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/precache.h>
20 #include <linux/task_io_accounting_ops.h>
21 #include <linux/buffer_head.h> /* grr. try_to_release_page,
27 * do_invalidatepage - invalidate part or all of a page
28 * @page: the page which is affected
29 * @offset: the index of the truncation point
31 * do_invalidatepage() is called when all or part of the page has become
32 * invalidated by a truncate operation.
34 * do_invalidatepage() does not have to release all buffers, but it must
35 * ensure that no dirty buffer is left outside @offset and that no I/O
36 * is underway against any of the blocks which are outside the truncation
37 * point. Because the caller is about to free (and possibly reuse) those
40 void do_invalidatepage(struct page *page, unsigned long offset)
42 void (*invalidatepage)(struct page *, unsigned long);
43 invalidatepage = page->mapping->a_ops->invalidatepage;
46 invalidatepage = block_invalidatepage;
49 (*invalidatepage)(page, offset);
52 static inline void truncate_partial_page(struct page *page, unsigned partial)
54 zero_user_segment(page, partial, PAGE_CACHE_SIZE);
55 precache_flush(page->mapping, page->index);
56 if (page_has_private(page))
57 do_invalidatepage(page, partial);
61 * This cancels just the dirty bit on the kernel page itself, it
62 * does NOT actually remove dirty bits on any mmap's that may be
63 * around. It also leaves the page tagged dirty, so any sync
64 * activity will still find it on the dirty lists, and in particular,
65 * clear_page_dirty_for_io() will still look at the dirty bits in
68 * Doing this should *normally* only ever be done when a page
69 * is truncated, and is not actually mapped anywhere at all. However,
70 * fs/buffer.c does this when it notices that somebody has cleaned
71 * out all the buffers on a page without actually doing it through
72 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
74 void cancel_dirty_page(struct page *page, unsigned int account_size)
76 if (TestClearPageDirty(page)) {
77 struct address_space *mapping = page->mapping;
78 if (mapping && mapping_cap_account_dirty(mapping)) {
79 dec_zone_page_state(page, NR_FILE_DIRTY);
80 dec_bdi_stat(mapping->backing_dev_info,
83 task_io_account_cancelled_write(account_size);
87 EXPORT_SYMBOL(cancel_dirty_page);
90 * If truncate cannot remove the fs-private metadata from the page, the page
91 * becomes orphaned. It will be left on the LRU and may even be mapped into
92 * user pagetables if we're racing with filemap_fault().
94 * We need to bale out if page->mapping is no longer equal to the original
95 * mapping. This happens a) when the VM reclaimed the page while we waited on
96 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
97 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
100 truncate_complete_page(struct address_space *mapping, struct page *page)
102 if (page->mapping != mapping)
105 if (page_has_private(page))
106 do_invalidatepage(page, 0);
108 cancel_dirty_page(page, PAGE_CACHE_SIZE);
110 clear_page_mlock(page);
111 remove_from_page_cache(page);
112 ClearPageMappedToDisk(page);
113 /* this must be after the remove_from_page_cache which
116 precache_flush(mapping, page->index);
117 page_cache_release(page); /* pagecache ref */
122 * This is for invalidate_mapping_pages(). That function can be called at
123 * any time, and is not supposed to throw away dirty pages. But pages can
124 * be marked dirty at any time too, so use remove_mapping which safely
125 * discards clean, unused pages.
127 * Returns non-zero if the page was successfully invalidated.
130 invalidate_complete_page(struct address_space *mapping, struct page *page)
134 if (page->mapping != mapping)
137 if (page_has_private(page) && !try_to_release_page(page, 0))
140 clear_page_mlock(page);
141 ret = remove_mapping(mapping, page);
146 int truncate_inode_page(struct address_space *mapping, struct page *page)
148 if (page_mapped(page)) {
149 unmap_mapping_range(mapping,
150 (loff_t)page->index << PAGE_CACHE_SHIFT,
153 return truncate_complete_page(mapping, page);
157 * Used to get rid of pages on hardware memory corruption.
159 int generic_error_remove_page(struct address_space *mapping, struct page *page)
164 * Only punch for normal data pages for now.
165 * Handling other types like directories would need more auditing.
167 if (!S_ISREG(mapping->host->i_mode))
169 return truncate_inode_page(mapping, page);
171 EXPORT_SYMBOL(generic_error_remove_page);
174 * Safely invalidate one page from its pagecache mapping.
175 * It only drops clean, unused pages. The page must be locked.
177 * Returns 1 if the page is successfully invalidated, otherwise 0.
