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 /* Handled by shmem itself */
29 if (shmem_mapping(mapping
))
32 spin_lock_irq(&mapping
->tree_lock
);
34 * Regular page slots are stabilized by the page lock even
35 * without the tree itself locked. These unlocked entries
36 * need verification under the tree lock.
38 if (radix_tree_delete_item(&mapping
->page_tree
, index
, entry
) == entry
)
40 spin_unlock_irq(&mapping
->tree_lock
);
44 * do_invalidatepage - invalidate part or all of a page
45 * @page: the page which is affected
46 * @offset: start of the range to invalidate
47 * @length: length of the range to invalidate
49 * do_invalidatepage() is called when all or part of the page has become
50 * invalidated by a truncate operation.
52 * do_invalidatepage() does not have to release all buffers, but it must
53 * ensure that no dirty buffer is left outside @offset and that no I/O
54 * is underway against any of the blocks which are outside the truncation
55 * point. Because the caller is about to free (and possibly reuse) those
58 void do_invalidatepage(struct page
*page
, unsigned int offset
,
61 void (*invalidatepage
)(struct page
*, unsigned int, unsigned int);
63 invalidatepage
= page
->mapping
->a_ops
->invalidatepage
;
66 invalidatepage
= block_invalidatepage
;
69 (*invalidatepage
)(page
, offset
, length
);
73 * This cancels just the dirty bit on the kernel page itself, it
74 * does NOT actually remove dirty bits on any mmap's that may be
75 * around. It also leaves the page tagged dirty, so any sync
76 * activity will still find it on the dirty lists, and in particular,
77 * clear_page_dirty_for_io() will still look at the dirty bits in
80 * Doing this should *normally* only ever be done when a page
81 * is truncated, and is not actually mapped anywhere at all. However,
82 * fs/buffer.c does this when it notices that somebody has cleaned
83 * out all the buffers on a page without actually doing it through
84 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
86 void cancel_dirty_page(struct page
*page
, unsigned int account_size
)
88 if (TestClearPageDirty(page
)) {
89 struct address_space
*mapping
= page
->mapping
;
90 if (mapping
&& mapping_cap_account_dirty(mapping
)) {
91 dec_zone_page_state(page
, NR_FILE_DIRTY
);
92 dec_bdi_stat(mapping
->backing_dev_info
,
95 task_io_account_cancelled_write(account_size
);
99 EXPORT_SYMBOL(cancel_dirty_page
);
102 * If truncate cannot remove the fs-private metadata from the page, the page
103 * becomes orphaned. It will be left on the LRU and may even be mapped into
104 * user pagetables if we're racing with filemap_fault().
106 * We need to bale out if page->mapping is no longer equal to the original
107 * mapping. This happens a) when the VM reclaimed the page while we waited on
108 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
109 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
112 truncate_complete_page(struct address_space
*mapping
, struct page
*page
)
114 if (page
->mapping
!= mapping
)
117 if (page_has_private(page
))
118 do_invalidatepage(page
, 0, PAGE_CACHE_SIZE
);
120 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
122 ClearPageMappedToDisk(page
);
123 delete_from_page_cache(page
);
128 * This is for invalidate_mapping_pages(). That function can be called at
129 * any time, and is not supposed to throw away dirty pages. But pages can
130 * be marked dirty at any time too, so use remove_mapping which safely
131 * discards clean, unused pages.
133 * Returns non-zero if the page was successfully invalidated.
136 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
140 if (page
->mapping
!= mapping
)
143 if (page_has_private(page
) && !try_to_release_page(page
, 0))
146 ret
= remove_mapping(mapping
, page
);
151 int truncate_inode_page(struct address_space
*mapping
, struct page
*page
)
153 if (page_mapped(page
)) {
154 unmap_mapping_range(mapping
,
155 (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
,
158 return truncate_complete_page(mapping
, page
);
162 * Used to get rid of pages on hardware memory corruption.
164 int generic_error_remove_page(struct address_space
*mapping
, struct page
*page
)
169 * Only punch for normal data pages for now.
