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>
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 cleancache_flush_page(page
->mapping
, page
);
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 ClearPageMappedToDisk(page
);
112 delete_from_page_cache(page
);
117 * This is for invalidate_mapping_pages(). That function can be called at
118 * any time, and is not supposed to throw away dirty pages. But pages can
119 * be marked dirty at any time too, so use remove_mapping which safely
120 * discards clean, unused pages.
122 * Returns non-zero if the page was successfully invalidated.
125 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
129 if (page
->mapping
!= mapping
)
132 if (page_has_private(page
) && !try_to_release_page(page
, 0))
135 clear_page_mlock(page
);
136 ret
= remove_mapping(mapping
, page
);
141 int truncate_inode_page(struct address_space
*mapping
, struct page
*page
)
143 if (page_mapped(page
)) {
144 unmap_mapping_range(mapping
,
145 (loff_t
)page
->index
<< PAGE_CACHE_SHIFT
,
148 return truncate_complete_page(mapping
, page
);
152 * Used to get rid of pages on hardware memory corruption.
154 int generic_error_remove_page(struct address_space
*mapping
, struct page
*page
)
159 * Only punch for normal data pages for now.
160 * Handling other types like directories would need more auditing.
162 if (!S_ISREG(mapping
->host
->i_mode
))
164 return truncate_inode_page(mapping
, page
);
166 EXPORT_SYMBOL(generic_error_remove_page
);
169 * Safely invalidate one page from its pagecache mapping.
170 * It only drops clean, unused pages. The page must be locked.
172 * Returns 1 if the page is successfully invalidated, otherwise 0.
174 int invalidate_inode_page(struct page
*page
)
176 struct address_space
*mapping
= page_mapping(page
);
179 if (PageDirty(page
) || PageWriteback(page
))
181 if (page_mapped(page
))
183 return invalidate_complete_page(mapping
, page
);
187 * truncate_inode_pages - truncate range of pages specified by start & end byte offsets
188 * @mapping: mapping to truncate
189 * @lstart: offset from which to truncate
190 * @lend: offset to which to truncate
192 * Truncate the page cache, removing the pages that are between
193 * specified offsets (and zeroing out partial page
194 * (if lstart is not page aligned)).
196 * Truncate takes two passes - the first pass is nonblocking. It will not
197 * block on page locks and it will not block on writeback. The second pass
198 * will wait. This is to prevent as much IO as possible in the affected region.
199 * The first pass will remove most pages, so the search cost of the second pass
202 * We pass down the cache-hot hint to the page freeing code. Even if the
203 * mapping is large, it is probably the case that the final pages are the most
204 * recently touched, and freeing happens in ascending file offset order.
206 void truncate_inode_pages_range(struct address_space
*mapping
,
207 loff_t lstart
, loff_t lend
)
209 const pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
210 const unsigned partial
= lstart
& (PAGE_CACHE_SIZE
- 1);
216 cleancache_flush_inode(mapping
);
217 if (mapping
->nrpages
== 0)
220 BUG_ON((lend
& (PAGE_CACHE_SIZE
- 1)) != (PAGE_CACHE_SIZE
- 1));
221 end
= (lend
>> PAGE_CACHE_SHIFT
);
223 pagevec_init(&pvec
, 0);
225 while (index
<= end
&& pagevec_lookup(&pvec
, mapping
, index
,
226 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
227 mem_cgroup_uncharge_start();
228 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
229 struct page
*page
= pvec
.pages
[i
];
231 /* We rely upon deletion not changing page->index */
236 if (!trylock_page(page
))
238 WARN_ON(page
->index
!= index
);
239 if (PageWriteback(page
)) {
243 truncate_inode_page(mapping
, page
);
246 pagevec_release(&pvec
);
247 mem_cgroup_uncharge_end();
253 struct page
*page
= find_lock_page(mapping
, start
- 1);
255 wait_on_page_writeback(page
);
256 truncate_partial_page(page
, partial
);
258 page_cache_release(page
);
265 if (!pagevec_lookup(&pvec
, mapping
, index
,
266 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
272 if (index
== start
&& pvec
.pages
[0]->index
> end
) {
273 pagevec_release(&pvec
);
276 mem_cgroup_uncharge_start();
277 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
278 struct page
*page
= pvec
.pages
[i
];
280 /* We rely upon deletion not changing page->index */
286 WARN_ON(page
->index
!= index
);
287 wait_on_page_writeback(page
);
288 truncate_inode_page(mapping
, page
);
291 pagevec_release(&pvec
);
292 mem_cgroup_uncharge_end();
295 cleancache_flush_inode(mapping
);
297 EXPORT_SYMBOL(truncate_inode_pages_range
);
300 * truncate_inode_pages - truncate *all* the pages from an offset
301 * @mapping: mapping to truncate
302 * @lstart: offset from which to truncate
304 * Called under (and serialised by) inode->i_mutex.