179 int invalidate_inode_page(struct page *page)
181 struct address_space *mapping = page_mapping(page);
184 if (PageDirty(page) || PageWriteback(page))
186 if (page_mapped(page))
188 return invalidate_complete_page(mapping, page);
192 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
193 * @mapping: mapping to truncate
194 * @lstart: offset from which to truncate
195 * @lend: offset to which to truncate
197 * Truncate the page cache, removing the pages that are between
198 * specified offsets (and zeroing out partial page
199 * (if lstart is not page aligned)).
201 * Truncate takes two passes - the first pass is nonblocking. It will not
202 * block on page locks and it will not block on writeback. The second pass
203 * will wait. This is to prevent as much IO as possible in the affected region.
204 * The first pass will remove most pages, so the search cost of the second pass
207 * When looking at page->index outside the page lock we need to be careful to
208 * copy it into a local to avoid races (it could change at any time).
210 * We pass down the cache-hot hint to the page freeing code. Even if the
211 * mapping is large, it is probably the case that the final pages are the most
212 * recently touched, and freeing happens in ascending file offset order.
214 void truncate_inode_pages_range(struct address_space *mapping,
215 loff_t lstart, loff_t lend)
217 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
219 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
224 precache_flush_inode(mapping);
225 if (mapping->nrpages == 0)
228 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
229 end = (lend >> PAGE_CACHE_SHIFT);
231 pagevec_init(&pvec, 0);
233 while (next <= end &&
234 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
235 for (i = 0; i < pagevec_count(&pvec); i++) {
236 struct page *page = pvec.pages[i];
237 pgoff_t page_index = page->index;
239 if (page_index > end) {
244 if (page_index > next)
247 if (!trylock_page(page))
249 if (PageWriteback(page)) {
253 truncate_inode_page(mapping, page);
256 pagevec_release(&pvec);
261 struct page *page = find_lock_page(mapping, start - 1);
263 wait_on_page_writeback(page);
264 truncate_partial_page(page, partial);
266 page_cache_release(page);
273 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
279 if (pvec.pages[0]->index > end) {
280 pagevec_release(&pvec);
283 mem_cgroup_uncharge_start();
284 for (i = 0; i < pagevec_count(&pvec); i++) {
285 struct page *page = pvec.pages[i];
287 if (page->index > end)
290 wait_on_page_writeback(page);
291 truncate_inode_page(mapping, page);
292 if (page->index > next)
297 pagevec_release(&pvec);
298 mem_cgroup_uncharge_end();
300 precache_flush_inode(mapping);
302 EXPORT_SYMBOL(truncate_inode_pages_range);
305 * truncate_inode_pages - truncate *all* the pages from an offset
306 * @mapping: mapping to truncate
307 * @lstart: offset from which to truncate
309 * Called under (and serialised by) inode->i_mutex.
311 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
313 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
315 EXPORT_SYMBOL(truncate_inode_pages);
318 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
319 * @mapping: the address_space which holds the pages to invalidate
320 * @start: the offset 'from' which to invalidate
321 * @end: the offset 'to' which to invalidate (inclusive)
323 * This function only removes the unlocked pages, if you want to
324 * remove all the pages of one inode, you must call truncate_inode_pages.
326 * invalidate_mapping_pages() will not block on IO activity. It will not
327 * invalidate pages which are dirty, locked, under writeback or mapped into
330 unsigned long invalidate_mapping_pages(struct address_space *mapping,
331 pgoff_t start, pgoff_t end)
334 pgoff_t next = start;
335 unsigned long ret = 0;
338 pagevec_init(&pvec, 0);
339 while (next <= end &&
340 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
341 mem_cgroup_uncharge_start();
342 for (i = 0; i < pagevec_count(&pvec); i++) {
343 struct page *page = pvec.pages[i];
347 lock_failed = !trylock_page(page);
350 * We really shouldn't be looking at the ->index of an
351 * unlocked page. But we're not allowed to lock these
352 * pages. So we rely upon nobody altering the ->index
353 * of this (pinned-by-us) page.
362 ret += invalidate_inode_page(page);
368 pagevec_release(&pvec);
369 mem_cgroup_uncharge_end();
374 EXPORT_SYMBOL(invalidate_mapping_pages);
377 * This is like invalidate_complete_page(), except it ignores the page's
378 * refcount. We do this because invalidate_inode_pages2() needs stronger
379 * invalidation guarantees, and cannot afford to leave pages behind because
380 * shrink_page_list() has a temp ref on them, or because they're transiently
381 * sitting in the lru_cache_add() pagevecs.