170 * Handling other types like directories would need more auditing.
172 if (!S_ISREG(mapping
->host
->i_mode
))
174 return truncate_inode_page(mapping
, page
);
176 EXPORT_SYMBOL(generic_error_remove_page
);
179 * Safely invalidate one page from its pagecache mapping.
180 * It only drops clean, unused pages. The page must be locked.
182 * Returns 1 if the page is successfully invalidated, otherwise 0.
184 int invalidate_inode_page(struct page
*page
)
186 struct address_space
*mapping
= page_mapping(page
);
189 if (PageDirty(page
) || PageWriteback(page
))
191 if (page_mapped(page
))
193 return invalidate_complete_page(mapping
, page
);
197 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
198 * @mapping: mapping to truncate
199 * @lstart: offset from which to truncate
200 * @lend: offset to which to truncate (inclusive)
202 * Truncate the page cache, removing the pages that are between
203 * specified offsets (and zeroing out partial pages
204 * if lstart or lend + 1 is not page aligned).
206 * Truncate takes two passes - the first pass is nonblocking. It will not
207 * block on page locks and it will not block on writeback. The second pass
208 * will wait. This is to prevent as much IO as possible in the affected region.
209 * The first pass will remove most pages, so the search cost of the second pass
212 * We pass down the cache-hot hint to the page freeing code. Even if the
213 * mapping is large, it is probably the case that the final pages are the most
214 * recently touched, and freeing happens in ascending file offset order.
216 * Note that since ->invalidatepage() accepts range to invalidate
217 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
218 * page aligned properly.
220 void truncate_inode_pages_range(struct address_space
*mapping
,
221 loff_t lstart
, loff_t lend
)
223 pgoff_t start
; /* inclusive */
224 pgoff_t end
; /* exclusive */
225 unsigned int partial_start
; /* inclusive */
226 unsigned int partial_end
; /* exclusive */
228 pgoff_t indices
[PAGEVEC_SIZE
];
232 cleancache_invalidate_inode(mapping
);
233 if (mapping
->nrpages
== 0 && mapping
->nrshadows
== 0)
236 /* Offsets within partial pages */
237 partial_start
= lstart
& (PAGE_CACHE_SIZE
- 1);
238 partial_end
= (lend
+ 1) & (PAGE_CACHE_SIZE
- 1);
241 * 'start' and 'end' always covers the range of pages to be fully
242 * truncated. Partial pages are covered with 'partial_start' at the
243 * start of the range and 'partial_end' at the end of the range.
244 * Note that 'end' is exclusive while 'lend' is inclusive.
246 start
= (lstart
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
249 * lend == -1 indicates end-of-file so we have to set 'end'
250 * to the highest possible pgoff_t and since the type is
251 * unsigned we're using -1.
255 end
= (lend
+ 1) >> PAGE_CACHE_SHIFT
;
257 pagevec_init(&pvec
, 0);
259 while (index
< end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
260 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
262 mem_cgroup_uncharge_start();
263 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
264 struct page
*page
= pvec
.pages
[i
];
266 /* We rely upon deletion not changing page->index */
271 if (radix_tree_exceptional_entry(page
)) {
272 clear_exceptional_entry(mapping
, index
, page
);
276 if (!trylock_page(page
))
278 WARN_ON(page
->index
!= index
);
279 if (PageWriteback(page
)) {
283 truncate_inode_page(mapping
, page
);
286 pagevec_remove_exceptionals(&pvec
);
287 pagevec_release(&pvec
);
288 mem_cgroup_uncharge_end();
294 struct page
*page
= find_lock_page(mapping
, start
- 1);
296 unsigned int top
= PAGE_CACHE_SIZE
;
298 /* Truncation within a single page */
302 wait_on_page_writeback(page
);
303 zero_user_segment(page
, partial_start
, top
);
304 cleancache_invalidate_page(mapping
, page
);
305 if (page_has_private(page
))
306 do_invalidatepage(page
, partial_start
,
307 top
- partial_start
);
309 page_cache_release(page
);
313 struct page
*page
= find_lock_page(mapping
, end
);
315 wait_on_page_writeback(page
);
316 zero_user_segment(page
, 0, partial_end
);
317 cleancache_invalidate_page(mapping
, page
);
318 if (page_has_private(page
))
319 do_invalidatepage(page
, 0,
322 page_cache_release(page
);
326 * If the truncation happened within a single page no pages
327 * will be released, just zeroed, so we can bail out now.
