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/export.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,
22 #include <linux/cleancache.h>
25 static void clear_exceptional_entry(struct address_space
*mapping
,
26 pgoff_t index
, void *entry
)
28 struct radix_tree_node
*node
;
31 /* Handled by shmem itself */
32 if (shmem_mapping(mapping
))
35 spin_lock_irq(&mapping
->tree_lock
);
37 * Regular page slots are stabilized by the page lock even
38 * without the tree itself locked. These unlocked entries
39 * need verification under the tree lock.
41 if (!__radix_tree_lookup(&mapping
->page_tree
, index
, &node
, &slot
))
45 radix_tree_replace_slot(slot
, NULL
);
49 workingset_node_shadows_dec(node
);
51 * Don't track node without shadow entries.
53 * Avoid acquiring the list_lru lock if already untracked.
54 * The list_empty() test is safe as node->private_list is
55 * protected by mapping->tree_lock.
57 if (!workingset_node_shadows(node
) &&
58 !list_empty(&node
->private_list
))
59 list_lru_del(&workingset_shadow_nodes
, &node
->private_list
);
60 __radix_tree_delete_node(&mapping
->page_tree
, node
);
62 spin_unlock_irq(&mapping
->tree_lock
);
66 * do_invalidatepage - invalidate part or all of a page
67 * @page: the page which is affected
68 * @offset: start of the range to invalidate
69 * @length: length of the range to invalidate
71 * do_invalidatepage() is called when all or part of the page has become
72 * invalidated by a truncate operation.
74 * do_invalidatepage() does not have to release all buffers, but it must
75 * ensure that no dirty buffer is left outside @offset and that no I/O
76 * is underway against any of the blocks which are outside the truncation
77 * point. Because the caller is about to free (and possibly reuse) those
80 void do_invalidatepage(struct page
*page
, unsigned int offset
,
83 void (*invalidatepage
)(struct page
*, unsigned int, unsigned int);
85 invalidatepage
= page
->mapping
->a_ops
->invalidatepage
;
88 invalidatepage
= block_invalidatepage
;
91 (*invalidatepage
)(page
, offset
, length
);
95 * This cancels just the dirty bit on the kernel page itself, it
96 * does NOT actually remove dirty bits on any mmap's that may be
97 * around. It also leaves the page tagged dirty, so any sync
98 * activity will still find it on the dirty lists, and in particular,
99 * clear_page_dirty_for_io() will still look at the dirty bits in
102 * Doing this should *normally* only ever be done when a page
103 * is truncated, and is not actually mapped anywhere at all. However,
104 * fs/buffer.c does this when it notices that somebody has cleaned
105 * out all the buffers on a page without actually doing it through
106 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
108 void cancel_dirty_page(struct page
*page
, unsigned int account_size
)
110 if (TestClearPageDirty(page
)) {
111 struct address_space
*mapping
= page
->mapping
;
112 if (mapping
&& mapping_cap_account_dirty(mapping
)) {
113 dec_zone_page_state(page
, NR_FILE_DIRTY
);
114 dec_bdi_stat(mapping
->backing_dev_info
,
117 task_io_account_cancelled_write(account_size
);
121 EXPORT_SYMBOL(cancel_dirty_page
);
124 * If truncate cannot remove the fs-private metadata from the page, the page
125 * becomes orphaned. It will be left on the LRU and may even be mapped into
126 * user pagetables if we're racing with filemap_fault().
128 * We need to bale out if page->mapping is no longer equal to the original
129 * mapping. This happens a) when the VM reclaimed the page while we waited on
130 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
131 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
134 truncate_complete_page(struct address_space
*mapping
, struct page
*page
)
136 if (page
->mapping
!= mapping
)
139 if (page_has_private(page
))
140 do_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
142 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
144 ClearPageMappedToDisk(page
);
145 delete_from_page_cache(page
);
150 * This is for invalidate_mapping_pages(). That function can be called at
151 * any time, and is not supposed to throw away dirty pages. But pages can
152 * be marked dirty at any time too, so use remove_mapping which safely
153 * discards clean, unused pages.