306 * Note: When this function returns, there can be a page in the process of
307 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
308 * mapping->nrpages can be non-zero when this function returns even after
309 * truncation of the whole mapping.
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 index
= start
;
336 unsigned long count
= 0;
340 * Note: this function may get called on a shmem/tmpfs mapping:
341 * pagevec_lookup() might then return 0 prematurely (because it
342 * got a gangful of swap entries); but it's hardly worth worrying
343 * about - it can rarely have anything to free from such a mapping
344 * (most pages are dirty), and already skips over any difficulties.
347 pagevec_init(&pvec
, 0);
348 while (index
<= end
&& pagevec_lookup(&pvec
, mapping
, index
,
349 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
350 mem_cgroup_uncharge_start();
351 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
352 struct page
*page
= pvec
.pages
[i
];
354 /* We rely upon deletion not changing page->index */
359 if (!trylock_page(page
))
361 WARN_ON(page
->index
!= index
);
362 ret
= invalidate_inode_page(page
);
365 * Invalidation is a hint that the page is no longer
366 * of interest and try to speed up its reclaim.
369 deactivate_page(page
);
372 pagevec_release(&pvec
);
373 mem_cgroup_uncharge_end();
379 EXPORT_SYMBOL(invalidate_mapping_pages
);
382 * This is like invalidate_complete_page(), except it ignores the page's
383 * refcount. We do this because invalidate_inode_pages2() needs stronger
384 * invalidation guarantees, and cannot afford to leave pages behind because
385 * shrink_page_list() has a temp ref on them, or because they're transiently
386 * sitting in the lru_cache_add() pagevecs.
389 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
391 if (page
->mapping
!= mapping
)
394 if (page_has_private(page
) && !try_to_release_page(page
, GFP_KERNEL
))
397 clear_page_mlock(page
);
399 spin_lock_irq(&mapping
->tree_lock
);
403 BUG_ON(page_has_private(page
));
404 __delete_from_page_cache(page
);
405 spin_unlock_irq(&mapping
->tree_lock
);
406 mem_cgroup_uncharge_cache_page(page
);
408 if (mapping
->a_ops
->freepage
)
409 mapping
->a_ops
->freepage(page
);
411 page_cache_release(page
); /* pagecache ref */
414 spin_unlock_irq(&mapping
->tree_lock
);
418 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
420 if (!PageDirty(page
))
422 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
424 return mapping
->a_ops
->launder_page(page
);
428 * invalidate_inode_pages2_range - remove range of pages from an address_space
429 * @mapping: the address_space
430 * @start: the page offset 'from' which to invalidate
431 * @end: the page offset 'to' which to invalidate (inclusive)
433 * Any pages which are found to be mapped into pagetables are unmapped prior to
436 * Returns -EBUSY if any pages could not be invalidated.
438 int invalidate_inode_pages2_range(struct address_space
*mapping
,
439 pgoff_t start
, pgoff_t end
)
446 int did_range_unmap
= 0;
448 cleancache_flush_inode(mapping
);
449 pagevec_init(&pvec
, 0);
451 while (index
<= end
&& pagevec_lookup(&pvec
, mapping
, index
,
452 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
453 mem_cgroup_uncharge_start();
454 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
455 struct page
*page
= pvec
.pages
[i
];
457 /* We rely upon deletion not changing page->index */
463 WARN_ON(page
->index
!= index
);
464 if (page
->mapping
!= mapping
) {
468 wait_on_page_writeback(page
);
469 if (page_mapped(page
)) {
470 if (!did_range_unmap
) {
472 * Zap the rest of the file in one hit.
