xref: /linux/mm/truncate.c (revision b0148a98ec5151fec82064d95f11eb9efbc628ea)
1 /*
2  * mm/truncate.c - code for taking down pages from address_spaces
3  *
4  * Copyright (C) 2002, Linus Torvalds
5  *
6  * 10Sep2002	akpm@zip.com.au
7  *		Initial version.
8  */
9 
10 #include <linux/kernel.h>
11 #include <linux/mm.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,
18 				   do_invalidatepage */
19 
20 
21 /**
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
25  *
26  * do_invalidatepage() is called when all or part of the page has become
27  * invalidated by a truncate operation.
28  *
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
33  * blocks on-disk.
34  */
35 void do_invalidatepage(struct page *page, unsigned long offset)
36 {
37 	void (*invalidatepage)(struct page *, unsigned long);
38 	invalidatepage = page->mapping->a_ops->invalidatepage;
39 #ifdef CONFIG_BLOCK
40 	if (!invalidatepage)
41 		invalidatepage = block_invalidatepage;
42 #endif
43 	if (invalidatepage)
44 		(*invalidatepage)(page, offset);
45 }
46 
47 static inline void truncate_partial_page(struct page *page, unsigned partial)
48 {
49 	memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
50 	if (PagePrivate(page))
51 		do_invalidatepage(page, partial);
52 }
53 
54 /*
55  * This cancels just the dirty bit on the kernel page itself, it
56  * does NOT actually remove dirty bits on any mmap's that may be
57  * around. It also leaves the page tagged dirty, so any sync
58  * activity will still find it on the dirty lists, and in particular,
59  * clear_page_dirty_for_io() will still look at the dirty bits in
60  * the VM.
61  *
62  * Doing this should *normally* only ever be done when a page
63  * is truncated, and is not actually mapped anywhere at all. However,
64  * fs/buffer.c does this when it notices that somebody has cleaned
65  * out all the buffers on a page without actually doing it through
66  * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
67  */
68 void cancel_dirty_page(struct page *page, unsigned int account_size)
69 {
70 	if (TestClearPageDirty(page)) {
71 		struct address_space *mapping = page->mapping;
72 		if (mapping && mapping_cap_account_dirty(mapping)) {
73 			dec_zone_page_state(page, NR_FILE_DIRTY);
74 			if (account_size)
75 				task_io_account_cancelled_write(account_size);
76 		}
77 	}
78 }
79 EXPORT_SYMBOL(cancel_dirty_page);
80 
81 /*
82  * If truncate cannot remove the fs-private metadata from the page, the page
83  * becomes anonymous.  It will be left on the LRU and may even be mapped into
84  * user pagetables if we're racing with filemap_nopage().
85  *
86  * We need to bale out if page->mapping is no longer equal to the original
87  * mapping.  This happens a) when the VM reclaimed the page while we waited on
88  * its lock, b) when a concurrent invalidate_inode_pages got there first and
89  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
90  */
91 static void
92 truncate_complete_page(struct address_space *mapping, struct page *page)
93 {
94 	if (page->mapping != mapping)
95 		return;
96 
97 	cancel_dirty_page(page, PAGE_CACHE_SIZE);
98 
99 	if (PagePrivate(page))
100 		do_invalidatepage(page, 0);
101 
102 	ClearPageUptodate(page);
103 	ClearPageMappedToDisk(page);
104 	remove_from_page_cache(page);
105 	page_cache_release(page);	/* pagecache ref */
106 }
107 
108 /*
109  * This is for invalidate_inode_pages().  That function can be called at
110  * any time, and is not supposed to throw away dirty pages.  But pages can
111  * be marked dirty at any time too, so use remove_mapping which safely
112  * discards clean, unused pages.
113  *
114  * Returns non-zero if the page was successfully invalidated.
115  */
116 static int
117 invalidate_complete_page(struct address_space *mapping, struct page *page)
118 {
119 	int ret;
120 
121 	if (page->mapping != mapping)
122 		return 0;
123 
124 	if (PagePrivate(page) && !try_to_release_page(page, 0))
125 		return 0;
126 
127 	ret = remove_mapping(mapping, page);
128 
129 	return ret;
130 }
131 
132 /**
133  * truncate_inode_pages - truncate range of pages specified by start and
134  * end byte offsets
135  * @mapping: mapping to truncate
136  * @lstart: offset from which to truncate
137  * @lend: offset to which to truncate
138  *
139  * Truncate the page cache, removing the pages that are between
140  * specified offsets (and zeroing out partial page
141  * (if lstart is not page aligned)).
142  *
143  * Truncate takes two passes - the first pass is nonblocking.  It will not
144  * block on page locks and it will not block on writeback.  The second pass
145  * will wait.  This is to prevent as much IO as possible in the affected region.
