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/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 task_io_account_cancelled_write(account_size);
69 * If truncate cannot remove the fs-private metadata from the page, the page
70 * becomes anonymous. It will be left on the LRU and may even be mapped into
71 * user pagetables if we're racing with filemap_nopage().
73 * We need to bale out if page->mapping is no longer equal to the original
74 * mapping. This happens a) when the VM reclaimed the page while we waited on
75 * its lock, b) when a concurrent invalidate_inode_pages got there first and
76 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
79 truncate_complete_page(struct address_space *mapping, struct page *page)
81 if (page->mapping != mapping)
84 if (PagePrivate(page))
85 do_invalidatepage(page, 0);
87 cancel_dirty_page(page, PAGE_CACHE_SIZE);
89 ClearPageUptodate(page);
90 ClearPageMappedToDisk(page);
91 remove_from_page_cache(page);
92 page_cache_release(page); /* pagecache ref */
96 * This is for invalidate_inode_pages(). That function can be called at
97 * any time, and is not supposed to throw away dirty pages. But pages can
98 * be marked dirty at any time too, so use remove_mapping which safely
99 * discards clean, unused pages.
101 * Returns non-zero if the page was successfully invalidated.
104 invalidate_complete_page(struct address_space *mapping, struct page *page)
108 if (page->mapping != mapping)
111 if (PagePrivate(page) && !try_to_release_page(page, 0))
114 ret = remove_mapping(mapping, page);
120 * truncate_inode_pages - truncate range of pages specified by start and
122 * @mapping: mapping to truncate
123 * @lstart: offset from which to truncate
124 * @lend: offset to which to truncate
126 * Truncate the page cache, removing the pages that are between
127 * specified offsets (and zeroing out partial page
128 * (if lstart is not page aligned)).
130 * Truncate takes two passes - the first pass is nonblocking. It will not
131 * block on page locks and it will not block on writeback. The second pass
132 * will wait. This is to prevent as much IO as possible in the affected region.
133 * The first pass will remove most pages, so the search cost of the second pass
136 * When looking at page->index outside the page lock we need to be careful to
137 * copy it into a local to avoid races (it could change at any time).
139 * We pass down the cache-hot hint to the page freeing code. Even if the
140 * mapping is large, it is probably the case that the final pages are the most
141 * recently touched, and freeing happens in ascending file offset order.
143 void truncate_inode_pages_range(struct address_space *mapping,
144 loff_t lstart, loff_t lend)
146 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
148 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
153 if (mapping->nrpages == 0)
156 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
157 end = (lend >> PAGE_CACHE_SHIFT);
159 pagevec_init(&pvec, 0);
161 while (next <= end &&
162 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
163 for (i = 0; i < pagevec_count(&pvec); i++) {
164 struct page *page = pvec.pages[i];
165 pgoff_t page_index = page->index;
167 if (page_index > end) {
172 if (page_index > next)
175 if (TestSetPageLocked(page))
177 if (PageWriteback(page)) {
181 truncate_complete_page(mapping, page);
184 pagevec_release(&pvec);
189 struct page *page = find_lock_page(mapping, start - 1);
191 wait_on_page_writeback(page);
192 truncate_partial_page(page, partial);
194 page_cache_release(page);
201 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
207 if (pvec.pages[0]->index > end) {
208 pagevec_release(&pvec);
211 for (i = 0; i < pagevec_count(&pvec); i++) {
212 struct page *page = pvec.pages[i];
214 if (page->index > end)
217 wait_on_page_writeback(page);
218 if (page->index > next)
221 truncate_complete_page(mapping, page);
224 pagevec_release(&pvec);
227 EXPORT_SYMBOL(truncate_inode_pages_range);
230 * truncate_inode_pages - truncate *all* the pages from an offset
231 * @mapping: mapping to truncate
232 * @lstart: offset from which to truncate
234 * Called under (and serialised by) inode->i_mutex.