335 if (!pagevec_lookup_entries(&pvec
, mapping
, index
,
336 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
),
343 if (index
== start
&& indices
[0] >= end
) {
344 pagevec_remove_exceptionals(&pvec
);
345 pagevec_release(&pvec
);
348 mem_cgroup_uncharge_start();
349 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
350 struct page
*page
= pvec
.pages
[i
];
352 /* We rely upon deletion not changing page->index */
357 if (radix_tree_exceptional_entry(page
)) {
358 clear_exceptional_entry(mapping
, index
, page
);
363 WARN_ON(page
->index
!= index
);
364 wait_on_page_writeback(page
);
365 truncate_inode_page(mapping
, page
);
368 pagevec_remove_exceptionals(&pvec
);
369 pagevec_release(&pvec
);
370 mem_cgroup_uncharge_end();
373 cleancache_invalidate_inode(mapping
);
375 EXPORT_SYMBOL(truncate_inode_pages_range
);
378 * truncate_inode_pages - truncate *all* the pages from an offset
379 * @mapping: mapping to truncate
380 * @lstart: offset from which to truncate
382 * Called under (and serialised by) inode->i_mutex.
384 * Note: When this function returns, there can be a page in the process of
385 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
386 * mapping->nrpages can be non-zero when this function returns even after
387 * truncation of the whole mapping.
389 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
391 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
393 EXPORT_SYMBOL(truncate_inode_pages
);
396 * truncate_inode_pages_final - truncate *all* pages before inode dies
397 * @mapping: mapping to truncate
399 * Called under (and serialized by) inode->i_mutex.
401 * Filesystems have to use this in the .evict_inode path to inform the
402 * VM that this is the final truncate and the inode is going away.
404 void truncate_inode_pages_final(struct address_space
*mapping
)
406 unsigned long nrshadows
;
407 unsigned long nrpages
;
410 * Page reclaim can not participate in regular inode lifetime
411 * management (can't call iput()) and thus can race with the
412 * inode teardown. Tell it when the address space is exiting,
413 * so that it does not install eviction information after the
414 * final truncate has begun.
416 mapping_set_exiting(mapping
);
419 * When reclaim installs eviction entries, it increases
420 * nrshadows first, then decreases nrpages. Make sure we see
421 * this in the right order or we might miss an entry.
423 nrpages
= mapping
->nrpages
;
425 nrshadows
= mapping
->nrshadows
;
427 if (nrpages
|| nrshadows
) {
429 * As truncation uses a lockless tree lookup, cycle
430 * the tree lock to make sure any ongoing tree
431 * modification that does not see AS_EXITING is
432 * completed before starting the final truncate.
434 spin_lock_irq(&mapping
->tree_lock
);
435 spin_unlock_irq(&mapping
->tree_lock
);
437 truncate_inode_pages(mapping
, 0);
440 EXPORT_SYMBOL(truncate_inode_pages_final
);
443 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
444 * @mapping: the address_space which holds the pages to invalidate
445 * @start: the offset 'from' which to invalidate
446 * @end: the offset 'to' which to invalidate (inclusive)
448 * This function only removes the unlocked pages, if you want to
449 * remove all the pages of one inode, you must call truncate_inode_pages.
