2 * mm/truncate.c - code for taking down pages from address_spaces
4 * Copyright (C) 2002, Linus Torvalds
6 * 10Sep2002 akpm@zip.com.au
10 #include <linux/kernel.h>
11 #include <linux/backing-dev.h>
13 #include <linux/swap.h>
14 #include <linux/module.h>
15 #include <linux/pagemap.h>
16 #include <linux/highmem.h>
17 #include <linux/pagevec.h>
18 #include <linux/task_io_accounting_ops.h>
19 #include <linux/buffer_head.h> /* grr. try_to_release_page,
24 * do_invalidatepage - invalidate part or all of a page
25 * @page: the page which is affected
26 * @offset: the index of the truncation point
28 * do_invalidatepage() is called when all or part of the page has become
29 * invalidated by a truncate operation.
31 * do_invalidatepage() does not have to release all buffers, but it must
32 * ensure that no dirty buffer is left outside @offset and that no I/O
33 * is underway against any of the blocks which are outside the truncation
34 * point. Because the caller is about to free (and possibly reuse) those
37 void do_invalidatepage(struct page
*page
, unsigned long offset
)
39 void (*invalidatepage
)(struct page
*, unsigned long);
40 invalidatepage
= page
->mapping
->a_ops
->invalidatepage
;
43 invalidatepage
= block_invalidatepage
;
46 (*invalidatepage
)(page
, offset
);
49 static inline void truncate_partial_page(struct page
*page
, unsigned partial
)
51 zero_user_segment(page
, partial
, PAGE_CACHE_SIZE
);
52 if (PagePrivate(page
))
53 do_invalidatepage(page
, partial
);
57 * This cancels just the dirty bit on the kernel page itself, it
58 * does NOT actually remove dirty bits on any mmap's that may be
59 * around. It also leaves the page tagged dirty, so any sync
60 * activity will still find it on the dirty lists, and in particular,
61 * clear_page_dirty_for_io() will still look at the dirty bits in
64 * Doing this should *normally* only ever be done when a page
65 * is truncated, and is not actually mapped anywhere at all. However,
66 * fs/buffer.c does this when it notices that somebody has cleaned
67 * out all the buffers on a page without actually doing it through
68 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
70 void cancel_dirty_page(struct page
*page
, unsigned int account_size
)
72 if (TestClearPageDirty(page
)) {
73 struct address_space
*mapping
= page
->mapping
;
74 if (mapping
&& mapping_cap_account_dirty(mapping
)) {
75 dec_zone_page_state(page
, NR_FILE_DIRTY
);
76 dec_bdi_stat(mapping
->backing_dev_info
,
79 task_io_account_cancelled_write(account_size
);
83 EXPORT_SYMBOL(cancel_dirty_page
);
86 * If truncate cannot remove the fs-private metadata from the page, the page
87 * becomes orphaned. It will be left on the LRU and may even be mapped into
88 * user pagetables if we're racing with filemap_fault().
90 * We need to bale out if page->mapping is no longer equal to the original
91 * mapping. This happens a) when the VM reclaimed the page while we waited on
92 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
93 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
96 truncate_complete_page(struct address_space
*mapping
, struct page
*page
)
98 if (page
->mapping
!= mapping
)
101 if (PagePrivate(page
))
102 do_invalidatepage(page
, 0);
104 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
106 remove_from_page_cache(page
);
107 ClearPageUptodate(page
);
108 ClearPageMappedToDisk(page
);
109 page_cache_release(page
); /* pagecache ref */
113 * This is for invalidate_mapping_pages(). That function can be called at
114 * any time, and is not supposed to throw away dirty pages. But pages can
115 * be marked dirty at any time too, so use remove_mapping which safely
116 * discards clean, unused pages.
118 * Returns non-zero if the page was successfully invalidated.
121 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
125 if (page
->mapping
!= mapping
)
128 if (PagePrivate(page
) && !try_to_release_page(page
, 0))
131 ret
= remove_mapping(mapping
, page
);
137 * truncate_inode_pages - truncate range of pages specified by start and
139 * @mapping: mapping to truncate
140 * @lstart: offset from which to truncate
141 * @lend: offset to which to truncate
143 * Truncate the page cache, removing the pages that are between
144 * specified offsets (and zeroing out partial page
145 * (if lstart is not page aligned)).
147 * Truncate takes two passes - the first pass is nonblocking. It will not
148 * block on page locks and it will not block on writeback. The second pass
149 * will wait. This is to prevent as much IO as possible in the affected region.
150 * The first pass will remove most pages, so the search cost of the second pass
153 * When looking at page->index outside the page lock we need to be careful to
154 * copy it into a local to avoid races (it could change at any time).
156 * We pass down the cache-hot hint to the page freeing code. Even if the
157 * mapping is large, it is probably the case that the final pages are the most
158 * recently touched, and freeing happens in ascending file offset order.