155 * Returns non-zero if the page was successfully invalidated.
158 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
162 if (page
->mapping
!= mapping
)
165 if (page_has_private(page
) && !try_to_release_page(page
, 0))
168 ret
= remove_mapping(mapping
, page
);
173 int truncate_inode_page(struct address_space
*mapping
, struct page
*page
)
175 if (page_mapped(page
)) {
176 unmap_mapping_range(mapping
,
177 (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
,
180 return truncate_complete_page(mapping
, page
);
184 * Used to get rid of pages on hardware memory corruption.
186 int generic_error_remove_page(struct address_space
*mapping
, struct page
*page
)
191 * Only punch for normal data pages for now.
192 * Handling other types like directories would need more auditing.
194 if (!S_ISREG(mapping
->host
->i_mode
))
196 return truncate_inode_page(mapping
, page
);
198 EXPORT_SYMBOL(generic_error_remove_page
);
201 * Safely invalidate one page from its pagecache mapping.
202 * It only drops clean, unused pages. The page must be locked.
204 * Returns 1 if the page is successfully invalidated, otherwise 0.
206 int invalidate_inode_page(struct page
*page
)
208 struct address_space
*mapping
= page_mapping(page
);
211 if (PageDirty(page
) || PageWriteback(page
))
213 if (page_mapped(page
))
215 return invalidate_complete_page(mapping
, page
);
219 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
220 * @mapping: mapping to truncate
221 * @lstart: offset from which to truncate
222 * @lend: offset to which to truncate (inclusive)
224 * Truncate the page cache, removing the pages that are between
225 * specified offsets (and zeroing out partial pages
226 * if lstart or lend + 1 is not page aligned).
228 * Truncate takes two passes - the first pass is nonblocking. It will not
229 * block on page locks and it will not block on writeback. The second pass
230 * will wait. This is to prevent as much IO as possible in the affected region.
231 * The first pass will remove most pages, so the search cost of the second pass
234 * We pass down the cache-hot hint to the page freeing code. Even if the
235 * mapping is large, it is probably the case that the final pages are the most
236 * recently touched, and freeing happens in ascending file offset order.
238 * Note that since ->invalidatepage() accepts range to invalidate
239 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
240 * page aligned properly.
242 void truncate_inode_pages_range(struct address_space
*mapping
,
243 loff_t lstart
, loff_t lend
)
245 pgoff_t start
; /* inclusive */
246 pgoff_t end
; /* exclusive */
247 unsigned int partial_start
; /* inclusive */
248 unsigned int partial_end
; /* exclusive */
250 pgoff_t indices
[PAGEVEC_SIZE
];
254 cleancache_invalidate_inode(mapping
);
255 if (mapping
->nrpages
== 0 && mapping
->nrshadows
== 0)
258 /* Offsets within partial pages */
259 partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
260 partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
263 * 'start' and 'end' always covers the range of pages to be fully
264 * truncated. Partial pages are covered with 'partial_start' at the
265 * start of the range and 'partial_end' at the end of the range.
266 * Note that 'end' is exclusive while 'lend' is inclusive.
268 start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
271 * lend == -1 indicates end-of-file so we have to set 'end'
272 * to the highest possible pgoff_t and since the type is
273 * unsigned we're using -1.
