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>
12 #include <linux/swap.h>
13 #include <linux/module.h>
14 #include <linux/pagemap.h>
15 #include <linux/pagevec.h>
16 #include <linux/task_io_accounting_ops.h>
17 #include <linux/buffer_head.h> /* grr. try_to_release_page,
22 * do_invalidatepage - invalidate part of all of a page
23 * @page: the page which is affected
24 * @offset: the index of the truncation point
26 * do_invalidatepage() is called when all or part of the page has become
27 * invalidated by a truncate operation.
29 * do_invalidatepage() does not have to release all buffers, but it must
30 * ensure that no dirty buffer is left outside @offset and that no I/O
31 * is underway against any of the blocks which are outside the truncation
32 * point. Because the caller is about to free (and possibly reuse) those
35 void do_invalidatepage(struct page
*page
, unsigned long offset
)
37 void (*invalidatepage
)(struct page
*, unsigned long);
38 invalidatepage
= page
->mapping
->a_ops
->invalidatepage
;
41 invalidatepage
= block_invalidatepage
;
44 (*invalidatepage
)(page
, offset
);
47 static inline void truncate_partial_page(struct page
*page
, unsigned partial
)
49 memclear_highpage_flush(page
, partial
, PAGE_CACHE_SIZE
-partial
);
50 if (PagePrivate(page
))
51 do_invalidatepage(page
, partial
);
54 void cancel_dirty_page(struct page
*page
, unsigned int account_size
)
56 /* If we're cancelling the page, it had better not be mapped any more */
57 if (page_mapped(page
)) {
58 static unsigned int warncount
;
60 WARN_ON(++warncount
< 5);
63 if (TestClearPageDirty(page
)) {
64 struct address_space
*mapping
= page
->mapping
;
65 if (mapping
&& mapping_cap_account_dirty(mapping
)) {
66 dec_zone_page_state(page
, NR_FILE_DIRTY
);
68 task_io_account_cancelled_write(account_size
);
72 EXPORT_SYMBOL(cancel_dirty_page
);
75 * If truncate cannot remove the fs-private metadata from the page, the page
76 * becomes anonymous. It will be left on the LRU and may even be mapped into
77 * user pagetables if we're racing with filemap_nopage().
79 * We need to bale out if page->mapping is no longer equal to the original
80 * mapping. This happens a) when the VM reclaimed the page while we waited on
81 * its lock, b) when a concurrent invalidate_inode_pages got there first and
82 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
85 truncate_complete_page(struct address_space
*mapping
, struct page
*page
)
87 if (page
->mapping
!= mapping
)
90 cancel_dirty_page(page
, PAGE_CACHE_SIZE
);
92 if (PagePrivate(page
))
93 do_invalidatepage(page
, 0);
95 ClearPageUptodate(page
);
96 ClearPageMappedToDisk(page
);
97 remove_from_page_cache(page
);
98 page_cache_release(page
); /* pagecache ref */
102 * This is for invalidate_inode_pages(). That function can be called at
103 * any time, and is not supposed to throw away dirty pages. But pages can
104 * be marked dirty at any time too, so use remove_mapping which safely
105 * discards clean, unused pages.
107 * Returns non-zero if the page was successfully invalidated.
110 invalidate_complete_page(struct address_space
*mapping
, struct page
*page
)
114 if (page
->mapping
!= mapping
)
117 if (PagePrivate(page
) && !try_to_release_page(page
, 0))
120 ret
= remove_mapping(mapping
, page
);
126 * truncate_inode_pages - truncate range of pages specified by start and
128 * @mapping: mapping to truncate
129 * @lstart: offset from which to truncate
130 * @lend: offset to which to truncate
132 * Truncate the page cache, removing the pages that are between
133 * specified offsets (and zeroing out partial page
134 * (if lstart is not page aligned)).
136 * Truncate takes two passes - the first pass is nonblocking. It will not
137 * block on page locks and it will not block on writeback. The second pass
138 * will wait. This is to prevent as much IO as possible in the affected region.
