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 mem_cgroup_uncharge_start();
236 for (i = 0; i < pagevec_count(&pvec); i++) {
237 struct page *page = pvec.pages[i];
238 pgoff_t page_index = page->index;
240 if (page_index > end) {
245 if (page_index > next)
248 if (!trylock_page(page))
250 if (PageWriteback(page)) {
254 truncate_inode_page(mapping, page);
257 pagevec_release(&pvec);
258 mem_cgroup_uncharge_end();
263 struct page *page = find_lock_page(mapping, start - 1);
265 wait_on_page_writeback(page);
266 truncate_partial_page(page, partial);
268 page_cache_release(page);
275 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
281 if (pvec.pages[0]->index > end) {
282 pagevec_release(&pvec);
285 mem_cgroup_uncharge_start();
286 for (i = 0; i < pagevec_count(&pvec); i++) {
287 struct page *page = pvec.pages[i];
289 if (page->index > end)
292 wait_on_page_writeback(page);
293 truncate_inode_page(mapping, page);
294 if (page->index > next)
299 pagevec_release(&pvec);
300 mem_cgroup_uncharge_end();
302 precache_flush_inode(mapping);
304 EXPORT_SYMBOL(truncate_inode_pages_range);
307 * truncate_inode_pages - truncate *all* the pages from an offset
308 * @mapping: mapping to truncate
309 * @lstart: offset from which to truncate
311 * Called under (and serialised by) inode->i_mutex.
313 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
315 truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
317 EXPORT_SYMBOL(truncate_inode_pages);
320 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
321 * @mapping: the address_space which holds the pages to invalidate
322 * @start: the offset 'from' which to invalidate
323 * @end: the offset 'to' which to invalidate (inclusive)
325 * This function only removes the unlocked pages, if you want to
326 * remove all the pages of one inode, you must call truncate_inode_pages.
328 * invalidate_mapping_pages() will not block on IO activity. It will not
329 * invalidate pages which are dirty, locked, under writeback or mapped into
332 unsigned long invalidate_mapping_pages(struct address_space *mapping,
333 pgoff_t start, pgoff_t end)
336 pgoff_t next = start;
337 unsigned long ret = 0;
340 pagevec_init(&pvec, 0);
341 while (next <= end &&
342 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
343 mem_cgroup_uncharge_start();
344 for (i = 0; i < pagevec_count(&pvec); i++) {
345 struct page *page = pvec.pages[i];
349 lock_failed = !trylock_page(page);
352 * We really shouldn't be looking at the ->index of an
353 * unlocked page. But we're not allowed to lock these
354 * pages. So we rely upon nobody altering the ->index
355 * of this (pinned-by-us) page.
364 ret += invalidate_inode_page(page);
370 pagevec_release(&pvec);
371 mem_cgroup_uncharge_end();
376 EXPORT_SYMBOL(invalidate_mapping_pages);
379 * This is like invalidate_complete_page(), except it ignores the page's
380 * refcount. We do this because invalidate_inode_pages2() needs stronger
381 * invalidation guarantees, and cannot afford to leave pages behind because
382 * shrink_page_list() has a temp ref on them, or because they're transiently
383 * sitting in the lru_cache_add() pagevecs.
386 invalidate_complete_page2(struct address_space *mapping, struct page *page)
388 if (page->mapping != mapping)
391 if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
394 spin_lock_irq(&mapping->tree_lock);
398 clear_page_mlock(page);
399 BUG_ON(page_has_private(page));
400 __remove_from_page_cache(page);
401 spin_unlock_irq(&mapping->tree_lock);
402 mem_cgroup_uncharge_cache_page(page);
404 if (mapping->a_ops->freepage)
405 mapping->a_ops->freepage(page);
407 page_cache_release(page); /* pagecache ref */
410 spin_unlock_irq(&mapping->tree_lock);
414 static int do_launder_page(struct address_space *mapping, struct page *page)
416 if (!PageDirty(page))
418 if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
420 return mapping->a_ops->launder_page(page);
424 * invalidate_inode_pages2_range - remove range of pages from an address_space
425 * @mapping: the address_space
426 * @start: the page offset 'from' which to invalidate
427 * @end: the page offset 'to' which to invalidate (inclusive)
429 * Any pages which are found to be mapped into pagetables are unmapped prior to
432 * Returns -EBUSY if any pages could not be invalidated.
