which to truncate
110 * Truncate the page cache, removing the pages that are between
111 * specified offsets (and zeroing out partial page
112 * (if lstart is not page aligned)).
114 * Truncate takes two passes - the first pass is nonblocking. It will not
115 * block on page locks and it will not block on writeback. The second pass
116 * will wait. This is to prevent as much IO as possible in the affected region.
117 * The first pass will remove most pages, so the search cost of the second pass
120 * When looking at page->index outside the page lock we need to be careful to
121 * copy it into a local to avoid races (it could change at any time).
123 * We pass down the cache-hot hint to the page freeing code. Even if the
124 * mapping is large, it is probably the case that the final pages are the most
125 * recently touched, and freeing happens in ascending file offset order.
127 void truncate_inode_pages_range(struct address_space *mapping,
128 loff_t lstart, loff_t lend)
130 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
132 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
137 if (mapping->nrpages == 0)
140 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
141 end = (lend >> PAGE_CACHE_SHIFT);
143 pagevec_init(&pvec, 0);
145 while (next <= end &&
146 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
147 for (i = 0; i < pagevec_count(&pvec); i++) {
148 struct page *page = pvec.pages[i];
149 pgoff_t page_index = page->index;
151 if (page_index > end) {
156 if (page_index > next)
159 if (TestSetPageLocked(page))
161 if (PageWriteback(page)) {
165 truncate_complete_page(mapping, page);
168 pagevec_release(&pvec);
173 struct page *page = find_lock_page(mapping, start - 1);
175 wait_on_page_writeback(page);
176 truncate_partial_page(page, partial);
178 page_cache_release(page);
185 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
191 if (pvec.pages[0]->index > end) {
192 pagevec_release(&pvec);
195 for (i = 0; i < pagevec_count(&pvec); i++) {
196 struct page *page = pvec.pages[i];
198 if (page->index > end)
201 wait_on_page_writeback(page);
202 if (page->index > next)
205 truncate_complete_page(mapping, page);
208 pagevec_release(&pvec);
211 EXPORT_SYMBOL(truncate_inode_pages_range);
214 * truncate_inode_pages - truncate *all* the pages from an offset
215 * @mapping: mapping to truncate
216 * @lstart: offset from which to truncate
218 * Called under (and serialised by) inode->i_mutex.
220 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
222 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
224 EXPORT_SYMBOL(truncate_inode_pages);
227 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
228 * @mapping: the address_space which holds the pages to invalidate
229 * @start: the offset 'from' which to invalidate
230 * @end: the offset 'to' which to invalidate (inclusive)
232 * This function only removes the unlocked pages, if you want to
233 * remove all the pages of one inode, you must call truncate_inode_pages.
235 * invalidate_mapping_pages() will not block on IO activity. It will not
236 * invalidate pages which are dirty, locked, under writeback or mapped into
239 unsigned long invalidate_mapping_pages(struct address_space *mapping,
240 pgoff_t start, pgoff_t end)
243 pgoff_t next = start;
244 unsigned long ret = 0;
247 pagevec_init(&pvec, 0);
248 while (next <= end &&
249 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
250 for (i = 0; i < pagevec_count(&pvec); i++) {
251 struct page *page = pvec.pages[i];
255 lock_failed = TestSetPageLocked(page);
258 * We really shouldn't be looking at the ->index of an
259 * unlocked page. But we're not allowed to lock these
260 * pages. So we rely upon nobody altering the ->index
261 * of this (pinned-by-us) page.
270 if (PageDirty(page) || PageWriteback(page))
272 if (page_mapped(page))
274 ret += invalidate_complete_page(mapping, page);
280 pagevec_release(&pvec);
285 unsigned long invalidate_inode_pages(struct address_space *mapping)
287 return invalidate_mapping_pages(mapping, 0, ~0UL);
289 EXPORT_SYMBOL(invalidate_inode_pages);
292 * This is like invalidate_complete_page(), except it ignores the page's
293 * refcount. We do this because invalidate_inode_pages2() needs stronger
294 * invalidation guarantees, and cannot afford to leave pages behind because
295 * shrink_list() has a temp ref on them, or because they're transiently sitting
296 * in the lru_cache_add() pagevecs.
299 invalidate_complete_page2(struct address_space *mapping, struct page *page)
301 if (page->mapping != mapping)
304 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
307 write_lock_irq(&mapping->tree_lock);
311 BUG_ON(PagePrivate(page));
312 __remove_from_page_cache(page);
313 write_unlock_irq(&mapping->tree_lock);
314 ClearPageUptodate(page);
315 page_cache_release(page); /* pagecache ref */
318 write_unlock_irq(&mapping->tree_lock);
323 * invalidate_inode_pages2_range - remove range of pages from an address_space
324 * @mapping: the address_space
325 * @start: the page offset 'from' which to invalidate
326 * @end: the page offset 'to' which to invalidate (inclusive)
328 * Any pages which are found to be mapped into pagetables are unmapped prior to
331 * Returns -EIO if any pages could not be invalidated.
333 int invalidate_inode_pages2_range(struct address_space *mapping,
334 pgoff_t start, pgoff_t end)
340 int did_range_unmap = 0;
343 pagevec_init(&pvec, 0);
345 while (next <= end && !ret && !wrapped &&
346 pagevec_lookup(&pvec, mapping, next,
347 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
348 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
349 struct page *page = pvec.pages[i];
354 if (page->mapping != mapping) {
358 page_index = page->index;
359 next = page_index + 1;
362 if (page_index > end) {
366 wait_on_page_writeback(page);
367 while (page_mapped(page)) {
368 if (!did_range_unmap) {
370 * Zap the rest of the file in one hit.
372 unmap_mapping_range(mapping,
373 (loff_t)page_index<<PAGE_CACHE_SHIFT,
374 (loff_t)(end - page_index + 1)
382 unmap_mapping_range(mapping,
383 (loff_t)page_index<<PAGE_CACHE_SHIFT,
387 was_dirty = test_clear_page_dirty(page);
388 if (!invalidate_complete_page2(mapping, page)) {
390 set_page_dirty(page);
395 pagevec_release(&pvec);
401 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
404 * invalidate_inode_pages2 - remove all pages from an address_space
405 * @mapping: the address_space
407 * Any pages which are found to be mapped into pagetables are unmapped prior to
410 * Returns -EIO if any pages could not be invalidated.
412 int invalidate_inode_pages2(struct address_space *mapping)
414 return invalidate_inode_pages2_range(mapping, 0, -1);
416 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);