277 end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
279 pagevec_init(&pvec
, 0);
281 while (index
< end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
282 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
284 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
285 struct page
*page
= pvec
.pages
[i
];
287 /* We rely upon deletion not changing page->index */
292 if (radix_tree_exceptional_entry(page
)) {
293 clear_exceptional_entry(mapping
, index
, page
);
297 if (!trylock_page(page
))
299 WARN_ON(page
->index
!= index
);
300 if (PageWriteback(page
)) {
304 truncate_inode_page(mapping
, page
);
307 pagevec_remove_exceptionals(&pvec
);
308 pagevec_release(&pvec
);
314 struct page
*page
= find_lock_page(mapping
, start
- 1);
316 unsigned int top
= PAGE_CACHE_SIZE
;
318 /* Truncation within a single page */
322 wait_on_page_writeback(page
);
323 zero_user_segment(page
, partial_start
, top
);
324 cleancache_invalidate_page(mapping
, page
);
325 if (page_has_private(page
))
326 do_invalidatepage(page
, partial_start
,
327 top
- partial_start
);
329 page_cache_release(page
);
333 struct page
*page
= find_lock_page(mapping
, end
);
335 wait_on_page_writeback(page
);
336 zero_user_segment(page
, 0, partial_end
);
337 cleancache_invalidate_page(mapping
, page
);
338 if (page_has_private(page
))
339 do_invalidatepage(page
, 0,
342 page_cache_release(page
);
346 * If the truncation happened within a single page no pages
347 * will be released, just zeroed, so we can bail out now.
355 if (!pagevec_lookup_entries(&pvec
, mapping
, index
,
356 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
), indices
)) {
357 /* If all gone from start onwards, we're done */
360 /* Otherwise restart to make sure all gone */
364 if (index
== start
&& indices
[0] >= end
) {
365 /* All gone out of hole to be punched, we're done */
366 pagevec_remove_exceptionals(&pvec
);
367 pagevec_release(&pvec
);
370 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
371 struct page
*page
= pvec
.pages
[i
];
373 /* We rely upon deletion not changing page->index */
376 /* Restart punch to make sure all gone */
381 if (radix_tree_exceptional_entry(page
)) {
382 clear_exceptional_entry(mapping
, index
, page
);
387 WARN_ON(page
->index
!= index
);
388 wait_on_page_writeback(page
);
389 truncate_inode_page(mapping
, page
);
392 pagevec_remove_exceptionals(&pvec
);
393 pagevec_release(&pvec
);
396 cleancache_invalidate_inode(mapping
);
398 EXPORT_SYMBOL(truncate_inode_pages_range
);
401 * truncate_inode_pages - truncate *all* the pages from an offset
402 * @mapping: mapping to truncate
403 * @lstart: offset from which to truncate
405 * Called under (and serialised by) inode->i_mutex.
407 * Note: When this function returns, there can be a page in the process of
408 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
409 * mapping->nrpages can be non-zero when this function returns even after
410 * truncation of the whole mapping.
412 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
414 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
416 EXPORT_SYMBOL(truncate_inode_pages
);
419 * truncate_inode_pages_final - truncate *all* pages before inode dies
420 * @mapping: mapping to truncate
422 * Called under (and serialized by) inode->i_mutex.
424 * Filesystems have to use this in the .evict_inode path to inform the
425 * VM that this is the final truncate and the inode is going away.
427 void truncate_inode_pages_final(struct address_space
*mapping
)
429 unsigned long nrshadows
;
430 unsigned long nrpages
;
433 * Page reclaim can not participate in regular inode lifetime
434 * management (can't call iput()) and thus can race with the
435 * inode teardown. Tell it when the address space is exiting,
436 * so that it does not install eviction information after the
437 * final truncate has begun.
439 mapping_set_exiting(mapping
);
442 * When reclaim installs eviction entries, it increases
443 * nrshadows first, then decreases nrpages. Make sure we see
444 * this in the right order or we might miss an entry.
446 nrpages
= mapping
->nrpages
;
448 nrshadows
= mapping
->nrshadows
;
450 if (nrpages
|| nrshadows
) {
452 * As truncation uses a lockless tree lookup, cycle
453 * the tree lock to make sure any ongoing tree
454 * modification that does not see AS_EXITING is
455 * completed before starting the final truncate.
457 spin_lock_irq(&mapping
->tree_lock
);
458 spin_unlock_irq(&mapping
->tree_lock
);
460 truncate_inode_pages(mapping
, 0);
463 EXPORT_SYMBOL(truncate_inode_pages_final
);
466 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
467 * @mapping: the address_space which holds the pages to invalidate
468 * @start: the offset 'from' which to invalidate
469 * @end: the offset 'to' which to invalidate (inclusive)
471 * This function only removes the unlocked pages, if you want to
472 * remove all the pages of one inode, you must call truncate_inode_pages.