236 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
238 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
240 EXPORT_SYMBOL(truncate_inode_pages);
243 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
244 * @mapping: the address_space which holds the pages to invalidate
245 * @start: the offset 'from' which to invalidate
246 * @end: the offset 'to' which to invalidate (inclusive)
248 * This function only removes the unlocked pages, if you want to
249 * remove all the pages of one inode, you must call truncate_inode_pages.
251 * invalidate_mapping_pages() will not block on IO activity. It will not
252 * invalidate pages which are dirty, locked, under writeback or mapped into
255 unsigned long invalidate_mapping_pages(struct address_space *mapping,
256 pgoff_t start, pgoff_t end)
259 pgoff_t next = start;
260 unsigned long ret = 0;
263 pagevec_init(&pvec, 0);
264 while (next <= end &&
265 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
266 for (i = 0; i < pagevec_count(&pvec); i++) {
267 struct page *page = pvec.pages[i];
271 lock_failed = TestSetPageLocked(page);
274 * We really shouldn't be looking at the ->index of an
275 * unlocked page. But we're not allowed to lock these
276 * pages. So we rely upon nobody altering the ->index
277 * of this (pinned-by-us) page.
286 if (PageDirty(page) || PageWriteback(page))
288 if (page_mapped(page))
290 ret += invalidate_complete_page(mapping, page);
296 pagevec_release(&pvec);
301 unsigned long invalidate_inode_pages(struct address_space *mapping)
303 return invalidate_mapping_pages(mapping, 0, ~0UL);
305 EXPORT_SYMBOL(invalidate_inode_pages);
308 * This is like invalidate_complete_page(), except it ignores the page's
309 * refcount. We do this because invalidate_inode_pages2() needs stronger
310 * invalidation guarantees, and cannot afford to leave pages behind because
311 * shrink_list() has a temp ref on them, or because they're transiently sitting
312 * in the lru_cache_add() pagevecs.
315 invalidate_complete_page2(struct address_space *mapping, struct page *page)
317 if (page->mapping != mapping)
320 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
323 write_lock_irq(&mapping->tree_lock);
327 BUG_ON(PagePrivate(page));
328 __remove_from_page_cache(page);
329 write_unlock_irq(&mapping->tree_lock);
330 ClearPageUptodate(page);
331 page_cache_release(page); /* pagecache ref */
334 write_unlock_irq(&mapping->tree_lock);
339 * invalidate_inode_pages2_range - remove range of pages from an address_space
340 * @mapping: the address_space
341 * @start: the page offset 'from' which to invalidate
342 * @end: the page offset 'to' which to invalidate (inclusive)
344 * Any pages which are found to be mapped into pagetables are unmapped prior to
347 * Returns -EIO if any pages could not be invalidated.
349 int invalidate_inode_pages2_range(struct address_space *mapping,
350 pgoff_t start, pgoff_t end)
356 int did_range_unmap = 0;
359 pagevec_init(&pvec, 0);
361 while (next <= end && !ret && !wrapped &&
362 pagevec_lookup(&pvec, mapping, next,
363 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
364 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
365 struct page *page = pvec.pages[i];
369 if (page->mapping != mapping) {
373 page_index = page->index;
374 next = page_index + 1;
377 if (page_index > end) {
381 wait_on_page_writeback(page);
382 while (page_mapped(page)) {
383 if (!did_range_unmap) {
385 * Zap the rest of the file in one hit.
387 unmap_mapping_range(mapping,
388 (loff_t)page_index<<PAGE_CACHE_SHIFT,
389 (loff_t)(end - page_index + 1)
397 unmap_mapping_range(mapping,
398 (loff_t)page_index<<PAGE_CACHE_SHIFT,
402 if (!invalidate_complete_page2(mapping, page))
406 pagevec_release(&pvec);
412 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
415 * invalidate_inode_pages2 - remove all pages from an address_space
416 * @mapping: the address_space
418 * Any pages which are found to be mapped into pagetables are unmapped prior to
421 * Returns -EIO if any pages could not be invalidated.
423 int invalidate_inode_pages2(struct address_space *mapping)
425 return invalidate_inode_pages2_range(mapping, 0, -1);
427 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);