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>
12 #include <linux/gfp.h>
14 #include <linux/swap.h>
15 #include <linux/module.h>
16 #include <linux/pagemap.h>
17 #include <linux/highmem.h>
18 #include <linux/pagevec.h>
19 #include <linux/task_io_accounting_ops.h>
20 #include <linux/buffer_head.h> /* grr. try_to_release_page,
26 * do_invalidatepage - invalidate part or all of a page
27 * @page: the page which is affected
28 * @offset: the index of the truncation point
30 * do_invalidatepage() is called when all or part of the page has become
31 * invalidated by a truncate operation.
33 * do_invalidatepage() does not have to release all buffers, but it must
34 * ensure that no dirty buffer is left outside @offset and that no I/O
35 * is underway against any of the blocks which are outside the truncation
36 * point. Because the caller is about to free (and possibly reuse) those
39 void do_invalidatepage(struct page *page, unsigned long offset)
41 void (*invalidatepage)(struct page *, unsigned long);
42 invalidatepage = page->mapping->a_ops->invalidatepage;
45 invalidatepage = block_invalidatepage;
48 (*invalidatepage)(page, offset);
51 static inline void truncate_partial_page(struct page *page, unsigned partial)
53 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
54 if (page_has_private(page))
55 do_invalidatepage(page, partial);
59 * This cancels just the dirty bit on the kernel page itself, it
60 * does NOT actually remove dirty bits on any mmap's that may be
61 * around. It also leaves the page tagged dirty, so any sync
62 * activity will still find it on the dirty lists, and in particular,
63 * clear_page_dirty_for_io() will still look at the dirty bits in
66 * Doing this should *normally* only ever be done when a page
67 * is truncated, and is not actually mapped anywhere at all. However,
68 * fs/buffer.c does this when it notices that somebody has cleaned
69 * out all the buffers on a page without actually doing it through
70 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
72 void cancel_dirty_page(struct page *page, unsigned int account_size)
74 if (TestClearPageDirty(page)) {
75 struct address_space *mapping = page->mapping;
76 if (mapping && mapping_cap_account_dirty(mapping)) {
77 dec_zone_page_state(page, NR_FILE_DIRTY);
78 dec_bdi_stat(mapping->backing_dev_info,
81 task_io_account_cancelled_write(account_size);
85 EXPORT_SYMBOL(cancel_dirty_page);
88 * If truncate cannot remove the fs-private metadata from the page, the page
89 * becomes orphaned. It will be left on the LRU and may even be mapped into
90 * user pagetables if we're racing with filemap_fault().
92 * We need to bale out if page->mapping is no longer equal to the original
93 * mapping. This happens a) when the VM reclaimed the page while we waited on
94 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
95 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
98 truncate_complete_page(struct address_space *mapping, struct page *page)
100 if (page->mapping != mapping)
103 if (page_has_private(page))
104 do_invalidatepage(page, 0);
106 cancel_dirty_page(page, PAGE_CACHE_SIZE);
108 clear_page_mlock(page);
109 ClearPageMappedToDisk(page);
110 delete_from_page_cache(page);
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 mem_cgroup_uncharge_start();
228 for (i = 0; i < pagevec_count(&pvec); i++) {
229 struct page *page = pvec.pages[i];
230 pgoff_t page_index = page->index;
232 if (page_index > end) {
237 if (page_index > next)
240 if (!trylock_page(page))
242 if (PageWriteback(page)) {
246 truncate_inode_page(mapping, page);
249 pagevec_release(&pvec);
250 mem_cgroup_uncharge_end();
255 struct page *page = find_lock_page(mapping, start - 1);
257 wait_on_page_writeback(page);
258 truncate_partial_page(page, partial);
260 page_cache_release(page);
267 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
273 if (pvec.pages[0]->index > end) {
274 pagevec_release(&pvec);
277 mem_cgroup_uncharge_start();
278 for (i = 0; i < pagevec_count(&pvec); i++) {
279 struct page *page = pvec.pages[i];
281 if (page->index > end)
284 wait_on_page_writeback(page);
285 truncate_inode_page(mapping, page);
286 if (page->index > next)
291 pagevec_release(&pvec);
292 mem_cgroup_uncharge_end();
295 EXPORT_SYMBOL(truncate_inode_pages_range);
298 * truncate_inode_pages - truncate *all* the pages from an offset
299 * @mapping: mapping to truncate
300 * @lstart: offset from which to truncate
302 * Called under (and serialised by) inode->i_mutex.
304 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
306 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
308 EXPORT_SYMBOL(truncate_inode_pages);
311 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
312 * @mapping: the address_space which holds the pages to invalidate
313 * @start: the offset 'from' which to invalidate
314 * @end: the offset 'to' which to invalidate (inclusive)
316 * This function only removes the unlocked pages, if you want to
317 * remove all the pages of one inode, you must call truncate_inode_pages.
319 * invalidate_mapping_pages() will not block on IO activity. It will not
320 * invalidate pages which are dirty, locked, under writeback or mapped into
323 unsigned long invalidate_mapping_pages(struct address_space *mapping,
324 pgoff_t start, pgoff_t end)
327 pgoff_t next = start;
328 unsigned long ret = 0;
331 pagevec_init(&pvec, 0);
332 while (next <= end &&
333 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
334 mem_cgroup_uncharge_start();
335 for (i = 0; i < pagevec_count(&pvec); i++) {
336 struct page *page = pvec.pages[i];
340 lock_failed = !trylock_page(page);
343 * We really shouldn't be looking at the ->index of an
344 * unlocked page. But we're not allowed to lock these
345 * pages. So we rely upon nobody altering the ->index
346 * of this (pinned-by-us) page.