146  * The first pass will remove most pages, so the search cost of the second pass
147  * is low.
148  *
149  * When looking at page->index outside the page lock we need to be careful to
150  * copy it into a local to avoid races (it could change at any time).
151  *
152  * We pass down the cache-hot hint to the page freeing code.  Even if the
153  * mapping is large, it is probably the case that the final pages are the most
154  * recently touched, and freeing happens in ascending file offset order.
155  */
156 void truncate_inode_pages_range(struct address_space *mapping,
157 				loff_t lstart, loff_t lend)
158 {
159 	const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
160 	pgoff_t end;
161 	const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
162 	struct pagevec pvec;
163 	pgoff_t next;
164 	int i;
165 
166 	if (mapping->nrpages == 0)
167 		return;
168 
169 	BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
170 	end = (lend >> PAGE_CACHE_SHIFT);
171 
172 	pagevec_init(&pvec, 0);
173 	next = start;
174 	while (next <= end &&
175 	       pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
176 		for (i = 0; i < pagevec_count(&pvec); i++) {
177 			struct page *page = pvec.pages[i];
178 			pgoff_t page_index = page->index;
179 
180 			if (page_index > end) {
181 				next = page_index;
182 				break;
183 			}
184 
185 			if (page_index > next)
186 				next = page_index;
187 			next++;
188 			if (TestSetPageLocked(page))
189 				continue;
190 			if (PageWriteback(page)) {
191 				unlock_page(page);
192 				continue;
193 			}
194 			truncate_complete_page(mapping, page);
195 			unlock_page(page);
196 		}
197 		pagevec_release(&pvec);
198 		cond_resched();
199 	}
200 
201 	if (partial) {
202 		struct page *page = find_lock_page(mapping, start - 1);
203 		if (page) {
204 			wait_on_page_writeback(page);
205 			truncate_partial_page(page, partial);
206 			unlock_page(page);
207 			page_cache_release(page);
208 		}
209 	}
210 
211 	next = start;
212 	for ( ; ; ) {
213 		cond_resched();
214 		if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
215 			if (next == start)
216 				break;
217 			next = start;
218 			continue;
219 		}
220 		if (pvec.pages[0]->index > end) {
221 			pagevec_release(&pvec);
222 			break;
223 		}
224 		for (i = 0; i < pagevec_count(&pvec); i++) {
225 			struct page *page = pvec.pages[i];
226 
227 			if (page->index > end)
228 				break;
229 			lock_page(page);
230 			wait_on_page_writeback(page);
231 			if (page->index > next)
232 				next = page->index;
233 			next++;
234 			truncate_complete_page(mapping, page);
235 			unlock_page(page);
236 		}
237 		pagevec_release(&pvec);
238 	}
239 }
240 EXPORT_SYMBOL(truncate_inode_pages_range);
241 
242 /**
243  * truncate_inode_pages - truncate *all* the pages from an offset
244  * @mapping: mapping to truncate
245  * @lstart: offset from which to truncate
246  *
247  * Called under (and serialised by) inode->i_mutex.
248  */
249 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
250 {
251 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
252 }
253 EXPORT_SYMBOL(truncate_inode_pages);
254 
255 /**
256  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
257  * @mapping: the address_space which holds the pages to invalidate
258  * @start: the offset 'from' which to invalidate
259  * @end: the offset 'to' which to invalidate (inclusive)
260  *
261  * This function only removes the unlocked pages, if you want to
262  * remove all the pages of one inode, you must call truncate_inode_pages.
263  *
264  * invalidate_mapping_pages() will not block on IO activity. It will not
265  * invalidate pages which are dirty, locked, under writeback or mapped into
266  * pagetables.
267  */
268 unsigned long invalidate_mapping_pages(struct address_space *mapping,
269 				pgoff_t start, pgoff_t end)
270 {
271 	struct pagevec pvec;
272 	pgoff_t next = start;
273 	unsigned long ret = 0;
274 	int i;
275 
276 	pagevec_init(&pvec, 0);
277 	while (next <= end &&
278 			pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
279 		for (i = 0; i < pagevec_count(&pvec); i++) {
280 			struct page *page = pvec.pages[i];
281 			pgoff_t index;
282 			int lock_failed;
283 
284 			lock_failed = TestSetPageLocked(page);
285 
286 			/*
287 			 * We really shouldn't be looking at the ->index of an
288 			 * unlocked page.  But we're not allowed to lock these
289 			 * pages.  So we rely upon nobody altering the ->index
290 			 * of this (pinned-by-us) page.