474 unmap_mapping_range(mapping
,
475 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
476 (loff_t
)(1 + end
- index
)
484 unmap_mapping_range(mapping
,
485 (loff_t
)index
<< PAGE_CACHE_SHIFT
,
489 BUG_ON(page_mapped(page
));
490 ret2
= do_launder_page(mapping
, page
);
492 if (!invalidate_complete_page2(mapping
, page
))
499 pagevec_release(&pvec
);
500 mem_cgroup_uncharge_end();
504 cleancache_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 * @oldsize: old file size
528 * @newsize: new file size
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 oldsize
, loff_t newsize
)
542 struct address_space
*mapping
= inode
->i_mapping
;
543 loff_t holebegin
= round_up(newsize
, PAGE_SIZE
);
546 * unmap_mapping_range is called twice, first simply for
547 * efficiency so that truncate_inode_pages does fewer
548 * single-page unmaps. However after this first call, and
549 * before truncate_inode_pages finishes, it is possible for
550 * private pages to be COWed, which remain after
551 * truncate_inode_pages finishes, hence the second
552 * unmap_mapping_range call must be made for correctness.
554 unmap_mapping_range(mapping
, holebegin
, 0, 1);
555 truncate_inode_pages(mapping
, newsize
);
556 unmap_mapping_range(mapping
, holebegin
, 0, 1);
558 EXPORT_SYMBOL(truncate_pagecache
);
561 * truncate_setsize - update inode and pagecache for a new file size
563 * @newsize: new file size
565 * truncate_setsize updates i_size and performs pagecache truncation (if
566 * necessary) to @newsize. It will be typically be called from the filesystem's
567 * setattr function when ATTR_SIZE is passed in.
569 * Must be called with inode_mutex held and before all filesystem specific
570 * block truncation has been performed.
572 void truncate_setsize(struct inode
*inode
, loff_t newsize
)
576 oldsize
= inode
->i_size
;
577 i_size_write(inode
, newsize
);
579 truncate_pagecache(inode
, oldsize
, newsize
);
581 EXPORT_SYMBOL(truncate_setsize
);
584 * vmtruncate - unmap mappings "freed" by truncate() syscall
585 * @inode: inode of the file used
586 * @newsize: file offset to start truncating
588 * This function is deprecated and truncate_setsize or truncate_pagecache
589 * should be used instead, together with filesystem specific block truncation.
591 int vmtruncate(struct inode
*inode
, loff_t newsize
)
595 error
= inode_newsize_ok(inode
, newsize
);
599 truncate_setsize(inode
, newsize
);
600 if (inode
->i_op
->truncate
)
601 inode
->i_op
->truncate(inode
);
604 EXPORT_SYMBOL(vmtruncate
);
606 int vmtruncate_range(struct inode
*inode
, loff_t lstart
, loff_t lend
)
608 struct address_space
*mapping
= inode
->i_mapping
;
609 loff_t holebegin
= round_up(lstart
, PAGE_SIZE
);
610 loff_t holelen
= 1 + lend
- holebegin
;
613 * If the underlying filesystem is not going to provide
614 * a way to truncate a range of blocks (punch a hole) -
615 * we should return failure right now.
617 if (!inode
->i_op
->truncate_range
)
620 mutex_lock(&inode
->i_mutex
);
621 inode_dio_wait(inode
);
622 unmap_mapping_range(mapping
, holebegin
, holelen
, 1);
623 inode
->i_op
->truncate_range(inode
, lstart
, lend
);
624 /* unmap again to remove racily COWed private pages */
625 unmap_mapping_range(mapping
, holebegin
, holelen
, 1);
626 mutex_unlock(&inode
->i_mutex
);