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 mem_cgroup_uncharge_start();
285 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
286 struct page
*page
= pvec
.pages
[i
];
288 /* We rely upon deletion not changing page->index */
293 if (radix_tree_exceptional_entry(page
)) {
294 clear_exceptional_entry(mapping
, index
, page
);
298 if (!trylock_page(page
))
300 WARN_ON(page
->index
!= index
);
301 if (PageWriteback(page
)) {
305 truncate_inode_page(mapping
, page
);
308 pagevec_remove_exceptionals(&pvec
);
309 pagevec_release(&pvec
);
310 mem_cgroup_uncharge_end();
316 struct page
*page
= find_lock_page(mapping
, start
- 1);
318 unsigned int top
= PAGE_CACHE_SIZE
;
320 /* Truncation within a single page */
324 wait_on_page_writeback(page
);
325 zero_user_segment(page
, partial_start
, top
);
326 cleancache_invalidate_page(mapping
, page
);
327 if (page_has_private(page
))
328 do_invalidatepage(page
, partial_start
,
329 top
- partial_start
);
331 page_cache_release(page
);
335 struct page
*page
= find_lock_page(mapping
, end
);
337 wait_on_page_writeback(page
);
338 zero_user_segment(page
, 0, partial_end
);
339 cleancache_invalidate_page(mapping
, page
);
340 if (page_has_private(page
))
341 do_invalidatepage(page
, 0,
344 page_cache_release(page
);
348 * If the truncation happened within a single page no pages
349 * will be released, just zeroed, so we can bail out now.
357 if (!pagevec_lookup_entries(&pvec
, mapping
, index
,
358 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
365 if (index
== start
&& indices
[0] >= end
) {
366 pagevec_remove_exceptionals(&pvec
);
367 pagevec_release(&pvec
);
370 mem_cgroup_uncharge_start();
371 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
372 struct page
*page
= pvec
.pages
[i
];
374 /* We rely upon deletion not changing page->index */
379 if (radix_tree_exceptional_entry(page
)) {
380 clear_exceptional_entry(mapping
, index
, page
);
385 WARN_ON(page
->index
!= index
);
386 wait_on_page_writeback(page
);
387 truncate_inode_page(mapping
, page
);
390 pagevec_remove_exceptionals(&pvec
);
391 pagevec_release(&pvec
);
392 mem_cgroup_uncharge_end();
395 cleancache_invalidate_inode(mapping
);
397 EXPORT_SYMBOL(truncate_inode_pages_range
);
400 * truncate_inode_pages - truncate *all* the pages from an offset
401 * @mapping: mapping to truncate
402 * @lstart: offset from which to truncate
404 * Called under (and serialised by) inode->i_mutex.
406 * Note: When this function returns, there can be a page in the process of
407 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
408 * mapping->nrpages can be non-zero when this function returns even after
409 * truncation of the whole mapping.
411 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
413 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
415 EXPORT_SYMBOL(truncate_inode_pages
);
418 * truncate_inode_pages_final - truncate *all* pages before inode dies
419 * @mapping: mapping to truncate
421 * Called under (and serialized by) inode->i_mutex.
423 * Filesystems have to use this in the .evict_inode path to inform the
424 * VM that this is the final truncate and the inode is going away.
426 void truncate_inode_pages_final(struct address_space
*mapping
)
428 unsigned long nrshadows
;
429 unsigned long nrpages
;
432 * Page reclaim can not participate in regular inode lifetime
433 * management (can't call iput()) and thus can race with the
434 * inode teardown. Tell it when the address space is exiting,
435 * so that it does not install eviction information after the
436 * final truncate has begun.
438 mapping_set_exiting(mapping
);
441 * When reclaim installs eviction entries, it increases
442 * nrshadows first, then decreases nrpages. Make sure we see
443 * this in the right order or we might miss an entry.
445 nrpages
= mapping
->nrpages
;
447 nrshadows
= mapping
->nrshadows
;
449 if (nrpages
|| nrshadows
) {
451 * As truncation uses a lockless tree lookup, cycle
452 * the tree lock to make sure any ongoing tree
453 * modification that does not see AS_EXITING is
454 * completed before starting the final truncate.