474 * invalidate_mapping_pages() will not block on IO activity. It will not
475 * invalidate pages which are dirty, locked, under writeback or mapped into
478 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
479 pgoff_t start
, pgoff_t end
)
481 pgoff_t indices
[PAGEVEC_SIZE
];
483 pgoff_t index
= start
;
485 unsigned long count
= 0;
488 pagevec_init(&pvec
, 0);
489 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
490 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
492 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
493 struct page
*page
= pvec
.pages
[i
];
495 /* We rely upon deletion not changing page->index */
500 if (radix_tree_exceptional_entry(page
)) {
501 clear_exceptional_entry(mapping
, index
, page
);
505 if (!trylock_page(page
))
507 WARN_ON(page
->index
!= index
);
508 ret
= invalidate_inode_page(page
);
511 * Invalidation is a hint that the page is no longer
512 * of interest and try to speed up its reclaim.
515 deactivate_page(page
);
518 pagevec_remove_exceptionals(&pvec
);
519 pagevec_release(&pvec
);
525 EXPORT_SYMBOL(invalidate_mapping_pages
);
528 * This is like invalidate_complete_page(), except it ignores the page's
529 * refcount. We do this because invalidate_inode_pages2() needs stronger
530 * invalidation guarantees, and cannot afford to leave pages behind because
531 * shrink_page_list() has a temp ref on them, or because they're transiently
532 * sitting in the lru_cache_add() pagevecs.
535 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
537 if (page
->mapping
!= mapping
)
540 if (page_has_private(page
) && !try_to_release_page(page
, GFP_KERNEL
))
543 spin_lock_irq(&mapping
->tree_lock
);
547 BUG_ON(page_has_private(page
));
548 __delete_from_page_cache(page
, NULL
);
549 spin_unlock_irq(&mapping
->tree_lock
);
551 if (mapping
->a_ops
->freepage
)
552 mapping
->a_ops
->freepage(page
);
554 page_cache_release(page
); /* pagecache ref */
557 spin_unlock_irq(&mapping
->tree_lock
);
561 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
563 if (!PageDirty(page
))
565 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
567 return mapping
->a_ops
->launder_page(page
);
571 * invalidate_inode_pages2_range - remove range of pages from an address_space
572 * @mapping: the address_space
573 * @start: the page offset 'from' which to invalidate
574 * @end: the page offset 'to' which to invalidate (inclusive)
576 * Any pages which are found to be mapped into pagetables are unmapped prior to
579 * Returns -EBUSY if any pages could not be invalidated.
581 int invalidate_inode_pages2_range(struct address_space
*mapping
,
582 pgoff_t start
, pgoff_t end
)
584 pgoff_t indices
[PAGEVEC_SIZE
];
590 int did_range_unmap
= 0;
592 cleancache_invalidate_inode(mapping
);
593 pagevec_init(&pvec
, 0);
595 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
596 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
598 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
599 struct page
*page
= pvec
.pages
[i
];
601 /* We rely upon deletion not changing page->index */
606 if (radix_tree_exceptional_entry(page
)) {
607 clear_exceptional_entry(mapping
, index
, page
);
612 WARN_ON(page
->index
!= index
);
613 if (page
->mapping
!= mapping
) {
617 wait_on_page_writeback(page
);
618 if (page_mapped(page
)) {
619 if (!did_range_unmap
) {
621 * Zap the rest of the file in one hit.