434 int invalidate_inode_pages2_range(struct address_space *mapping,
435 pgoff_t start, pgoff_t end)
442 int did_range_unmap = 0;
445 precache_flush_inode(mapping);
446 pagevec_init(&pvec, 0);
448 while (next <= end && !wrapped &&
449 pagevec_lookup(&pvec, mapping, next,
450 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
451 mem_cgroup_uncharge_start();
452 for (i = 0; i < pagevec_count(&pvec); i++) {
453 struct page *page = pvec.pages[i];
457 if (page->mapping != mapping) {
461 page_index = page->index;
462 next = page_index + 1;
465 if (page_index > end) {
469 wait_on_page_writeback(page);
470 if (page_mapped(page)) {
471 if (!did_range_unmap) {
473 * Zap the rest of the file in one hit.
475 unmap_mapping_range(mapping,
476 (loff_t)page_index<<PAGE_CACHE_SHIFT,
477 (loff_t)(end - page_index + 1)
485 unmap_mapping_range(mapping,
486 (loff_t)page_index<<PAGE_CACHE_SHIFT,
490 BUG_ON(page_mapped(page));
491 ret2 = do_launder_page(mapping, page);
493 if (!invalidate_complete_page2(mapping, page))
500 pagevec_release(&pvec);
501 mem_cgroup_uncharge_end();
504 precache_flush_inode(mapping);
507 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
510 * invalidate_inode_pages2 - remove all pages from an address_space
511 * @mapping: the address_space
513 * Any pages which are found to be mapped into pagetables are unmapped prior to
516 * Returns -EBUSY if any pages could not be invalidated.
518 int invalidate_inode_pages2(struct address_space *mapping)
520 return invalidate_inode_pages2_range(mapping, 0, -1);
522 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
525 * truncate_pagecache - unmap and remove pagecache that has been truncated
527 * @old: old file offset
528 * @new: new file offset
530 * inode's new i_size must already be written before truncate_pagecache
533 * This function should typically be called before the filesystem
534 * releases resources associated with the freed range (eg. deallocates
535 * blocks). This way, pagecache will always stay logically coherent
536 * with on-disk format, and the filesystem would not have to deal with
537 * situations such as writepage being called for a page that has already
538 * had its underlying blocks deallocated.
540 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
542 struct address_space *mapping = inode->i_mapping;
545 * unmap_mapping_range is called twice, first simply for
546 * efficiency so that truncate_inode_pages does fewer
547 * single-page unmaps. However after this first call, and
548 * before truncate_inode_pages finishes, it is possible for
549 * private pages to be COWed, which remain after
550 * truncate_inode_pages finishes, hence the second
551 * unmap_mapping_range call must be made for correctness.
553 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
554 truncate_inode_pages(mapping, new);
555 unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
557 EXPORT_SYMBOL(truncate_pagecache);
560 * truncate_setsize - update inode and pagecache for a new file size
562 * @newsize: new file size
564 * truncate_setsize updates i_size and performs pagecache truncation (if
565 * necessary) to @newsize. It will be typically be called from the filesystem's
566 * setattr function when ATTR_SIZE is passed in.
568 * Must be called with inode_mutex held and before all filesystem specific
569 * block truncation has been performed.
571 void truncate_setsize(struct inode *inode, loff_t newsize)
575 oldsize = inode->i_size;
576 i_size_write(inode, newsize);
578 truncate_pagecache(inode, oldsize, newsize);
580 EXPORT_SYMBOL(truncate_setsize);
583 * vmtruncate - unmap mappings "freed" by truncate() syscall
584 * @inode: inode of the file used
585 * @offset: file offset to start truncating
587 * This function is deprecated and truncate_setsize or truncate_pagecache
588 * should be used instead, together with filesystem specific block truncation.
590 int vmtruncate(struct inode *inode, loff_t offset)
594 error = inode_newsize_ok(inode, offset);
598 truncate_setsize(inode, offset);
599 if (inode->i_op->truncate)
600 inode->i_op->truncate(inode);
603 EXPORT_SYMBOL(vmtruncate);