139 * The first pass will remove most pages, so the search cost of the second pass
142 * When looking at page->index outside the page lock we need to be careful to
143 * copy it into a local to avoid races (it could change at any time).
145 * We pass down the cache-hot hint to the page freeing code. Even if the
146 * mapping is large, it is probably the case that the final pages are the most
147 * recently touched, and freeing happens in ascending file offset order.
149 void truncate_inode_pages_range(struct address_space
*mapping
,
150 loff_t lstart
, loff_t lend
)
152 const pgoff_t start
= (lstart
+ PAGE_CACHE_SIZE
-1) >> PAGE_CACHE_SHIFT
;
154 const unsigned partial
= lstart
& (PAGE_CACHE_SIZE
- 1);
159 if (mapping
->nrpages
== 0)
162 BUG_ON((lend
& (PAGE_CACHE_SIZE
- 1)) != (PAGE_CACHE_SIZE
- 1));
163 end
= (lend
>> PAGE_CACHE_SHIFT
);
165 pagevec_init(&pvec
, 0);
167 while (next
<= end
&&
168 pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
169 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
170 struct page
*page
= pvec
.pages
[i
];
171 pgoff_t page_index
= page
->index
;
173 if (page_index
> end
) {
178 if (page_index
> next
)
181 if (TestSetPageLocked(page
))
183 if (PageWriteback(page
)) {
187 truncate_complete_page(mapping
, page
);
190 pagevec_release(&pvec
);
195 struct page
*page
= find_lock_page(mapping
, start
- 1);
197 wait_on_page_writeback(page
);
198 truncate_partial_page(page
, partial
);
200 page_cache_release(page
);
207 if (!pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
213 if (pvec
.pages
[0]->index
> end
) {
214 pagevec_release(&pvec
);
217 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
218 struct page
*page
= pvec
.pages
[i
];
220 if (page
->index
> end
)
223 wait_on_page_writeback(page
);
224 if (page
->index
> next
)
227 truncate_complete_page(mapping
, page
);
230 pagevec_release(&pvec
);
233 EXPORT_SYMBOL(truncate_inode_pages_range
);
236 * truncate_inode_pages - truncate *all* the pages from an offset
237 * @mapping: mapping to truncate
238 * @lstart: offset from which to truncate
240 * Called under (and serialised by) inode->i_mutex.
242 void truncate_inode_pages(struct address_space
*mapping
, loff_t lstart
)
244 truncate_inode_pages_range(mapping
, lstart
, (loff_t
)-1);
246 EXPORT_SYMBOL(truncate_inode_pages
);
249 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
250 * @mapping: the address_space which holds the pages to invalidate
251 * @start: the offset 'from' which to invalidate
252 * @end: the offset 'to' which to invalidate (inclusive)
254 * This function only removes the unlocked pages, if you want to
255 * remove all the pages of one inode, you must call truncate_inode_pages.
257 * invalidate_mapping_pages() will not block on IO activity. It will not
258 * invalidate pages which are dirty, locked, under writeback or mapped into
261 unsigned long invalidate_mapping_pages(struct address_space
*mapping
,
262 pgoff_t start
, pgoff_t end
)
265 pgoff_t next
= start
;
266 unsigned long ret
= 0;
269 pagevec_init(&pvec
, 0);
270 while (next
<= end
&&
271 pagevec_lookup(&pvec
, mapping
, next
, PAGEVEC_SIZE
)) {
272 for (i
= 0; i
< pagevec_count(&pvec
); i
++) {
273 struct page
*page
= pvec
.pages
[i
];
277 lock_failed
= TestSetPageLocked(page
);
280 * We really shouldn't be looking at the ->index of an
281 * unlocked page. But we're not allowed to lock these
282 * pages. So we rely upon nobody altering the ->index
283 * of this (pinned-by-us) page.