384 invalidate_complete_page2(struct address_space *mapping, struct page *page)
386 if (page->mapping != mapping)
389 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
392 spin_lock_irq(&mapping->tree_lock);
396 clear_page_mlock(page);
397 BUG_ON(page_has_private(page));
398 __remove_from_page_cache(page);
399 spin_unlock_irq(&mapping->tree_lock);
400 mem_cgroup_uncharge_cache_page(page);
401 page_cache_release(page); /* pagecache ref */
404 spin_unlock_irq(&mapping->tree_lock);
408 static int do_launder_page(struct address_space *mapping, struct page *page)
410 if (!PageDirty(page))
412 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
414 return mapping->a_ops->launder_page(page);
418 * invalidate_inode_pages2_range - remove range of pages from an address_space
419 * @mapping: the address_space
420 * @start: the page offset 'from' which to invalidate
421 * @end: the page offset 'to' which to invalidate (inclusive)
423 * Any pages which are found to be mapped into pagetables are unmapped prior to
426 * Returns -EBUSY if any pages could not be invalidated.
428 int invalidate_inode_pages2_range(struct address_space *mapping,
429 pgoff_t start, pgoff_t end)
436 int did_range_unmap = 0;
439 precache_flush_inode(mapping);
440 pagevec_init(&pvec, 0);
442 while (next <= end && !wrapped &&
443 pagevec_lookup(&pvec, mapping, next,
444 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
445 mem_cgroup_uncharge_start();
446 for (i = 0; i < pagevec_count(&pvec); i++) {
447 struct page *page = pvec.pages[i];
451 if (page->mapping != mapping) {
455 page_index = page->index;
456 next = page_index + 1;
459 if (page_index > end) {
463 wait_on_page_writeback(page);
464 if (page_mapped(page)) {
465 if (!did_range_unmap) {
467 * Zap the rest of the file in one hit.
469 unmap_mapping_range(mapping,
470 (loff_t)page_index<<PAGE_CACHE_SHIFT,
471 (loff_t)(end - page_index + 1)
479 unmap_mapping_range(mapping,
480 (loff_t)page_index<<PAGE_CACHE_SHIFT,
484 BUG_ON(page_mapped(page));
485 ret2 = do_launder_page(mapping, page);
487 if (!invalidate_complete_page2(mapping, page))
494 pagevec_release(&pvec);
495 mem_cgroup_uncharge_end();
498 precache_flush_inode(mapping);
501 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
504 * invalidate_inode_pages2 - remove all pages from an address_space
505 * @mapping: the address_space
507 * Any pages which are found to be mapped into pagetables are unmapped prior to
510 * Returns -EBUSY if any pages could not be invalidated.
512 int invalidate_inode_pages2(struct address_space *mapping)
514 return invalidate_inode_pages2_range(mapping, 0, -1);
516 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
519 * truncate_pagecache - unmap and remove pagecache that has been truncated
521 * @old: old file offset
522 * @new: new file offset
524 * inode's new i_size must already be written before truncate_pagecache
527 * This function should typically be called before the filesystem
528 * releases resources associated with the freed range (eg. deallocates
529 * blocks). This way, pagecache will always stay logically coherent
530 * with on-disk format, and the filesystem would not have to deal with
531 * situations such as writepage being called for a page that has already
532 * had its underlying blocks deallocated.
534 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
536 struct address_space *mapping = inode->i_mapping;
539 * unmap_mapping_range is called twice, first simply for
540 * efficiency so that truncate_inode_pages does fewer
541 * single-page unmaps. However after this first call, and
542 * before truncate_inode_pages finishes, it is possible for
543 * private pages to be COWed, which remain after
544 * truncate_inode_pages finishes, hence the second
545 * unmap_mapping_range call must be made for correctness.
547 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
548 truncate_inode_pages(mapping, new);
549 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
551 EXPORT_SYMBOL(truncate_pagecache);
554 * truncate_setsize - update inode and pagecache for a new file size
556 * @newsize: new file size
558 * truncate_setsize updastes i_size update and performs pagecache
559 * truncation (if necessary) for a file size updates. It will be
560 * typically be called from the filesystem's setattr function when
561 * ATTR_SIZE is passed in.
563 * Must be called with inode_mutex held and after all filesystem
564 * specific block truncation has been performed.
566 void truncate_setsize(struct inode *inode, loff_t newsize)
570 oldsize = inode->i_size;
571 i_size_write(inode, newsize);
573 truncate_pagecache(inode, oldsize, newsize);
575 EXPORT_SYMBOL(truncate_setsize);
578 * vmtruncate - unmap mappings "freed" by truncate() syscall
579 * @inode: inode of the file used
580 * @offset: file offset to start truncating
582 * This function is deprecated and truncate_setsize or truncate_pagecache
583 * should be used instead, together with filesystem specific block truncation.
585 int vmtruncate(struct inode *inode, loff_t offset)
589 error = inode_newsize_ok(inode, offset);
593 truncate_setsize(inode, offset);
594 if (inode->i_op->truncate)
595 inode->i_op->truncate(inode);
598 EXPORT_SYMBOL(vmtruncate);