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 * Used to get rid of pages on hardware memory corruption.
152 int generic_error_remove_page(struct address_space *mapping, struct page *page)
157 * Only punch for normal data pages for now.
158 * Handling other types like directories would need more auditing.
160 if (!S_ISREG(mapping->host->i_mode))
162 return truncate_inode_page(mapping, page);
164 EXPORT_SYMBOL(generic_error_remove_page);
167 * Safely invalidate one page from its pagecache mapping.
168 * It only drops clean, unused pages. The page must be locked.
170 * Returns 1 if the page is successfully invalidated, otherwise 0.
172 int invalidate_inode_page(struct page *page)
174 struct address_space *mapping = page_mapping(page);
177 if (PageDirty(page) || PageWriteback(page))
179 if (page_mapped(page))
181 return invalidate_complete_page(mapping, page);
185 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
186 * @mapping: mapping to truncate
187 * @lstart: offset from which to truncate
188 * @lend: offset to which to truncate
190 * Truncate the page cache, removing the pages that are between
191 * specified offsets (and zeroing out partial page
192 * (if lstart is not page aligned)).
194 * Truncate takes two passes - the first pass is nonblocking. It will not
195 * block on page locks and it will not block on writeback. The second pass
196 * will wait. This is to prevent as much IO as possible in the affected region.
197 * The first pass will remove most pages, so the search cost of the second pass
200 * When looking at page->index outside the page lock we need to be careful to
201 * copy it into a local to avoid races (it could change at any time).
203 * We pass down the cache-hot hint to the page freeing code. Even if the
204 * mapping is large, it is probably the case that the final pages are the most
205 * recently touched, and freeing happens in ascending file offset order.
207 void truncate_inode_pages_range(struct address_space *mapping,
208 loff_t lstart, loff_t lend)
210 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
212 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
217 if (mapping->nrpages == 0)
220 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
221 end = (lend >> PAGE_CACHE_SHIFT);
223 pagevec_init(&pvec, 0);
225 while (next <= end &&
226 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
227 for (i = 0; i < pagevec_count(&pvec); i++) {
228 struct page *page = pvec.pages[i];
229 pgoff_t page_index = page->index;
231 if (page_index > end) {
236 if (page_index > next)
239 if (!trylock_page(page))
241 if (PageWriteback(page)) {
245 truncate_inode_page(mapping, page);
248 pagevec_release(&pvec);
253 struct page *page = find_lock_page(mapping, start - 1);
255 wait_on_page_writeback(page);
256 truncate_partial_page(page, partial);
258 page_cache_release(page);
265 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
271 if (pvec.pages[0]->index > end) {
272 pagevec_release(&pvec);
275 mem_cgroup_uncharge_start();
276 for (i = 0; i < pagevec_count(&pvec); i++) {
277 struct page *page = pvec.pages[i];
279 if (page->index > end)
282 wait_on_page_writeback(page);
283 truncate_inode_page(mapping, page);
284 if (page->index > next)
289 pagevec_release(&pvec);
290 mem_cgroup_uncharge_end();
293 EXPORT_SYMBOL(truncate_inode_pages_range);
296 * truncate_inode_pages - truncate *all* the pages from an offset
297 * @mapping: mapping to truncate
298 * @lstart: offset from which to truncate
300 * Called under (and serialised by) inode->i_mutex.
302 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
304 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
306 EXPORT_SYMBOL(truncate_inode_pages);
309 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
310 * @mapping: the address_space which holds the pages to invalidate
311 * @start: the offset 'from' which to invalidate
312 * @end: the offset 'to' which to invalidate (inclusive)
314 * This function only removes the unlocked pages, if you want to
315 * remove all the pages of one inode, you must call truncate_inode_pages.
317 * invalidate_mapping_pages() will not block on IO activity. It will not
318 * invalidate pages which are dirty, locked, under writeback or mapped into
321 unsigned long invalidate_mapping_pages(struct address_space *mapping,
322 pgoff_t start, pgoff_t end)
325 pgoff_t next = start;
326 unsigned long ret = 0;
329 pagevec_init(&pvec, 0);
330 while (next <= end &&
331 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
332 mem_cgroup_uncharge_start();
333 for (i = 0; i < pagevec_count(&pvec); i++) {
334 struct page *page = pvec.pages[i];
338 lock_failed = !trylock_page(page);
341 * We really shouldn't be looking at the ->index of an
342 * unlocked page. But we're not allowed to lock these
343 * pages. So we rely upon nobody altering the ->index
344 * of this (pinned-by-us) page.