456 spin_lock_irq(&mapping
->tree_lock
);
457 spin_unlock_irq(&mapping
->tree_lock
);
459 truncate_inode_pages(mapping
, 0);
462 EXPORT_SYMBOL(truncate_inode_pages_final
);
465 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
466 * @mapping: the address_space which holds the pages to invalidate
467 * @start: the offset 'from' which to invalidate
468 * @end: the offset 'to' which to invalidate (inclusive)
470 * This function only removes the unlocked pages, if you want to
471 * remove all the pages of one inode, you must call truncate_inode_pages.
473 * invalidate_mapping_pages() will not block on IO activity. It will not
474 * invalidate pages which are dirty, locked, under writeback or mapped into
477 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
478 pgoff_t start
, pgoff_t end
)
480 pgoff_t indices
[PAGEVEC_SIZE
];
482 pgoff_t index
= start
;
484 unsigned long count
= 0;
487 pagevec_init(&pvec
, 0);
488 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
489 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
491 mem_cgroup_uncharge_start();
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
);
520 mem_cgroup_uncharge_end();
526 EXPORT_SYMBOL(invalidate_mapping_pages
);
529 * This is like invalidate_complete_page(), except it ignores the page's
530 * refcount. We do this because invalidate_inode_pages2() needs stronger
531 * invalidation guarantees, and cannot afford to leave pages behind because
532 * shrink_page_list() has a temp ref on them, or because they're transiently
533 * sitting in the lru_cache_add() pagevecs.
536 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
538 if (page
->mapping
!= mapping
)
541 if (page_has_private(page
) && !try_to_release_page(page
, GFP_KERNEL
))
544 spin_lock_irq(&mapping
->tree_lock
);
548 BUG_ON(page_has_private(page
));
549 __delete_from_page_cache(page
, NULL
);
550 spin_unlock_irq(&mapping
->tree_lock
);
551 mem_cgroup_uncharge_cache_page(page
);
553 if (mapping
->a_ops
->freepage
)
554 mapping
->a_ops
->freepage(page
);
556 page_cache_release(page
); /* pagecache ref */
559 spin_unlock_irq(&mapping
->tree_lock
);
563 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
565 if (!PageDirty(page
))
567 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
569 return mapping
->a_ops
->launder_page(page
);
573 * invalidate_inode_pages2_range - remove range of pages from an address_space
574 * @mapping: the address_space
575 * @start: the page offset 'from' which to invalidate
576 * @end: the page offset 'to' which to invalidate (inclusive)
578 * Any pages which are found to be mapped into pagetables are unmapped prior to
581 * Returns -EBUSY if any pages could not be invalidated.
583 int invalidate_inode_pages2_range(struct address_space
*mapping
,
584 pgoff_t start
, pgoff_t end
)
586 pgoff_t indices
[PAGEVEC_SIZE
];
592 int did_range_unmap
= 0;
594 cleancache_invalidate_inode(mapping
);
595 pagevec_init(&pvec
, 0);
597 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
598 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
600 mem_cgroup_uncharge_start();
601 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
602 struct page
*page
= pvec
.pages
[i
];
604 /* We rely upon deletion not changing page->index */
609 if (radix_tree_exceptional_entry(page
)) {
610 clear_exceptional_entry(mapping
, index
, page
);
615 WARN_ON(page
->index
!= index
);
616 if (page
->mapping
!= mapping
) {
620 wait_on_page_writeback(page
);
621 if (page_mapped(page
)) {
622 if (!did_range_unmap
) {
624 * Zap the rest of the file in one hit.