355 ret += invalidate_inode_page(page);
361 pagevec_release(&pvec);
362 mem_cgroup_uncharge_end();
367 EXPORT_SYMBOL(invalidate_mapping_pages);
370 * This is like invalidate_complete_page(), except it ignores the page's
371 * refcount. We do this because invalidate_inode_pages2() needs stronger
372 * invalidation guarantees, and cannot afford to leave pages behind because
373 * shrink_page_list() has a temp ref on them, or because they're transiently
374 * sitting in the lru_cache_add() pagevecs.
377 invalidate_complete_page2(struct address_space *mapping, struct page *page)
379 if (page->mapping != mapping)
382 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
385 spin_lock_irq(&mapping->tree_lock);
389 clear_page_mlock(page);
390 BUG_ON(page_has_private(page));
391 __remove_from_page_cache(page);
392 spin_unlock_irq(&mapping->tree_lock);
393 mem_cgroup_uncharge_cache_page(page);
395 if (mapping->a_ops->freepage)
396 mapping->a_ops->freepage(page);
398 page_cache_release(page); /* pagecache ref */
401 spin_unlock_irq(&mapping->tree_lock);
405 static int do_launder_page(struct address_space *mapping, struct page *page)
407 if (!PageDirty(page))
409 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
411 return mapping->a_ops->launder_page(page);
415 * invalidate_inode_pages2_range - remove range of pages from an address_space
416 * @mapping: the address_space
417 * @start: the page offset 'from' which to invalidate
418 * @end: the page offset 'to' which to invalidate (inclusive)
420 * Any pages which are found to be mapped into pagetables are unmapped prior to
423 * Returns -EBUSY if any pages could not be invalidated.
425 int invalidate_inode_pages2_range(struct address_space *mapping,
426 pgoff_t start, pgoff_t end)
433 int did_range_unmap = 0;
436 pagevec_init(&pvec, 0);
438 while (next <= end && !wrapped &&
439 pagevec_lookup(&pvec, mapping, next,
440 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
441 mem_cgroup_uncharge_start();
442 for (i = 0; i < pagevec_count(&pvec); i++) {
443 struct page *page = pvec.pages[i];
447 if (page->mapping != mapping) {
451 page_index = page->index;
452 next = page_index + 1;
455 if (page_index > end) {
459 wait_on_page_writeback(page);
460 if (page_mapped(page)) {
461 if (!did_range_unmap) {
463 * Zap the rest of the file in one hit.
465 unmap_mapping_range(mapping,
466 (loff_t)page_index<<PAGE_CACHE_SHIFT,
467 (loff_t)(end - page_index + 1)
475 unmap_mapping_range(mapping,
476 (loff_t)page_index<<PAGE_CACHE_SHIFT,
480 BUG_ON(page_mapped(page));
481 ret2 = do_launder_page(mapping, page);
483 if (!invalidate_complete_page2(mapping, page))
490 pagevec_release(&pvec);
491 mem_cgroup_uncharge_end();
496 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
499 * invalidate_inode_pages2 - remove all pages from an address_space
500 * @mapping: the address_space
502 * Any pages which are found to be mapped into pagetables are unmapped prior to
505 * Returns -EBUSY if any pages could not be invalidated.
507 int invalidate_inode_pages2(struct address_space *mapping)
509 return invalidate_inode_pages2_range(mapping, 0, -1);
511 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
514 * truncate_pagecache - unmap and remove pagecache that has been truncated
516 * @old: old file offset
517 * @new: new file offset
519 * inode's new i_size must already be written before truncate_pagecache
522 * This function should typically be called before the filesystem
523 * releases resources associated with the freed range (eg. deallocates
524 * blocks). This way, pagecache will always stay logically coherent
525 * with on-disk format, and the filesystem would not have to deal with
526 * situations such as writepage being called for a page that has already
527 * had its underlying blocks deallocated.
529 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
531 struct address_space *mapping = inode->i_mapping;
534 * unmap_mapping_range is called twice, first simply for
535 * efficiency so that truncate_inode_pages does fewer
536 * single-page unmaps. However after this first call, and
537 * before truncate_inode_pages finishes, it is possible for
538 * private pages to be COWed, which remain after
539 * truncate_inode_pages finishes, hence the second
540 * unmap_mapping_range call must be made for correctness.
542 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
543 truncate_inode_pages(mapping, new);
544 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
546 EXPORT_SYMBOL(truncate_pagecache);
549 * truncate_setsize - update inode and pagecache for a new file size
551 * @newsize: new file size
553 * truncate_setsize updates i_size and performs pagecache truncation (if
554 * necessary) to @newsize. It will be typically be called from the filesystem's
555 * setattr function when ATTR_SIZE is passed in.
557 * Must be called with inode_mutex held and before all filesystem specific
558 * block truncation has been performed.
560 void truncate_setsize(struct inode *inode, loff_t newsize)
564 oldsize = inode->i_size;
565 i_size_write(inode, newsize);
567 truncate_pagecache(inode, oldsize, newsize);
569 EXPORT_SYMBOL(truncate_setsize);
572 * vmtruncate - unmap mappings "freed" by truncate() syscall
573 * @inode: inode of the file used
574 * @offset: file offset to start truncating
576 * This function is deprecated and truncate_setsize or truncate_pagecache
577 * should be used instead, together with filesystem specific block truncation.
579 int vmtruncate(struct inode *inode, loff_t offset)
583 error = inode_newsize_ok(inode, offset);
587 truncate_setsize(inode, offset);
588 if (inode->i_op->truncate)
589 inode->i_op->truncate(inode);
592 EXPORT_SYMBOL(vmtruncate);