623 unmap_mapping_range(mapping
,
624 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
625 (loff_t
)(1 + end
- index
)
633 unmap_mapping_range(mapping
,
634 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
638 BUG_ON(page_mapped(page
));
639 ret2
= do_launder_page(mapping
, page
);
641 if (!invalidate_complete_page2(mapping
, page
))
648 pagevec_remove_exceptionals(&pvec
);
649 pagevec_release(&pvec
);
653 cleancache_invalidate_inode(mapping
);
656 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
659 * invalidate_inode_pages2 - remove all pages from an address_space
660 * @mapping: the address_space
662 * Any pages which are found to be mapped into pagetables are unmapped prior to
665 * Returns -EBUSY if any pages could not be invalidated.
667 int invalidate_inode_pages2(struct address_space
*mapping
)
669 return invalidate_inode_pages2_range(mapping
, 0, -1);
671 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);
674 * truncate_pagecache - unmap and remove pagecache that has been truncated
676 * @newsize: new file size
678 * inode's new i_size must already be written before truncate_pagecache
681 * This function should typically be called before the filesystem
682 * releases resources associated with the freed range (eg. deallocates
683 * blocks). This way, pagecache will always stay logically coherent
684 * with on-disk format, and the filesystem would not have to deal with
685 * situations such as writepage being called for a page that has already
686 * had its underlying blocks deallocated.
688 void truncate_pagecache(struct inode
*inode
, loff_t newsize
)
690 struct address_space
*mapping
= inode
->i_mapping
;
691 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
694 * unmap_mapping_range is called twice, first simply for
695 * efficiency so that truncate_inode_pages does fewer
696 * single-page unmaps. However after this first call, and
697 * before truncate_inode_pages finishes, it is possible for
698 * private pages to be COWed, which remain after
699 * truncate_inode_pages finishes, hence the second
700 * unmap_mapping_range call must be made for correctness.
702 unmap_mapping_range(mapping
, holebegin
, 0, 1);
703 truncate_inode_pages(mapping
, newsize
);
704 unmap_mapping_range(mapping
, holebegin
, 0, 1);
706 EXPORT_SYMBOL(truncate_pagecache
);
709 * truncate_setsize - update inode and pagecache for a new file size
711 * @newsize: new file size
713 * truncate_setsize updates i_size and performs pagecache truncation (if
714 * necessary) to @newsize. It will be typically be called from the filesystem's
715 * setattr function when ATTR_SIZE is passed in.
717 * Must be called with inode_mutex held and before all filesystem specific
718 * block truncation has been performed.
720 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
722 i_size_write(inode
, newsize
);
723 truncate_pagecache(inode
, newsize
);
725 EXPORT_SYMBOL(truncate_setsize
);
728 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
730 * @lstart: offset of beginning of hole
731 * @lend: offset of last byte of hole
733 * This function should typically be called before the filesystem
734 * releases resources associated with the freed range (eg. deallocates
735 * blocks). This way, pagecache will always stay logically coherent
736 * with on-disk format, and the filesystem would not have to deal with
737 * situations such as writepage being called for a page that has already
738 * had its underlying blocks deallocated.
740 void truncate_pagecache_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
742 struct address_space
*mapping
= inode
->i_mapping
;
743 loff_t unmap_start
= round_up(lstart
, PAGE_SIZE
);
744 loff_t unmap_end
= round_down(1 + lend
, PAGE_SIZE
) - 1;
746 * This rounding is currently just for example: unmap_mapping_range
747 * expands its hole outwards, whereas we want it to contract the hole
748 * inwards. However, existing callers of truncate_pagecache_range are
749 * doing their own page rounding first. Note that unmap_mapping_range
750 * allows holelen 0 for all, and we allow lend -1 for end of file.
754 * Unlike in truncate_pagecache, unmap_mapping_range is called only
755 * once (before truncating pagecache), and without "even_cows" flag:
756 * hole-punching should not remove private COWed pages from the hole.
758 if ((u64
)unmap_end
> (u64
)unmap_start
)
759 unmap_mapping_range(mapping
, unmap_start
,
760 1 + unmap_end
- unmap_start
, 0);
761 truncate_inode_pages_range(mapping
, lstart
, lend
);
763 EXPORT_SYMBOL(truncate_pagecache_range
);