291 			 */
292 			index = page->index;
293 			if (index > next)
294 				next = index;
295 			next++;
296 			if (lock_failed)
297 				continue;
298 
299 			if (PageDirty(page) || PageWriteback(page))
300 				goto unlock;
301 			if (page_mapped(page))
302 				goto unlock;
303 			ret += invalidate_complete_page(mapping, page);
304 unlock:
305 			unlock_page(page);
306 			if (next > end)
307 				break;
308 		}
309 		pagevec_release(&pvec);
310 	}
311 	return ret;
312 }
313 
314 unsigned long invalidate_inode_pages(struct address_space *mapping)
315 {
316 	return invalidate_mapping_pages(mapping, 0, ~0UL);
317 }
318 EXPORT_SYMBOL(invalidate_inode_pages);
319 
320 /*
321  * This is like invalidate_complete_page(), except it ignores the page's
322  * refcount.  We do this because invalidate_inode_pages2() needs stronger
323  * invalidation guarantees, and cannot afford to leave pages behind because
324  * shrink_list() has a temp ref on them, or because they're transiently sitting
325  * in the lru_cache_add() pagevecs.
326  */
327 static int
328 invalidate_complete_page2(struct address_space *mapping, struct page *page)
329 {
330 	if (page->mapping != mapping)
331 		return 0;
332 
333 	if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
334 		return 0;
335 
336 	write_lock_irq(&mapping->tree_lock);
337 	if (PageDirty(page))
338 		goto failed;
339 
340 	BUG_ON(PagePrivate(page));
341 	__remove_from_page_cache(page);
342 	write_unlock_irq(&mapping->tree_lock);
343 	ClearPageUptodate(page);
344 	page_cache_release(page);	/* pagecache ref */
345 	return 1;
346 failed:
347 	write_unlock_irq(&mapping->tree_lock);
348 	return 0;
349 }
350 
351 static int do_launder_page(struct address_space *mapping, struct page *page)
352 {
353 	if (!PageDirty(page))
354 		return 0;
355 	if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
356 		return 0;
357 	return mapping->a_ops->launder_page(page);
358 }
359 
360 /**
361  * invalidate_inode_pages2_range - remove range of pages from an address_space
362  * @mapping: the address_space
363  * @start: the page offset 'from' which to invalidate
364  * @end: the page offset 'to' which to invalidate (inclusive)
365  *
366  * Any pages which are found to be mapped into pagetables are unmapped prior to
367  * invalidation.
368  *
369  * Returns -EIO if any pages could not be invalidated.
370  */
371 int invalidate_inode_pages2_range(struct address_space *mapping,
372 				  pgoff_t start, pgoff_t end)
373 {
374 	struct pagevec pvec;
375 	pgoff_t next;
376 	int i;
377 	int ret = 0;
378 	int did_range_unmap = 0;
379 	int wrapped = 0;
380 
381 	pagevec_init(&pvec, 0);
382 	next = start;
383 	while (next <= end && !ret && !wrapped &&
384 		pagevec_lookup(&pvec, mapping, next,
385 			min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
386 		for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
387 			struct page *page = pvec.pages[i];
388 			pgoff_t page_index;
389 
390 			lock_page(page);
391 			if (page->mapping != mapping) {
392 				unlock_page(page);
393 				continue;
394 			}
395 			page_index = page->index;
396 			next = page_index + 1;
397 			if (next == 0)
398 				wrapped = 1;
399 			if (page_index > end) {
400 				unlock_page(page);
401 				break;
402 			}
403 			wait_on_page_writeback(page);
404 			while (page_mapped(page)) {
405 				if (!did_range_unmap) {
406 					/*
407 					 * Zap the rest of the file in one hit.
408 					 */
409 					unmap_mapping_range(mapping,
410 					   (loff_t)page_index<<PAGE_CACHE_SHIFT,
411 					   (loff_t)(end - page_index + 1)
412 							<< PAGE_CACHE_SHIFT,
413 					    0);
414 					did_range_unmap = 1;
415 				} else {
416 					/*
417 					 * Just zap this page
418 					 */
419 					unmap_mapping_range(mapping,
420 					  (loff_t)page_index<<PAGE_CACHE_SHIFT,
421 					  PAGE_CACHE_SIZE, 0);
422 				}
423 			}
424 			ret = do_launder_page(mapping, page);
425 			if (ret == 0 && !invalidate_complete_page2(mapping, page))
426 				ret = -EIO;
427 			unlock_page(page);
428 		}
429 		pagevec_release(&pvec);
430 		cond_resched();
431 	}
432 	return ret;
433 }
434 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
435 
436 /**
437  * invalidate_inode_pages2 - remove all pages from an address_space
438  * @mapping: the address_space
439  *
440  * Any pages which are found to be mapped into pagetables are unmapped prior to
441  * invalidation.
442  *
443  * Returns -EIO if any pages could not be invalidated.
444  */
445 int invalidate_inode_pages2(struct address_space *mapping)
446 {
447 	return invalidate_inode_pages2_range(mapping, 0, -1);
448 }
449 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
450