451 * invalidate_mapping_pages() will not block on IO activity. It will not
452 * invalidate pages which are dirty, locked, under writeback or mapped into
455 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
456 pgoff_t start
, pgoff_t end
)
458 pgoff_t indices
[PAGEVEC_SIZE
];
460 pgoff_t index
= start
;
462 unsigned long count
= 0;
466 * Note: this function may get called on a shmem/tmpfs mapping:
467 * pagevec_lookup() might then return 0 prematurely (because it
468 * got a gangful of swap entries); but it's hardly worth worrying
469 * about - it can rarely have anything to free from such a mapping
470 * (most pages are dirty), and already skips over any difficulties.
473 pagevec_init(&pvec
, 0);
474 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
475 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
477 mem_cgroup_uncharge_start();
478 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
479 struct page
*page
= pvec
.pages
[i
];
481 /* We rely upon deletion not changing page->index */
486 if (radix_tree_exceptional_entry(page
)) {
487 clear_exceptional_entry(mapping
, index
, page
);
491 if (!trylock_page(page
))
493 WARN_ON(page
->index
!= index
);
494 ret
= invalidate_inode_page(page
);
497 * Invalidation is a hint that the page is no longer
498 * of interest and try to speed up its reclaim.
501 deactivate_page(page
);
504 pagevec_remove_exceptionals(&pvec
);
505 pagevec_release(&pvec
);
506 mem_cgroup_uncharge_end();
512 EXPORT_SYMBOL(invalidate_mapping_pages
);
515 * This is like invalidate_complete_page(), except it ignores the page's
516 * refcount. We do this because invalidate_inode_pages2() needs stronger
517 * invalidation guarantees, and cannot afford to leave pages behind because
518 * shrink_page_list() has a temp ref on them, or because they're transiently
519 * sitting in the lru_cache_add() pagevecs.
522 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
524 if (page
->mapping
!= mapping
)
527 if (page_has_private(page
) && !try_to_release_page(page
, GFP_KERNEL
))
530 spin_lock_irq(&mapping
->tree_lock
);
534 BUG_ON(page_has_private(page
));
535 __delete_from_page_cache(page
, NULL
);
536 spin_unlock_irq(&mapping
->tree_lock
);
537 mem_cgroup_uncharge_cache_page(page
);
539 if (mapping
->a_ops
->freepage
)
540 mapping
->a_ops
->freepage(page
);
542 page_cache_release(page
); /* pagecache ref */
545 spin_unlock_irq(&mapping
->tree_lock
);
549 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
551 if (!PageDirty(page
))
553 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
555 return mapping
->a_ops
->launder_page(page
);
559 * invalidate_inode_pages2_range - remove range of pages from an address_space
560 * @mapping: the address_space
561 * @start: the page offset 'from' which to invalidate
562 * @end: the page offset 'to' which to invalidate (inclusive)
564 * Any pages which are found to be mapped into pagetables are unmapped prior to
567 * Returns -EBUSY if any pages could not be invalidated.
569 int invalidate_inode_pages2_range(struct address_space
*mapping
,
570 pgoff_t start
, pgoff_t end
)
572 pgoff_t indices
[PAGEVEC_SIZE
];
578 int did_range_unmap
= 0;
580 cleancache_invalidate_inode(mapping
);
581 pagevec_init(&pvec
, 0);
583 while (index
<= end
&& pagevec_lookup_entries(&pvec
, mapping
, index
,
584 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1,
586 mem_cgroup_uncharge_start();
587 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
588 struct page
*page
= pvec
.pages
[i
];
590 /* We rely upon deletion not changing page->index */
595 if (radix_tree_exceptional_entry(page
)) {
596 clear_exceptional_entry(mapping
, index
, page
);
601 WARN_ON(page
->index
!= index
);
602 if (page
->mapping
!= mapping
) {
606 wait_on_page_writeback(page
);
607 if (page_mapped(page
)) {
608 if (!did_range_unmap
) {
610 * Zap the rest of the file in one hit.