626 unmap_mapping_range(mapping
,
627 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
628 (loff_t
)(1 + end
- index
)
636 unmap_mapping_range(mapping
,
637 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
641 BUG_ON(page_mapped(page
));
642 ret2
= do_launder_page(mapping
, page
);
644 if (!invalidate_complete_page2(mapping
, page
))
651 pagevec_remove_exceptionals(&pvec
);
652 pagevec_release(&pvec
);
653 mem_cgroup_uncharge_end();
657 cleancache_invalidate_inode(mapping
);
660 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
663 * invalidate_inode_pages2 - remove all pages from an address_space
664 * @mapping: the address_space
666 * Any pages which are found to be mapped into pagetables are unmapped prior to
669 * Returns -EBUSY if any pages could not be invalidated.
671 int invalidate_inode_pages2(struct address_space
*mapping
)
673 return invalidate_inode_pages2_range(mapping
, 0, -1);
675 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);
678 * truncate_pagecache - unmap and remove pagecache that has been truncated
680 * @newsize: new file size
682 * inode's new i_size must already be written before truncate_pagecache
685 * This function should typically be called before the filesystem
686 * releases resources associated with the freed range (eg. deallocates
687 * blocks). This way, pagecache will always stay logically coherent
688 * with on-disk format, and the filesystem would not have to deal with
689 * situations such as writepage being called for a page that has already
690 * had its underlying blocks deallocated.
692 void truncate_pagecache(struct inode
*inode
, loff_t newsize
)
694 struct address_space
*mapping
= inode
->i_mapping
;
695 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
698 * unmap_mapping_range is called twice, first simply for
699 * efficiency so that truncate_inode_pages does fewer
700 * single-page unmaps. However after this first call, and
701 * before truncate_inode_pages finishes, it is possible for
702 * private pages to be COWed, which remain after
703 * truncate_inode_pages finishes, hence the second
704 * unmap_mapping_range call must be made for correctness.
706 unmap_mapping_range(mapping
, holebegin
, 0, 1);
707 truncate_inode_pages(mapping
, newsize
);
708 unmap_mapping_range(mapping
, holebegin
, 0, 1);
710 EXPORT_SYMBOL(truncate_pagecache
);
713 * truncate_setsize - update inode and pagecache for a new file size
715 * @newsize: new file size
717 * truncate_setsize updates i_size and performs pagecache truncation (if
718 * necessary) to @newsize. It will be typically be called from the filesystem's
719 * setattr function when ATTR_SIZE is passed in.
721 * Must be called with inode_mutex held and before all filesystem specific
722 * block truncation has been performed.
724 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
726 i_size_write(inode
, newsize
);
727 truncate_pagecache(inode
, newsize
);
729 EXPORT_SYMBOL(truncate_setsize
);
732 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
734 * @lstart: offset of beginning of hole
735 * @lend: offset of last byte of hole
737 * This function should typically be called before the filesystem
738 * releases resources associated with the freed range (eg. deallocates
739 * blocks). This way, pagecache will always stay logically coherent
740 * with on-disk format, and the filesystem would not have to deal with
741 * situations such as writepage being called for a page that has already
742 * had its underlying blocks deallocated.
744 void truncate_pagecache_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
746 struct address_space
*mapping
= inode
->i_mapping
;
747 loff_t unmap_start
= round_up(lstart
, PAGE_SIZE
);
748 loff_t unmap_end
= round_down(1 + lend
, PAGE_SIZE
) - 1;
750 * This rounding is currently just for example: unmap_mapping_range
751 * expands its hole outwards, whereas we want it to contract the hole
752 * inwards. However, existing callers of truncate_pagecache_range are
753 * doing their own page rounding first. Note that unmap_mapping_range
754 * allows holelen 0 for all, and we allow lend -1 for end of file.
758 * Unlike in truncate_pagecache, unmap_mapping_range is called only
759 * once (before truncating pagecache), and without "even_cows" flag:
760 * hole-punching should not remove private COWed pages from the hole.
762 if ((u64
)unmap_end
> (u64
)unmap_start
)
763 unmap_mapping_range(mapping
, unmap_start
,
764 1 + unmap_end
- unmap_start
, 0);
765 truncate_inode_pages_range(mapping
, lstart
, lend
);
767 EXPORT_SYMBOL(truncate_pagecache_range
);