160 void truncate_inode_pages_range(struct address_space
*mapping
,
161 loff_t lstart
, loff_t lend
)
163 const pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
165 const unsigned partial
= lstart
& (PAGE_CACHE_SIZE
- 1);
170 if (mapping
->nrpages
== 0)
173 BUG_ON((lend
& (PAGE_CACHE_SIZE
- 1)) != (PAGE_CACHE_SIZE
- 1));
174 end
= (lend
>> PAGE_CACHE_SHIFT
);
176 pagevec_init(&pvec
, 0);
178 while (next
<= end
&&
179 pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
180 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
181 struct page
*page
= pvec
.pages
[i
];
182 pgoff_t page_index
= page
->index
;
184 if (page_index
> end
) {
189 if (page_index
> next
)
192 if (TestSetPageLocked(page
))
194 if (PageWriteback(page
)) {
198 if (page_mapped(page
)) {
199 unmap_mapping_range(mapping
,
200 (loff_t
)page_index
<<PAGE_CACHE_SHIFT
,
203 truncate_complete_page(mapping
, page
);
206 pagevec_release(&pvec
);
211 struct page
*page
= find_lock_page(mapping
, start
- 1);
213 wait_on_page_writeback(page
);
214 truncate_partial_page(page
, partial
);
216 page_cache_release(page
);
223 if (!pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
229 if (pvec
.pages
[0]->index
> end
) {
230 pagevec_release(&pvec
);
233 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
234 struct page
*page
= pvec
.pages
[i
];
236 if (page
->index
> end
)
239 wait_on_page_writeback(page
);
240 if (page_mapped(page
)) {
241 unmap_mapping_range(mapping
,
242 (loff_t
)page
->index
<<PAGE_CACHE_SHIFT
,
245 if (page
->index
> next
)
248 truncate_complete_page(mapping
, page
);
251 pagevec_release(&pvec
);
254 EXPORT_SYMBOL(truncate_inode_pages_range
);
257 * truncate_inode_pages - truncate *all* the pages from an offset
258 * @mapping: mapping to truncate
259 * @lstart: offset from which to truncate
261 * Called under (and serialised by) inode->i_mutex.
263 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
265 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
267 EXPORT_SYMBOL(truncate_inode_pages
);
269 unsigned long __invalidate_mapping_pages(struct address_space
*mapping
,
270 pgoff_t start
, pgoff_t end
, bool be_atomic
)
273 pgoff_t next
= start
;
274 unsigned long ret
= 0;
277 pagevec_init(&pvec
, 0);
278 while (next
<= end
&&
279 pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
280 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
281 struct page
*page
= pvec
.pages
[i
];
285 lock_failed
= TestSetPageLocked(page
);
288 * We really shouldn't be looking at the ->index of an
289 * unlocked page. But we're not allowed to lock these
290 * pages. So we rely upon nobody altering the ->index
291 * of this (pinned-by-us) page.
300 if (PageDirty(page
) || PageWriteback(page
))
302 if (page_mapped(page
))
304 ret
+= invalidate_complete_page(mapping
, page
);
310 pagevec_release(&pvec
);
311 if (likely(!be_atomic
))
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
)
333 return __invalidate_mapping_pages(mapping
, start
, end
, false);
335 EXPORT_SYMBOL(invalidate_mapping_pages
);
338 * This is like invalidate_complete_page(), except it ignores the page's
339 * refcount. We do this because invalidate_inode_pages2() needs stronger
340 * invalidation guarantees, and cannot afford to leave pages behind because
341 * shrink_page_list() has a temp ref on them, or because they're transiently
342 * sitting in the lru_cache_add() pagevecs.
345 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
347 if (page
->mapping
!= mapping
)
350 if (PagePrivate(page
) && !try_to_release_page(page
, GFP_KERNEL
))
353 write_lock_irq(&mapping
->tree_lock
);
357 BUG_ON(PagePrivate(page
));
358 __remove_from_page_cache(page
);
359 write_unlock_irq(&mapping
->tree_lock
);
360 ClearPageUptodate(page
);
361 page_cache_release(page
); /* pagecache ref */
364 write_unlock_irq(&mapping
->tree_lock
);
368 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
370 if (!PageDirty(page
))
372 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
374 return mapping
->a_ops
->launder_page(page
);
378 * invalidate_inode_pages2_range - remove range of pages from an address_space
379 * @mapping: the address_space
380 * @start: the page offset 'from' which to invalidate
381 * @end: the page offset 'to' which to invalidate (inclusive)
383 * Any pages which are found to be mapped into pagetables are unmapped prior to
386 * Returns -EIO if any pages could not be invalidated.
388 int invalidate_inode_pages2_range(struct address_space
*mapping
,
389 pgoff_t start
, pgoff_t end
)
395 int did_range_unmap
= 0;
398 pagevec_init(&pvec
, 0);
400 while (next
<= end
&& !wrapped
&&
401 pagevec_lookup(&pvec
, mapping
, next
,
402 min(end
- next
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
403 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
404 struct page
*page
= pvec
.pages
[i
];
408 if (page
->mapping
!= mapping
) {
412 page_index
= page
->index
;
413 next
= page_index
+ 1;
416 if (page_index
> end
) {
420 wait_on_page_writeback(page
);
421 if (page_mapped(page
)) {
422 if (!did_range_unmap
) {
424 * Zap the rest of the file in one hit.
426 unmap_mapping_range(mapping
,
427 (loff_t
)page_index
<<PAGE_CACHE_SHIFT
,
428 (loff_t
)(end
- page_index
+ 1)
436 unmap_mapping_range(mapping
,
437 (loff_t
)page_index
<<PAGE_CACHE_SHIFT
,
441 BUG_ON(page_mapped(page
));
442 ret
= do_launder_page(mapping
, page
);
443 if (ret
== 0 && !invalidate_complete_page2(mapping
, page
))
447 pagevec_release(&pvec
);
452 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
455 * invalidate_inode_pages2 - remove all pages from an address_space
456 * @mapping: the address_space
458 * Any pages which are found to be mapped into pagetables are unmapped prior to
461 * Returns -EIO if any pages could not be invalidated.
463 int invalidate_inode_pages2(struct address_space
*mapping
)
465 return invalidate_inode_pages2_range(mapping
, 0, -1);
467 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);