612 unmap_mapping_range(mapping
,
613 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
614 (loff_t
)(1 + end
- index
)
622 unmap_mapping_range(mapping
,
623 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
627 BUG_ON(page_mapped(page
));
628 ret2
= do_launder_page(mapping
, page
);
630 if (!invalidate_complete_page2(mapping
, page
))
637 pagevec_remove_exceptionals(&pvec
);
638 pagevec_release(&pvec
);
639 mem_cgroup_uncharge_end();
643 cleancache_invalidate_inode(mapping
);
646 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
649 * invalidate_inode_pages2 - remove all pages from an address_space
650 * @mapping: the address_space
652 * Any pages which are found to be mapped into pagetables are unmapped prior to
655 * Returns -EBUSY if any pages could not be invalidated.
657 int invalidate_inode_pages2(struct address_space
*mapping
)
659 return invalidate_inode_pages2_range(mapping
, 0, -1);
661 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);
664 * truncate_pagecache - unmap and remove pagecache that has been truncated
666 * @newsize: new file size
668 * inode's new i_size must already be written before truncate_pagecache
671 * This function should typically be called before the filesystem
672 * releases resources associated with the freed range (eg. deallocates
673 * blocks). This way, pagecache will always stay logically coherent
674 * with on-disk format, and the filesystem would not have to deal with
675 * situations such as writepage being called for a page that has already
676 * had its underlying blocks deallocated.
678 void truncate_pagecache(struct inode
*inode
, loff_t newsize
)
680 struct address_space
*mapping
= inode
->i_mapping
;
681 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
684 * unmap_mapping_range is called twice, first simply for
685 * efficiency so that truncate_inode_pages does fewer
686 * single-page unmaps. However after this first call, and
687 * before truncate_inode_pages finishes, it is possible for
688 * private pages to be COWed, which remain after
689 * truncate_inode_pages finishes, hence the second
690 * unmap_mapping_range call must be made for correctness.
692 unmap_mapping_range(mapping
, holebegin
, 0, 1);
693 truncate_inode_pages(mapping
, newsize
);
694 unmap_mapping_range(mapping
, holebegin
, 0, 1);
696 EXPORT_SYMBOL(truncate_pagecache
);
699 * truncate_setsize - update inode and pagecache for a new file size
701 * @newsize: new file size
703 * truncate_setsize updates i_size and performs pagecache truncation (if
704 * necessary) to @newsize. It will be typically be called from the filesystem's
705 * setattr function when ATTR_SIZE is passed in.
707 * Must be called with inode_mutex held and before all filesystem specific
708 * block truncation has been performed.
710 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
712 i_size_write(inode
, newsize
);
713 truncate_pagecache(inode
, newsize
);
715 EXPORT_SYMBOL(truncate_setsize
);
718 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
720 * @lstart: offset of beginning of hole
721 * @lend: offset of last byte of hole
723 * This function should typically be called before the filesystem
724 * releases resources associated with the freed range (eg. deallocates
725 * blocks). This way, pagecache will always stay logically coherent
726 * with on-disk format, and the filesystem would not have to deal with
727 * situations such as writepage being called for a page that has already
728 * had its underlying blocks deallocated.
730 void truncate_pagecache_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
732 struct address_space
*mapping
= inode
->i_mapping
;
733 loff_t unmap_start
= round_up(lstart
, PAGE_SIZE
);
734 loff_t unmap_end
= round_down(1 + lend
, PAGE_SIZE
) - 1;
736 * This rounding is currently just for example: unmap_mapping_range
737 * expands its hole outwards, whereas we want it to contract the hole
738 * inwards. However, existing callers of truncate_pagecache_range are
739 * doing their own page rounding first. Note that unmap_mapping_range
740 * allows holelen 0 for all, and we allow lend -1 for end of file.
744 * Unlike in truncate_pagecache, unmap_mapping_range is called only
745 * once (before truncating pagecache), and without "even_cows" flag:
746 * hole-punching should not remove private COWed pages from the hole.
748 if ((u64
)unmap_end
> (u64
)unmap_start
)
749 unmap_mapping_range(mapping
, unmap_start
,
750 1 + unmap_end
- unmap_start
, 0);
751 truncate_inode_pages_range(mapping
, lstart
, lend
);
753 EXPORT_SYMBOL(truncate_pagecache_range
);