353 ret += invalidate_inode_page(page);
359 pagevec_release(&pvec);
360 mem_cgroup_uncharge_end();
365 EXPORT_SYMBOL(invalidate_mapping_pages);
368 * This is like invalidate_complete_page(), except it ignores the page's
369 * refcount. We do this because invalidate_inode_pages2() needs stronger
370 * invalidation guarantees, and cannot afford to leave pages behind because
371 * shrink_page_list() has a temp ref on them, or because they're transiently
372 * sitting in the lru_cache_add() pagevecs.
375 invalidate_complete_page2(struct address_space *mapping, struct page *page)
377 if (page->mapping != mapping)
380 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
383 spin_lock_irq(&mapping->tree_lock);
387 clear_page_mlock(page);
388 BUG_ON(page_has_private(page));
389 __remove_from_page_cache(page);
390 spin_unlock_irq(&mapping->tree_lock);
391 mem_cgroup_uncharge_cache_page(page);
392 page_cache_release(page); /* pagecache ref */
395 spin_unlock_irq(&mapping->tree_lock);
399 static int do_launder_page(struct address_space *mapping, struct page *page)
401 if (!PageDirty(page))
403 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
405 return mapping->a_ops->launder_page(page);
409 * invalidate_inode_pages2_range - remove range of pages from an address_space
410 * @mapping: the address_space
411 * @start: the page offset 'from' which to invalidate
412 * @end: the page offset 'to' which to invalidate (inclusive)
414 * Any pages which are found to be mapped into pagetables are unmapped prior to
417 * Returns -EBUSY if any pages could not be invalidated.
419 int invalidate_inode_pages2_range(struct address_space *mapping,
420 pgoff_t start, pgoff_t end)
427 int did_range_unmap = 0;
430 pagevec_init(&pvec, 0);
432 while (next <= end && !wrapped &&
433 pagevec_lookup(&pvec, mapping, next,
434 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
435 mem_cgroup_uncharge_start();
436 for (i = 0; i < pagevec_count(&pvec); i++) {
437 struct page *page = pvec.pages[i];
441 if (page->mapping != mapping) {
445 page_index = page->index;
446 next = page_index + 1;
449 if (page_index > end) {
453 wait_on_page_writeback(page);
454 if (page_mapped(page)) {
455 if (!did_range_unmap) {
457 * Zap the rest of the file in one hit.
459 unmap_mapping_range(mapping,
460 (loff_t)page_index<<PAGE_CACHE_SHIFT,
461 (loff_t)(end - page_index + 1)
469 unmap_mapping_range(mapping,
470 (loff_t)page_index<<PAGE_CACHE_SHIFT,
474 BUG_ON(page_mapped(page));
475 ret2 = do_launder_page(mapping, page);
477 if (!invalidate_complete_page2(mapping, page))
484 pagevec_release(&pvec);
485 mem_cgroup_uncharge_end();
490 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
493 * invalidate_inode_pages2 - remove all pages from an address_space
494 * @mapping: the address_space
496 * Any pages which are found to be mapped into pagetables are unmapped prior to
499 * Returns -EBUSY if any pages could not be invalidated.
501 int invalidate_inode_pages2(struct address_space *mapping)
503 return invalidate_inode_pages2_range(mapping, 0, -1);
505 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
508 * truncate_pagecache - unmap and remove pagecache that has been truncated
510 * @old: old file offset
511 * @new: new file offset
513 * inode's new i_size must already be written before truncate_pagecache
516 * This function should typically be called before the filesystem
517 * releases resources associated with the freed range (eg. deallocates
518 * blocks). This way, pagecache will always stay logically coherent
519 * with on-disk format, and the filesystem would not have to deal with
520 * situations such as writepage being called for a page that has already
521 * had its underlying blocks deallocated.
523 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
526 struct address_space *mapping = inode->i_mapping;
529 * unmap_mapping_range is called twice, first simply for
530 * efficiency so that truncate_inode_pages does fewer
531 * single-page unmaps. However after this first call, and
532 * before truncate_inode_pages finishes, it is possible for
533 * private pages to be COWed, which remain after
534 * truncate_inode_pages finishes, hence the second
535 * unmap_mapping_range call must be made for correctness.
537 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
538 truncate_inode_pages(mapping, new);
539 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
542 EXPORT_SYMBOL(truncate_pagecache);
545 * vmtruncate - unmap mappings "freed" by truncate() syscall
546 * @inode: inode of the file used
547 * @offset: file offset to start truncating
549 * NOTE! We have to be ready to update the memory sharing
550 * between the file and the memory map for a potential last
551 * incomplete page. Ugly, but necessary.
553 int vmtruncate(struct inode *inode, loff_t offset)
558 error = inode_newsize_ok(inode, offset);
561 oldsize = inode->i_size;
562 i_size_write(inode, offset);
563 truncate_pagecache(inode, oldsize, offset);
564 if (inode->i_op->truncate)
565 inode->i_op->truncate(inode);
569 EXPORT_SYMBOL(vmtruncate);