292 if (PageDirty(page
) || PageWriteback(page
))
294 if (page_mapped(page
))
296 ret
+= invalidate_complete_page(mapping
, page
);
302 pagevec_release(&pvec
);
307 unsigned long invalidate_inode_pages(struct address_space
*mapping
)
309 return invalidate_mapping_pages(mapping
, 0, ~0UL);
311 EXPORT_SYMBOL(invalidate_inode_pages
);
314 * This is like invalidate_complete_page(), except it ignores the page's
315 * refcount. We do this because invalidate_inode_pages2() needs stronger
316 * invalidation guarantees, and cannot afford to leave pages behind because
317 * shrink_list() has a temp ref on them, or because they're transiently sitting
318 * in the lru_cache_add() pagevecs.
321 invalidate_complete_page2(struct address_space
*mapping
, struct page
*page
)
323 if (page
->mapping
!= mapping
)
326 if (PagePrivate(page
) && !try_to_release_page(page
, GFP_KERNEL
))
329 write_lock_irq(&mapping
->tree_lock
);
333 BUG_ON(PagePrivate(page
));
334 __remove_from_page_cache(page
);
335 write_unlock_irq(&mapping
->tree_lock
);
336 ClearPageUptodate(page
);
337 page_cache_release(page
); /* pagecache ref */
340 write_unlock_irq(&mapping
->tree_lock
);
344 static int do_launder_page(struct address_space
*mapping
, struct page
*page
)
346 if (!PageDirty(page
))
348 if (page
->mapping
!= mapping
|| mapping
->a_ops
->launder_page
== NULL
)
350 return mapping
->a_ops
->launder_page(page
);
354 * invalidate_inode_pages2_range - remove range of pages from an address_space
355 * @mapping: the address_space
356 * @start: the page offset 'from' which to invalidate
357 * @end: the page offset 'to' which to invalidate (inclusive)
359 * Any pages which are found to be mapped into pagetables are unmapped prior to
362 * Returns -EIO if any pages could not be invalidated.
364 int invalidate_inode_pages2_range(struct address_space
*mapping
,
365 pgoff_t start
, pgoff_t end
)
371 int did_range_unmap
= 0;
374 pagevec_init(&pvec
, 0);
376 while (next
<= end
&& !ret
&& !wrapped
&&
377 pagevec_lookup(&pvec
, mapping
, next
,
378 min(end
- next
, (pgoff_t
)PAGEVEC_SIZE
- 1) + 1)) {
379 for (i
= 0; !ret
&& i
< pagevec_count(&pvec
); i
++) {
380 struct page
*page
= pvec
.pages
[i
];
384 if (page
->mapping
!= mapping
) {
388 page_index
= page
->index
;
389 next
= page_index
+ 1;
392 if (page_index
> end
) {
396 wait_on_page_writeback(page
);
397 while (page_mapped(page
)) {
398 if (!did_range_unmap
) {
400 * Zap the rest of the file in one hit.
402 unmap_mapping_range(mapping
,
403 (loff_t
)page_index
<<PAGE_CACHE_SHIFT
,
404 (loff_t
)(end
- page_index
+ 1)
412 unmap_mapping_range(mapping
,
413 (loff_t
)page_index
<<PAGE_CACHE_SHIFT
,
417 ret
= do_launder_page(mapping
, page
);
418 if (ret
== 0 && !invalidate_complete_page2(mapping
, page
))
422 pagevec_release(&pvec
);
428 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range
);
431 * invalidate_inode_pages2 - remove all pages from an address_space
432 * @mapping: the address_space
434 * Any pages which are found to be mapped into pagetables are unmapped prior to
437 * Returns -EIO if any pages could not be invalidated.
439 int invalidate_inode_pages2(struct address_space
*mapping
)
441 return invalidate_inode_pages2_range(mapping
, 0, -1);
443 EXPORT_SYMBOL_GPL(invalidate_inode_pages2
);